ChicoryTV rewrite — queue tracker (goal, arc, session protocol) #237

Closed
opened 2026-07-11 01:14:09 +02:00 by timothy · 111 comments
Owner

The prize

Finish the rewrite: ship ErsatzTV fully on the ChicoryTV SPA + REST API, delete the legacy Blazor UI, pass the API security review, go live. Everything in the queue serves that arc; anything that doesn't is backlog.

The arc (ordered — this IS the queue)

Status is NOT written in this arc. An item is done when its Gitea issue/milestone is closed — read that, never a marker here. Closed arc items move to "# Done (history)" at the bottom; this section lists only what is still open.

All six numbered arc items are closed; the ChicoryTV rewrite and go-live arc is complete. See "# Done (history)". Do not re-pick them.

Priority pickups (Blocker/High review findings — parallel to gates, do NOT reorder the numbered arc):

  • (none open — the #253 replace-all concurrency contract, the last standing priority pickup, is COMPLETE; see Done history. New Blocker/High review findings get slotted here.)

Backlog (not gates — do not reorder the arc for these): #99, #293 (unbounded all-items search, Low), #308 (concurrent same-item add 500), #330 (Cross-Origin-Resource-Policy header, ZAP Low), #333 (real-server media-source integration validation, non-gating Medium), #334 (local-library paths under the anonymous-read opt-out, Medium), #338 (ZAP warning-exit classification, Low), review nits from #216, playout-editor lock-aware UX (deferred from #215), add-items bucket-type strictness (deferred from #217). Also: App.tsx modularization epic #243–#247 (milestone "ChicoryTV modularization" — non-gate, post-go-live; #244 and the Libraries batch via #202 are done; #245/#246A/#247 are eligible but share App.tsx and must be coordinated).

Pending adversarial reviews (timothy/adversarial-reviewer)

The list is live, not here: query open ersatztv-labeled issues in that repo at every pickup. Pending audits are claimable work items like arc items: read-only, so parallel-safe against any code work; claim by commenting "picking this up" on the reviewer issue (that repo has no in-progress label). A comment claim remains active until a later comment explicitly releases or abandons it; a posted audit/review deliverable is completed work and is not a pickup merely because the reviewer issue stays open for implementer replies. Audit output feeds the triage rule below (findings → gate-blocker / priority pickup / backlog). (No status snapshot is kept here — a hand-refreshed one drifted the same way the arc's inline markers did. The repo is the source of truth; the audit→filed-issue history lives in the session-comment log below.)

Session protocol

STATUS SOURCE OF TRUTH = live Gitea issue state, NEVER this prose. The arc, and any status a session comment hand-writes, lag reality — worst across parallel sessions that narrate each other's merges as "the gate that unblocks X". ALWAYS reconcile an item's real open/closed state (the issue and its milestone) before claiming; when prose and live state disagree, live state wins — fix the prose, don't propagate it. This rule exists because #91b and the #251/#252 priority-pickups were repeatedly re-picked after they had closed (see the 2026-07-12 "stale done items" correction comment). Root cause + standing rule also captured in the kickoff file's workflow lore.

Pickup

  1. Read docs/handoffs/chicorytv-issue-queue.md (standing kickoff prompt + workflow lore), this body,
    and the last ~6 session comments (newest-first; stop at ~6). Build a bounded live shortlist by
    cascading through these tiers in this exact order: open arc → OPEN issues assigned to OPEN milestones → OPEN
    review issues and eligible adversarial-reviewer audits → unmilestoned/unreviewed
    priority: high → priority: medium → priority: low
    . Query each tier only while fewer
    than five eligible unique candidates have been collected; stop at five or when every tier is
    exhausted. A small packet is desired, but an empty or undersized higher tier MUST fall through to
    the next tier. The priority searches cover all OPEN repository issues carrying that label, not only
    IDs mentioned in tracker prose or recent comments. This is a shortlist search, not a full inventory.
    Every candidate row represents an issue (or eligible reviewer audit): milestone records provide tier
    metadata and are NEVER pickup candidates. An OPEN milestone with zero eligible OPEN issues contributes
    zero candidates. If the cheap worker lacks a repository-scoped issue-list tool, it immediately returns
    the literal result TOOL_LIMITATION; it must not rank tracker-mentioned issue IDs, turn milestone records
    into candidates, or infer that a tier is empty. The orchestrator then performs only the mechanical
    repository/state/milestone/label tier queries, passes at most the bounded raw ISSUE rows back to the cheap
    worker, and leaves filtering/ranking to that worker. Never substitute an owner-wide/global search or
    interpret the tool limitation as an empty tier.
  2. At every tier exclude pull requests, CLOSED issues, in-progress claims, parked issues, issues
    blocked by a CLOSED milestone, umbrella/epic containers whose children are the pickups, and reviewer
    audits that are already claimed or contain a posted deliverable. parked means excluded from every
    automatic tier and may be worked only when the user explicitly names that issue. Deduplicate issues
    found in multiple tiers.
    Order arc candidates by arc order; within milestone and review tiers use priority: high >
    medium > low; for otherwise equal candidates preserve this body's explicit eligible order,
    then use lowest issue number. Return the winner plus up to four fallbacks. The orchestrator performs
    one focused live recheck of the winner and walks the supplied fallbacks if it changed; it does not
    rebuild or enumerate the full queue. Never use fix size, recency, or perceived relevance as a
    tiebreaker. Live Gitea state always overrides prose.
  3. Claim: add in-progress to your issue(s) + comment "claiming — session started <timestamp>". (Tiny read-claim race window is accepted; if you discover a duplicate claim, later claimant backs off.) Reviewer-repo audits: claim by comment only; treat that comment as the live claim until a later comment explicitly releases or abandons it, and exclude audits with a posted deliverable even while the issue remains open for implementer replies.
  4. Stale claims: in-progress with no branch commits and no issue comment for ~48h may be taken over — comment "taking over stale claim" first, branch fresh.

Session end — append ONE comment here, using this template

## Session <date> — <topic>
**Closed**: #… (PR #…)
**Filed**: #… — one-line each
**Triage**: <each new issue (yours or others') → gate-blocker / priority pickup / backlog, with severity>
**Reviews**: <adversarial-reviewer audits picked up / delivered / newly filed>
**Arc change**: <none | body edited: what/why>
**Dependency/conflict warnings**: <cross-item deps or shared-surface conflicts for the next session (e.g. "#234 after #231", "coordinate with #215") — NOT a "next pick": the reader computes that from live state via the ranking rule>

Then remove in-progress from your issue(s). (Keep the comment to Closed/Filed/Triage/Reviews facts — do not restate the PR body or a decisions.md entry here; one home per fact.)

Body edits: re-read the body immediately before editing; note the edit in your session comment. Never store session state in the body. When you close an arc/priority item, move it to "# Done (history)" in this body in the SAME session — a "done" that lives only in your comment is the drift that got #91b/#251/#252 re-picked.

Triage rule (keeps eyes on the prize): every new issue gets an explicit verdict at session end, by severity:

  • Gate-blocker — blocks an arc item (add to milestone, slot into the arc above).
  • Priority pickup — adversarial-review finding at Blocker/High severity: correctness bugs in the running product don't wait for a distant gate; slot as next-free-session work, parallel to gates (that's what parallel capacity is for).
  • Backlog — everything else (label only, leave the arc alone).

Mark cross-item dependencies when slotting (e.g. "#234 after #231", "coordinate with #215") — claiming prevents duplicate pickup, not overlapping code changes; the dependency notes do that.

Done (history — do NOT re-pick; kept for context only)

  • Arc 6 — #335 — ChicoryTV v26.8.0 go-live COMPLETE 2026-07-13 (985eed07, image sha256:ffadc227…): exact-image authenticated ZAP + Semgrep gate triaged clean; production trust restricted; Keychain-backed local admin bootstrapped and seed removed; Komodo/PBS backup + prod-copy migration smoke passed; SPA auth, M3U, and XMLTV verified. server-management#590 and milestone "ChicoryTV go-live" are CLOSED.
  • Arc 1 — #207 + #212 — schedules editor mutation depth + channel-editor gaps.
  • Arc 2 — #231 → #232 (+ #233 / #234; the #20 lock-audit tail was #235) — EntityLocker soundness, then scan-lifecycle honesty.
  • Arc 3 — #202 — media-source management (write API + SPA).
  • Arc 4 — #91 phase (b) — Blazor removal, landed 2026-07-11 (408b0deb + cleanup 4407dd53/3c5d0590); #91 + milestone "Blazor removal (#91 phase b)" both CLOSED; 0 .razor files left; #204/#205/#206 folded in; the #235 F9 parity APIs (deep-scan, external-collections via #271 on 2026-07-12) shipped. The 2026-07-12 "now unblocked / recommended next" framing of #91b was stale — it had already merged; #271 was a follow-up parity gap, not a blocker.
  • Arc 5 — #197 — cold API contract + security review COMPLETE 2026-07-13. Owner verdict: GO-WITH-NONBLOCKING-FOLLOWUPS for the application/code gate on 977f9125; original Blocker/High remediation is merged and the authenticated scan reported 0 High/Med/Critical. Operational release/exposure completed in Arc 6 (#335). #334 (Medium), #330 (Low), and #293 (Low) remain backlog.
  • Priority pickups #251 + #252 — PR #258 (87060aa3, 2026-07-11): deco/deco-template/default-deco invalidation enqueues BuildPlayout(…, Reset) for exactly the affected playouts on CancellationToken.None; positional in-place reconcile (CurrentValues.SetValues) in ReplaceProgramScheduleItemsHandler (full content-aware stable identity deferred to #253).
  • #254 — mutation-hardening low-severity tail (a1bd303c).
  • #269 (priority pickup — #253 cross-editor rotation) — PR #307 (7a9b30de, 2026-07-12): every non-If-Match config-sibling writer bumps Version (guarded no-op) + the SaveChangesForcingVersion delta-rebase fix (Codex F1) so rotation survives a race. Filed #308 (concurrent same-item add 500 — pre-existing, backlog).
  • #253 (audit adversarial-reviewer#23 F1 — replace-all optimistic-concurrency contract) — COMPLETE, closed 2026-07-12. Full contract shipped across all 10 replace-all aggregates: Block reference (PR1) → Template/DecoTemplate/Playlist/schedule-items/Collection/Playout×2/MultiCollection/RerunCollection (PR2/PR3) → #269 cross-editor ETag rotation → #265 RFC-7232 evaluation. Follow-up #308 (concurrent same-item add 500) stays backlog.
  • #265 (priority pickup — #253's RFC-7232 refinement) — PR #322 (c197de42, 2026-07-12): If-Match now evaluates per RFC 7232 — valid-but-non-matching/weak/list/comma-in-tag → 412 (not 400), only grammar violations → 400; IfMatchCondition.ExpectedVersions : Option<Seq<int>> + CheckVersion set membership. Cold fork + Codex (2 Highs: comma-in-tag, NBSP-trim→force-write) + Codex fix-commit re-review. decisions.md entry. No wire-contract change.
  • #197 sub-issues shipped: Bundle A #280–#285 (auth posture, PR #292) · Bundle B #283 (artwork content-type XSS, PR #291) · Bundle C #287/#288 (contract-freeze honesty, PR #296) · PR2 SPA cutover #295 (browser cookie-only auth, machine-key screen, logout) + #301 (side-effecting-GET POST-ification), PR #316, 2026-07-12 · #319 (enforcing CSP + Permissions-Policy, PR #323, 2026-07-12). The final owner verdict is recorded in Arc 5 above.
  • #286 (/api/v1 versioning — the C1 BLOCKER) — PR #326 (682dceec, 2026-07-13): whole /api surface mounted at /api/v1, uniform (incl. auth/scan/scripted); ApiVersionRewriteMiddleware rewrites legacy /api/*→/api/v1/* in-pipeline (Deprecation/Sunset); /api/v1 frozen additive-only; route convention standardized (leading-slash absolute per method) + enforced by ApiRouteVersioningTests. Cold fork + Codex + fix-commit re-review. #289 / PR #76 (v0 read-only MCP foundation) CLOSED as superseded — the MCP is rebuilt fresh under #58 against the frozen /api/v1 (read+write), sequenced after #286 (see #58 re-scope + #76 close comments).
  • #314 (/api/v1 authenticated black-box security scan — the #197 exit criterion) — PR #331 (a042623a, 2026-07-13): authenticated OWASP ZAP api-scan (imports /openapi/v1.json, injects X-Api-Key) vs a throwaway :latest container → 0 High/Med/Crit across all 160 ops; semgrep SAST 0. Shipped scripts/security-scan.sh (repeatable per-release host gate) + docs/ci-cd.md. Fixed a go-live-blocking dep-vuln found en route: Microsoft.OpenApi 2.0.0→2.7.5 (GHSA-v5pm-xwqc-g5wc High, direct-pinned; v1.json byte-identical). CORP-header Low → #330 (backlog). Cold review + fix-commit re-review.
  • #213 / #215 / #217 / #218 / #219 / #220 / #221 — CLOSED 2026-07-11 via PRs #222–#228 / integration #239.

Protocol adopted 2026-07-11 (decisions.md entry of same date) replacing per-session rewrites of docs/handoffs/chicorytv-issue-queue.md. That file keeps the static kickoff prompt + workflow lore only. 2026-07-12: arc-status-in-prose pruned to "Done (history)", the adversarial-review status snapshot removed, and "Recommended next" cut to dependency-warnings-only — all after a queue-drift + methodology review (status now lives only in live Gitea state).

# The prize **Finish the rewrite**: ship ErsatzTV fully on the ChicoryTV SPA + REST API, delete the legacy Blazor UI, pass the API security review, go live. Everything in the queue serves that arc; anything that doesn't is backlog. # The arc (ordered — this IS the queue) > **Status is NOT written in this arc.** An item is done when its Gitea issue/milestone is closed — read that, never a marker here. Closed arc items move to **"# Done (history)"** at the bottom; this section lists only what is still open. *All six numbered arc items are closed; the ChicoryTV rewrite and go-live arc is complete. See "# Done (history)". Do not re-pick them.* **Priority pickups (Blocker/High review findings — parallel to gates, do NOT reorder the numbered arc):** - *(none open — the #253 replace-all concurrency contract, the last standing priority pickup, is COMPLETE; see Done history. New Blocker/High review findings get slotted here.)* **Backlog (not gates — do not reorder the arc for these)**: #99, #293 (unbounded all-items search, Low), #308 (concurrent same-item add 500), #330 (Cross-Origin-Resource-Policy header, ZAP Low), #333 (real-server media-source integration validation, non-gating Medium), #334 (local-library paths under the anonymous-read opt-out, Medium), #338 (ZAP warning-exit classification, Low), review nits from #216, playout-editor lock-aware UX (deferred from #215), add-items bucket-type strictness (deferred from #217). Also: **App.tsx modularization epic #243–#247** (milestone "ChicoryTV modularization" — non-gate, post-go-live; #244 and the Libraries batch via #202 are done; #245/#246A/#247 are eligible but share `App.tsx` and must be coordinated). # Pending adversarial reviews (timothy/adversarial-reviewer) **The list is live, not here: query open `ersatztv`-labeled issues in that repo at every pickup.** Pending **audits are claimable work items** like arc items: read-only, so parallel-safe against any code work; claim by commenting "picking this up" on the reviewer issue (that repo has no `in-progress` label). A comment claim remains active until a later comment explicitly releases or abandons it; a posted audit/review deliverable is completed work and is not a pickup merely because the reviewer issue stays open for implementer replies. Audit output feeds the triage rule below (findings → gate-blocker / priority pickup / backlog). *(No status snapshot is kept here — a hand-refreshed one drifted the same way the arc's inline markers did. The repo is the source of truth; the audit→filed-issue history lives in the session-comment log below.)* # Session protocol > **STATUS SOURCE OF TRUTH = live Gitea issue state, NEVER this prose.** The arc, and any status a session comment hand-writes, lag reality — worst across parallel sessions that narrate each other's merges as "the gate that unblocks X". ALWAYS reconcile an item's real open/closed state (the issue **and** its milestone) before claiming; when prose and live state disagree, live state wins — fix the prose, don't propagate it. This rule exists because #91b and the #251/#252 priority-pickups were repeatedly re-picked *after* they had closed (see the 2026-07-12 "stale done items" correction comment). Root cause + standing rule also captured in the kickoff file's workflow lore. **Pickup** 1. Read `docs/handoffs/chicorytv-issue-queue.md` (standing kickoff prompt + workflow lore), this body, and the **last ~6 session comments** (newest-first; stop at ~6). Build a bounded live shortlist by cascading through these tiers in this exact order: **open arc → OPEN issues assigned to OPEN milestones → OPEN `review` issues and eligible adversarial-reviewer audits → unmilestoned/unreviewed `priority: high` → `priority: medium` → `priority: low`**. Query each tier only while fewer than five eligible unique candidates have been collected; stop at five or when every tier is exhausted. A small packet is desired, but an empty or undersized higher tier MUST fall through to the next tier. The priority searches cover all OPEN repository issues carrying that label, not only IDs mentioned in tracker prose or recent comments. This is a shortlist search, not a full inventory. Every candidate row represents an issue (or eligible reviewer audit): milestone records provide tier metadata and are NEVER pickup candidates. An OPEN milestone with zero eligible OPEN issues contributes zero candidates. If the cheap worker lacks a repository-scoped issue-list tool, it immediately returns the literal result `TOOL_LIMITATION`; it must not rank tracker-mentioned issue IDs, turn milestone records into candidates, or infer that a tier is empty. The orchestrator then performs only the mechanical repository/state/milestone/label tier queries, passes at most the bounded raw ISSUE rows back to the cheap worker, and leaves filtering/ranking to that worker. Never substitute an owner-wide/global search or interpret the tool limitation as an empty tier. 2. At every tier exclude pull requests, CLOSED issues, `in-progress` claims, `parked` issues, issues blocked by a CLOSED milestone, umbrella/epic containers whose children are the pickups, and reviewer audits that are already claimed or contain a posted deliverable. `parked` means excluded from every automatic tier and may be worked only when the user explicitly names that issue. Deduplicate issues found in multiple tiers. Order arc candidates by arc order; within milestone and review tiers use `priority: high` > `medium` > `low`; for otherwise equal candidates preserve this body's explicit eligible order, then use lowest issue number. Return the winner plus up to four fallbacks. The orchestrator performs one focused live recheck of the winner and walks the supplied fallbacks if it changed; it does not rebuild or enumerate the full queue. Never use fix size, recency, or perceived relevance as a tiebreaker. Live Gitea state always overrides prose. 3. **Claim**: add `in-progress` to your issue(s) + comment "claiming — session started \<timestamp\>". (Tiny read-claim race window is accepted; if you discover a duplicate claim, later claimant backs off.) Reviewer-repo audits: claim by comment only; treat that comment as the live claim until a later comment explicitly releases or abandons it, and exclude audits with a posted deliverable even while the issue remains open for implementer replies. 4. **Stale claims**: `in-progress` with no branch commits and no issue comment for ~48h may be taken over — comment "taking over stale claim" first, branch fresh. **Session end — append ONE comment here, using this template** ``` ## Session <date> — <topic> **Closed**: #… (PR #…) **Filed**: #… — one-line each **Triage**: <each new issue (yours or others') → gate-blocker / priority pickup / backlog, with severity> **Reviews**: <adversarial-reviewer audits picked up / delivered / newly filed> **Arc change**: <none | body edited: what/why> **Dependency/conflict warnings**: <cross-item deps or shared-surface conflicts for the next session (e.g. "#234 after #231", "coordinate with #215") — NOT a "next pick": the reader computes that from live state via the ranking rule> ``` Then remove `in-progress` from your issue(s). *(Keep the comment to Closed/Filed/Triage/Reviews facts — do not restate the PR body or a decisions.md entry here; one home per fact.)* **Body edits**: re-read the body immediately before editing; note the edit in your session comment. Never store session state in the body. **When you close an arc/priority item, move it to "# Done (history)" in this body in the SAME session** — a "done" that lives only in your comment is the drift that got #91b/#251/#252 re-picked. **Triage rule (keeps eyes on the prize)**: every new issue gets an explicit verdict at session end, by severity: - **Gate-blocker** — blocks an arc item (add to milestone, slot into the arc above). - **Priority pickup** — adversarial-review finding at Blocker/High severity: correctness bugs in the running product don't wait for a distant gate; slot as next-free-session work, parallel to gates (that's what parallel capacity is for). - **Backlog** — everything else (label only, leave the arc alone). Mark cross-item dependencies when slotting (e.g. "#234 after #231", "coordinate with #215") — claiming prevents duplicate pickup, not overlapping code changes; the dependency notes do that. # Done (history — do NOT re-pick; kept for context only) - **Arc 6 — #335** — ChicoryTV v26.8.0 go-live COMPLETE 2026-07-13 (`985eed07`, image `sha256:ffadc227…`): exact-image authenticated ZAP + Semgrep gate triaged clean; production trust restricted; Keychain-backed local admin bootstrapped and seed removed; Komodo/PBS backup + prod-copy migration smoke passed; SPA auth, M3U, and XMLTV verified. server-management#590 and milestone "ChicoryTV go-live" are CLOSED. - **Arc 1 — #207 + #212** — schedules editor mutation depth + channel-editor gaps. - **Arc 2 — #231 → #232** (+ #233 / #234; the #20 lock-audit tail was #235) — EntityLocker soundness, then scan-lifecycle honesty. - **Arc 3 — #202** — media-source management (write API + SPA). - **Arc 4 — #91 phase (b)** — Blazor removal, landed **2026-07-11** (`408b0deb` + cleanup `4407dd53`/`3c5d0590`); #91 + milestone "Blazor removal (#91 phase b)" both CLOSED; 0 `.razor` files left; #204/#205/#206 folded in; the #235 F9 parity APIs (deep-scan, external-collections via #271 on 2026-07-12) shipped. *The 2026-07-12 "now unblocked / recommended next" framing of #91b was stale — it had already merged; #271 was a follow-up parity gap, not a blocker.* - **Arc 5 — #197** — cold API contract + security review COMPLETE 2026-07-13. Owner verdict: **GO-WITH-NONBLOCKING-FOLLOWUPS** for the application/code gate on `977f9125`; original Blocker/High remediation is merged and the authenticated scan reported 0 High/Med/Critical. Operational release/exposure completed in Arc 6 (#335). #334 (Medium), #330 (Low), and #293 (Low) remain backlog. - **Priority pickups #251 + #252** — PR #258 (`87060aa3`, 2026-07-11): deco/deco-template/default-deco invalidation enqueues `BuildPlayout(…, Reset)` for exactly the affected playouts on `CancellationToken.None`; positional in-place reconcile (`CurrentValues.SetValues`) in `ReplaceProgramScheduleItemsHandler` (full content-aware stable identity deferred to #253). - **#254** — mutation-hardening low-severity tail (`a1bd303c`). - **#269** (priority pickup — #253 cross-editor rotation) — PR #307 (`7a9b30de`, 2026-07-12): every non-If-Match config-sibling writer bumps `Version` (guarded no-op) + the `SaveChangesForcingVersion` delta-rebase fix (Codex F1) so rotation survives a race. Filed **#308** (concurrent same-item add 500 — pre-existing, backlog). - **#253** (audit adversarial-reviewer#23 F1 — replace-all optimistic-concurrency contract) — **COMPLETE**, closed 2026-07-12. Full contract shipped across all 10 replace-all aggregates: Block reference (PR1) → Template/DecoTemplate/Playlist/schedule-items/Collection/Playout×2/MultiCollection/RerunCollection (PR2/PR3) → #269 cross-editor ETag rotation → #265 RFC-7232 evaluation. Follow-up #308 (concurrent same-item add 500) stays backlog. - **#265** (priority pickup — #253's RFC-7232 refinement) — PR #322 (`c197de42`, 2026-07-12): `If-Match` now evaluates per RFC 7232 — valid-but-non-matching/weak/list/comma-in-tag → **412** (not 400), only grammar violations → 400; `IfMatchCondition.ExpectedVersions : Option<Seq<int>>` + `CheckVersion` set membership. Cold fork + Codex (2 Highs: comma-in-tag, NBSP-trim→force-write) + Codex fix-commit re-review. decisions.md entry. No wire-contract change. - **#197 sub-issues shipped**: Bundle A #280–#285 (auth posture, PR #292) · Bundle B #283 (artwork content-type XSS, PR #291) · Bundle C #287/#288 (contract-freeze honesty, PR #296) · **PR2 SPA cutover #295** (browser cookie-only auth, machine-key screen, logout) **+ #301** (side-effecting-GET POST-ification), PR #316, 2026-07-12 · **#319** (enforcing CSP + Permissions-Policy, PR #323, 2026-07-12). The final owner verdict is recorded in Arc 5 above. - **#286** (`/api/v1` versioning — the C1 BLOCKER) — PR #326 (`682dceec`, 2026-07-13): whole `/api` surface mounted at `/api/v1`, uniform (incl. auth/scan/scripted); `ApiVersionRewriteMiddleware` rewrites legacy `/api/*`→`/api/v1/*` in-pipeline (Deprecation/Sunset); `/api/v1` frozen **additive-only**; route convention standardized (leading-slash absolute per method) + enforced by `ApiRouteVersioningTests`. Cold fork + Codex + fix-commit re-review. **#289 / PR #76 (v0 read-only MCP foundation) CLOSED as superseded** — the MCP is rebuilt fresh under **#58** against the frozen `/api/v1` (read+write), sequenced after #286 (see #58 re-scope + #76 close comments). - **#314** (`/api/v1` authenticated black-box security scan — the #197 exit criterion) — PR #331 (`a042623a`, 2026-07-13): authenticated OWASP ZAP `api-scan` (imports `/openapi/v1.json`, injects `X-Api-Key`) vs a throwaway `:latest` container → **0 High/Med/Crit** across all 160 ops; semgrep SAST 0. Shipped `scripts/security-scan.sh` (repeatable per-release host gate) + `docs/ci-cd.md`. Fixed a go-live-blocking dep-vuln found en route: **Microsoft.OpenApi 2.0.0→2.7.5** (GHSA-v5pm-xwqc-g5wc High, direct-pinned; v1.json byte-identical). CORP-header Low → **#330** (backlog). Cold review + fix-commit re-review. - **#213 / #215 / #217 / #218 / #219 / #220 / #221** — CLOSED 2026-07-11 via PRs #222–#228 / integration #239. --- *Protocol adopted 2026-07-11 (decisions.md entry of same date) replacing per-session rewrites of `docs/handoffs/chicorytv-issue-queue.md`. That file keeps the static kickoff prompt + workflow lore only. 2026-07-12: arc-status-in-prose pruned to "Done (history)", the adversarial-review status snapshot removed, and "Recommended next" cut to dependency-warnings-only — all after a queue-drift + methodology review (status now lives only in live Gitea state).*
timothy pinned this 2026-07-11 01:14:19 +02:00
Author
Owner

Session 2026-07-11 — reviewer#20 audit + queue protocol v2 migration

Closed: none (adversarial-reviewer#20 deliverable posted; that issue stays open pending implementer replies)
Filed: #231 [PLAN-MODE] EntityLocker soundness · #232 scan lifecycle honesty (2 Blockers) · #233 troubleshooting lock leak · #234 subtitle-extraction/playout lock leaks (coordinate with #215) · #235 async contract normalization + playout build observability — all from the reviewer#20 lock-lifecycle audit, labeled review
Triage: #231–#235 → backlog until #197 prep (they're #197 inputs, listed in the arc's item 4; #234 additionally sequenced after #231 and behind #215's merge). No #91b gate impact.
Arc change: none to the ordering; this comment inaugurates the tracker (protocol adopted main @ dfd2134c, decisions.md 2026-07-11).
Recommended next: #207+#212 remain claimed/in flight; next free pickup after they land = #202 (pre-review its API design via adversarial-reviewer#19 before building).

## Session 2026-07-11 — reviewer#20 audit + queue protocol v2 migration **Closed**: none (adversarial-reviewer#20 deliverable posted; that issue stays open pending implementer replies) **Filed**: #231 `[PLAN-MODE]` EntityLocker soundness · #232 scan lifecycle honesty (2 Blockers) · #233 troubleshooting lock leak · #234 subtitle-extraction/playout lock leaks (coordinate with #215) · #235 async contract normalization + playout build observability — all from the reviewer#20 lock-lifecycle audit, labeled `review` **Triage**: #231–#235 → **backlog until #197 prep** (they're #197 inputs, listed in the arc's item 4; #234 additionally sequenced after #231 and behind #215's merge). No #91b gate impact. **Arc change**: none to the ordering; this comment inaugurates the tracker (protocol adopted `main @ dfd2134c`, decisions.md 2026-07-11). **Recommended next**: #207+#212 remain claimed/in flight; next free pickup after they land = #202 (pre-review its API design via adversarial-reviewer#19 before building).
Author
Owner

Arc change (amends the session comment above) — body edited on Timothy's call: review findings at Blocker/High severity are priority pickups, not backlog-until-#197.

  • New arc item 2: #231 → #232 (reviewer#20 Blockers — the API-lies-about-scans + permanent lock leaks live in the running product now), parallel-safe alongside #202; #233/#234 follow when free (#234 after #231 and after #215 merges).
  • #235 added to milestone 7: its F9 items (deep-scan + external-collections-scan have no API) gate deleting Libraries.razor, so the removal PR (arc item 4) is now explicitly gated on them.
  • Triage rule gains a third verdict: gate-blocker / priority pickup (Blocker/High review findings → next-free-session, parallel to gates) / backlog.
**Arc change (amends the session comment above)** — body edited on Timothy's call: review findings at Blocker/High severity are **priority pickups**, not backlog-until-#197. - New arc item 2: **#231 → #232** (reviewer#20 Blockers — the API-lies-about-scans + permanent lock leaks live in the running product now), parallel-safe alongside #202; #233/#234 follow when free (#234 after #231 and after #215 merges). - **#235 added to milestone 7**: its F9 items (deep-scan + external-collections-scan have no API) gate deleting `Libraries.razor`, so the removal PR (arc item 4) is now explicitly gated on them. - Triage rule gains a third verdict: gate-blocker / **priority pickup** (Blocker/High review findings → next-free-session, parallel to gates) / backlog.
Author
Owner

Session 2026-07-10/11 — adversarial-reviewer#18 gates batch (+ #213 remainder, CI fixes)

Closed: #221 (PR #222) · #220 (PR #223) · #217+#219 (PR #224) · #215 (PR #225) · #218 (PR #226) · #213 remainder (PRs #227+#228) — #223–#228 landed via integration PR #239 (one CI round; pairwise doc conflicts resolved there, incl. the #227/#228 parity-doc reconciliation). Also #236 (CI MySQL port collision — fixed on main ef8915f1 before that issue was filed; closed as duplicate-of-fix).

Filed: none new (this session worked the reviewer#18 findings #219–#221 + promotions, all pre-filed).

Triage:

  • #236 → closed, fixed (ef8915f1); all seven gate-batch branches + PR #229's branch got main merged in (plumbing merge for #229 — its worktree belongs to the parallel session, which should git pull).
  • Deferred-to-backlog from this batch (added to the body's backlog line): playout-editor lock-aware UX (#215's PUTs are server-gated; editors surface 409 as generic error) and add-items bucket-type strictness (#217 fixed existence-only, matching collections; strictness decision → #197).
  • Discrepancy note: this body had listed #215/#217/#220/#221 as backlog while adversarial-reviewer#18 had promoted them to #91b MUST-FIX gates — moot now (all closed), flagged for protocol awareness.

Arc change: body edited — backlog line updated (closed items removed, two deferrals added); #234's "#215 merges" dependency marked satisfied. No ordering change.

Recommended next: arc item 2 (#231 → #232) is fully unblocked and unclaimed; #207+#212 (PR #229) remain the parallel session's. Heads-up for #231: PR #225 added 409 build-lock guards consuming IsPlayoutLocked as-is — lock-lifecycle changes are inherited by those guards without code change, but re-run PlayoutControllerTests/ChannelControllerTests after touching the locker. CI note: migrations job saw one OOM-kill flake (csc exit 137) under 3-runner host load — rerun-and-watch, may need runner memory caps in server-management.

Reviewer#18 reply (done/deferred) posted on adversarial-reviewer#18. in-progress labels removed.

## Session 2026-07-10/11 — adversarial-reviewer#18 gates batch (+ #213 remainder, CI fixes) **Closed**: #221 (PR #222) · #220 (PR #223) · #217+#219 (PR #224) · #215 (PR #225) · #218 (PR #226) · #213 remainder (PRs #227+#228) — #223–#228 landed via integration PR #239 (one CI round; pairwise doc conflicts resolved there, incl. the #227/#228 parity-doc reconciliation). Also #236 (CI MySQL port collision — fixed on main `ef8915f1` before that issue was filed; closed as duplicate-of-fix). **Filed**: none new (this session worked the reviewer#18 findings #219–#221 + promotions, all pre-filed). **Triage**: - #236 → closed, fixed (`ef8915f1`); all seven gate-batch branches + PR #229's branch got main merged in (plumbing merge for #229 — its worktree belongs to the parallel session, which should `git pull`). - Deferred-to-backlog from this batch (added to the body's backlog line): playout-*editor* lock-aware UX (#215's PUTs are server-gated; editors surface 409 as generic error) and add-items bucket-type strictness (#217 fixed existence-only, matching collections; strictness decision → #197). - Discrepancy note: this body had listed #215/#217/#220/#221 as backlog while adversarial-reviewer#18 had promoted them to #91b MUST-FIX gates — moot now (all closed), flagged for protocol awareness. **Arc change**: body edited — backlog line updated (closed items removed, two deferrals added); #234's "#215 merges" dependency marked satisfied. No ordering change. **Recommended next**: arc item 2 (**#231 → #232**) is fully unblocked and unclaimed; #207+#212 (PR #229) remain the parallel session's. Heads-up for #231: PR #225 added 409 build-lock guards consuming `IsPlayoutLocked` as-is — lock-lifecycle changes are inherited by those guards without code change, but re-run `PlayoutControllerTests`/`ChannelControllerTests` after touching the locker. CI note: migrations job saw one OOM-kill flake (`csc` exit 137) under 3-runner host load — rerun-and-watch, may need runner memory caps in server-management. Reviewer#18 reply (done/deferred) posted on adversarial-reviewer#18. `in-progress` labels removed.
Author
Owner

Arc change — body edited (Timothy's call): added a "Pending adversarial reviews" section so the tracker also surfaces open audits in timothy/adversarial-reviewer (authoritative = open ersatztv-labeled issues there; snapshot kept fresh via session comments — current: #21/#22/#23 pending pickup, #19 pre-review open, #20 dispatched). Pickup step 2 now includes unclaimed audits as candidates (claim by comment in the reviewer repo); the session-end template gains a Reviews line. Kickoff prompt updated to match (same commit train as this note).

**Arc change** — body edited (Timothy's call): added a **"Pending adversarial reviews"** section so the tracker also surfaces open audits in timothy/adversarial-reviewer (authoritative = open `ersatztv`-labeled issues there; snapshot kept fresh via session comments — current: **#21/#22/#23 pending pickup**, #19 pre-review open, #20 dispatched). Pickup step 2 now includes unclaimed audits as candidates (claim by comment in the reviewer repo); the session-end template gains a **Reviews** line. Kickoff prompt updated to match (same commit train as this note).
Author
Owner

Session 2026-07-11 — #207+#212 schedules/channel-editor mutation depth (arc item 1)

Closed: #207, #212, #126 (PR #229, merged; in-progress labels removed)
Filed: #230 — schedules dirty-guard gaps (popstate bypass + shuffle-toggle draft normalization) · #238 — TopBar primaryAction button inert on every screen except SchedulesScreen (pre-existing; PR #229 added the ctv:primary-action event mechanism and opted only SchedulesScreen in)
Triage: #230 → backlog (low severity, Blazor had no guard at all — parity-neutral) · #238 → backlog (minor/UX, pre-existing; each screen's real save/action works)
Arc change: none
Recommended next: #202 (arc item 3) — pre-review its API design via adversarial-reviewer#19 (slice 2) BEFORE building, per the arc note. #231→#232 (arc item 2) is parallel-safe alongside it. Cross-session notes: (a) PR #229 heavily reshaped web/src/App.tsx (schedules screen extracted to screens/SchedulesScreen.tsx + web/src/schedules/; navigationGuard consulted in App's navigate) — rebase/merge main before touching App.tsx; (b) schedule items endpoints now return the flat ScheduleItemResponseModel (#126 fixed) — any code consuming the old polymorphic VM shape must migrate; (c) new reusable endpoints: GET /api/languages, GET /api/filler-presets?fillerKind=, GET /api/channels/{music-video-credits-templates,stream-selectors}; (d) I bumped the timeout on the #213 trash see-all paging test to 15s (CI VM hit the 5s default in run 686 — flake class, not a logic issue); (e) api-conventions §7 gained a write-path-projection convention (reload through read includes; beware lazy LanguageExt Map in tests) — worth reading before new mutation endpoints.

## Session 2026-07-11 — #207+#212 schedules/channel-editor mutation depth (arc item 1) **Closed**: #207, #212, #126 (PR #229, merged; `in-progress` labels removed) **Filed**: #230 — schedules dirty-guard gaps (popstate bypass + shuffle-toggle draft normalization) · #238 — TopBar `primaryAction` button inert on every screen except SchedulesScreen (pre-existing; PR #229 added the `ctv:primary-action` event mechanism and opted only SchedulesScreen in) **Triage**: #230 → **backlog** (low severity, Blazor had no guard at all — parity-neutral) · #238 → **backlog** (minor/UX, pre-existing; each screen's real save/action works) **Arc change**: none **Recommended next**: #202 (arc item 3) — pre-review its API design via adversarial-reviewer#19 (slice 2) BEFORE building, per the arc note. #231→#232 (arc item 2) is parallel-safe alongside it. Cross-session notes: (a) PR #229 heavily reshaped `web/src/App.tsx` (schedules screen extracted to `screens/SchedulesScreen.tsx` + `web/src/schedules/`; navigationGuard consulted in App's `navigate`) — rebase/merge main before touching App.tsx; (b) schedule items endpoints now return the flat `ScheduleItemResponseModel` (#126 fixed) — any code consuming the old polymorphic VM shape must migrate; (c) new reusable endpoints: `GET /api/languages`, `GET /api/filler-presets?fillerKind=`, `GET /api/channels/{music-video-credits-templates,stream-selectors}`; (d) I bumped the timeout on the #213 trash see-all paging test to 15s (CI VM hit the 5s default in run 686 — flake class, not a logic issue); (e) api-conventions §7 gained a write-path-projection convention (reload through read includes; beware lazy LanguageExt `Map` in tests) — worth reading before new mutation endpoints.
Author
Owner

Session 2026-07-11 — arc item 2 batch: EntityLocker soundness + scan-lifecycle honesty (#231 → #232)

Closed: #231 + #232 (PR #241, merged to main c42928e8; CI green, branch deleted)
Filed: none
Triage: no new issues. The fork + Codex adversarial review surfaced 3 findings (SPA stuck-scan-button, 409 re-enable-during-lag, scheduler enqueue-failure lock leak) — all fixed in-PR (commit 702b0121), so nothing to file.
Reviews: PR #241 got both a fork adversarial pass and an independent Codex pass (codex:rescue) — both delivered, findings reconciled and fixed. No adversarial-reviewer audits picked up this session; snapshot unchanged: #21/#22/#23 still pending pickup, #19 pre-review open, #20 dispatched — its #231/#232 items now closed, leaving #233/#234/#235 from that audit still open.
Arc change: none to ordering. Arc item 2's #231 → #232 is DONE; the item's #233/#234 follow when free remainder is now fully unblocked (#234's #231 dependency satisfied here; its #215 dependency merged earlier today). [PLAN-MODE] design for #231 was escalated to Fable per the updated handoff (kickoff §FABLE ESCALATION), decision recorded in docs/decisions.md 2026-07-11.
Recommended next: two parallel-safe tracks, different surfaces —

  • arc item 2 remainder (priority pickups, review Blocker/High): #233 (troubleshooting lock leak) and #234 (subtitle/BuildPlayout finally-gating — now consuming #231's contract; unblocked) and #235 (async contract normalization). #234 explicitly builds on the #231 ownership model just landed.
  • arc item 3 #202 (media-source write API + SPA) — pre-review its API design via adversarial-reviewer#19 first.
  • Or an unclaimed reviewer audit (#21/#22/#23, read-only, parallel-safe with anything).

No dependency conflicts between the above; in-progress labels on #231/#232 removed.

## Session 2026-07-11 — arc item 2 batch: EntityLocker soundness + scan-lifecycle honesty (#231 → #232) **Closed**: #231 + #232 (PR #241, merged to main `c42928e8`; CI green, branch deleted) **Filed**: none **Triage**: no new issues. The fork + Codex adversarial review surfaced 3 findings (SPA stuck-scan-button, 409 re-enable-during-lag, scheduler enqueue-failure lock leak) — **all fixed in-PR** (commit `702b0121`), so nothing to file. **Reviews**: PR #241 got both a fork adversarial pass and an independent **Codex** pass (`codex:rescue`) — both delivered, findings reconciled and fixed. No adversarial-reviewer audits picked up this session; snapshot unchanged: **#21/#22/#23 still pending pickup**, #19 pre-review open, #20 dispatched — its #231/#232 items now closed, leaving **#233/#234/#235** from that audit still open. **Arc change**: none to ordering. Arc item 2's `#231 → #232` is **DONE**; the item's `#233/#234 follow when free` remainder is now fully unblocked (#234's `#231` dependency satisfied here; its `#215` dependency merged earlier today). `[PLAN-MODE]` design for #231 was escalated to Fable per the updated handoff (kickoff §FABLE ESCALATION), decision recorded in `docs/decisions.md` 2026-07-11. **Recommended next**: two parallel-safe tracks, different surfaces — - **arc item 2 remainder** (priority pickups, review Blocker/High): **#233** (troubleshooting lock leak) and **#234** (subtitle/BuildPlayout finally-gating — now consuming #231's contract; unblocked) and **#235** (async contract normalization). #234 explicitly builds on the #231 ownership model just landed. - **arc item 3 #202** (media-source write API + SPA) — pre-review its API design via adversarial-reviewer#19 first. - Or an unclaimed reviewer audit (#21/#22/#23, read-only, parallel-safe with anything). No dependency conflicts between the above; `in-progress` labels on #231/#232 removed.
Author
Owner

Session 2026-07-11 — reviewer post-merge blockers (#212/#230)

Closed: #212, #230 (PR #242, merged eba22ce6; CI green)
Filed: #248 — SPA dialogs (overlay.tsx) are not focus-trapped / background-inerted
Triage: #248 → backlog (low severity: pre-existing, cross-cutting overlay a11y; the reachable button-driven variant of its underlying race is fixed in #242; residual path needs keyboard-escape mid-save). Not a gate. May fold into #197's a11y/contract pass if convenient.
Reviews: adversarial-reviewer#19 — replied with the #212/#230 implementer response; next slice = #202 API pre-review (unchanged, still open). No audits picked up or newly filed this session. #21/#22/#23 remain PENDING pickup; #20 still open awaiting implementer replies.
Arc change: none.
Recommended next: #231 → #232 (Reviewer#20 Blocker/High batch, arc item 2 — parallel-safe vs #202) and/or #202 pre-review via adversarial-reviewer#19 slice 2 before building (arc item 3). #234 after #231. All parallel-safe against the pending reviewer audits (#21/#22/#23) which any session can claim read-only.

Process note (workflow lore, appended to the handoff Lessons): re-review the fix commit, not just the initial PR diff, before closing — surfaced by the user this session and it earned its keep: the first Edit-gate fix was only PARTIALLY resolving the in-flight-save clobber (Delete path remained). Two Codex re-review passes drove it to a clean MERGEABLE verdict.

## Session 2026-07-11 — reviewer post-merge blockers (#212/#230) **Closed**: #212, #230 (PR #242, merged `eba22ce6`; CI green) **Filed**: #248 — SPA dialogs (overlay.tsx) are not focus-trapped / background-inerted **Triage**: #248 → **backlog** (low severity: pre-existing, cross-cutting overlay a11y; the reachable button-driven variant of its underlying race is fixed in #242; residual path needs keyboard-escape mid-save). Not a gate. May fold into #197's a11y/contract pass if convenient. **Reviews**: adversarial-reviewer#19 — replied with the #212/#230 implementer response; **next slice = #202 API pre-review** (unchanged, still open). No audits picked up or newly filed this session. #21/#22/#23 remain PENDING pickup; #20 still open awaiting implementer replies. **Arc change**: none. **Recommended next**: **#231 → #232** (Reviewer#20 Blocker/High batch, arc item 2 — parallel-safe vs #202) and/or **#202** pre-review via adversarial-reviewer#19 slice 2 before building (arc item 3). #234 after #231. All parallel-safe against the pending reviewer audits (#21/#22/#23) which any session can claim read-only. **Process note (workflow lore, appended to the handoff Lessons):** re-review the *fix commit*, not just the initial PR diff, before closing — surfaced by the user this session and it earned its keep: the first Edit-gate fix was only PARTIALLY resolving the in-flight-save clobber (Delete path remained). Two Codex re-review passes drove it to a clean MERGEABLE verdict.
Author
Owner

Session 2026-07-11 — #243 modularization phase 1 (#244 Extract ChannelsScreen)

Closed: #244 (PR #249, merged 533c6c1e) — first bounded extraction under epic #243. Behavior-preserving move of the Channels domain out of App.tsx → web/src/screens/ChannelsScreen.tsx + colocated test. Verified: vitest 587, lint/build clean, check:api no-drift, live E2E (0 console errors), independent adversarial review = SHIP.
Filed: none this session.
Triage:

  • #248 (overlay.tsx dialogs not focus-trapped / background-inerted — filed by the #242 session) → backlog. Genuine a11y/correctness gap but low-likelihood (requires defeating the modal via keyboard mid-save), pre-existing, cross-cutting; not a gate-blocker, not a running-product Blocker/High. Good input to note for #197 (cold API/security+posture review) as a known a11y gap; fold into a future shared-dialog a11y pass. Does not reorder the arc.
    Reviews: no change to the pending-audit snapshot — #21/#22/#23 still PENDING pickup, #19 pre-review open, #20 dispatched. I picked up no reviewer-repo audit this session.
    Arc change: none. #243/#244 live in the separate "ChicoryTV modularization" milestone, parallel to the Blazor-removal arc (epic's own AC: "No Blazor-removal milestone work is blocked on this refactor"). Worked at the user's explicit direction; the arc's prize (gates #231/#232 → #202 → #91b → #197) is unchanged and unblocked.
    Recommended next:
  • The prize is still the arc: #231/#232 (Blocker/High, in-flight in the batch/231-232 branch / PR #241) and #202 (pre-review via adversarial-reviewer#19 slice 2 FIRST) are the priority. Reviewer audits #21/#22/#23 are unclaimed parallel-safe pickups.
  • If continuing the #243 track: phase 2 = #245 (Extract Playouts) — but note it is a MUCH larger, riskier surface than Channels (Playouts is the biggest App.tsx domain and overlaps the playout async work in #231–#235; coordinate/rebase). Phase 3 = #246, phase 4 = #247 (folds in the #238 TopBar-action redesign). All rebase off 533c6c1e (App.tsx line anchors shifted ~+? after the Channels removal).
  • Cross-session note: PR #249 removed ~545 lines from App.tsx (Channels block) — any open branch touching App.tsx should merge main in.
## Session 2026-07-11 — #243 modularization phase 1 (#244 Extract ChannelsScreen) **Closed**: #244 (PR #249, merged `533c6c1e`) — first bounded extraction under epic #243. Behavior-preserving move of the Channels domain out of `App.tsx` → `web/src/screens/ChannelsScreen.tsx` + colocated test. Verified: vitest 587, lint/build clean, check:api no-drift, live E2E (0 console errors), independent adversarial review = SHIP. **Filed**: none this session. **Triage**: - **#248** (overlay.tsx dialogs not focus-trapped / background-inerted — filed by the #242 session) → **backlog**. Genuine a11y/correctness gap but low-likelihood (requires defeating the modal via keyboard mid-save), pre-existing, cross-cutting; not a gate-blocker, not a running-product Blocker/High. Good input to note for **#197** (cold API/security+posture review) as a known a11y gap; fold into a future shared-dialog a11y pass. Does not reorder the arc. **Reviews**: no change to the pending-audit snapshot — **#21/#22/#23 still PENDING pickup**, #19 pre-review open, #20 dispatched. I picked up no reviewer-repo audit this session. **Arc change**: none. #243/#244 live in the separate **"ChicoryTV modularization"** milestone, parallel to the Blazor-removal arc (epic's own AC: "No Blazor-removal milestone work is blocked on this refactor"). Worked at the user's explicit direction; the arc's prize (gates #231/#232 → #202 → #91b → #197) is unchanged and unblocked. **Recommended next**: - **The prize is still the arc**: #231/#232 (Blocker/High, in-flight in the batch/231-232 branch / PR #241) and #202 (pre-review via adversarial-reviewer#19 slice 2 FIRST) are the priority. Reviewer audits #21/#22/#23 are unclaimed parallel-safe pickups. - If continuing the #243 track: phase 2 = **#245 (Extract Playouts)** — but note it is a MUCH larger, riskier surface than Channels (Playouts is the biggest App.tsx domain and overlaps the playout async work in #231–#235; coordinate/rebase). Phase 3 = #246, phase 4 = #247 (folds in the #238 TopBar-action redesign). All rebase off `533c6c1e` (App.tsx line anchors shifted ~+? after the Channels removal). - Cross-session note: PR #249 removed ~545 lines from `App.tsx` (Channels block) — any open branch touching `App.tsx` should merge main in.
Author
Owner

Session 2026-07-11 — arc item 2 remainder: troubleshooting + playout lock leaks (#233 → #234)

Closed: #233 + #234 (PR #250, merged to main 77272ed5; CI green, branch deleted)
Filed: none
Triage: no new issues. Review surfaced findings but all were either fixed in-PR or map to existing #235.
Reviews: PR #250 got a fork adversarial pass and an independent Codex pass (codex:rescue) + a fork re-review of the follow-up refactor. Codex caught a High the fork missed — the troubleshooting handler's outer catch released the lock unconditionally → cross-release when an exception fired before this caller acquired (cancellation during validation / DB error). Fixed in-PR (8cdc9636) by centralizing lock ownership in Handle behind a lockAcquired flag (media-path acquisition moved out of GetProcess, which is now lock-free); non-vacuous regression test verified against a negative control. Deferred (with rationale, in the PR + issue close comments) → #235: worker-dispatch-failure lock leak, BuildPlayoutHandler silent-skip observability, TOCTOU race-loser 404-vs-409. adversarial-reviewer#20's #233/#234 items now done; #235 remains from that audit. Snapshot unchanged: reviewer #21/#22/#23 still pending pickup, #19 pre-review open, #20 open (its remaining item = #235).

Arc change: none to ordering. Arc item 2 (#231 → #232 + #233/#234 remainder) is now fully DONE. Only #235 (the async-contract-normalization + observability follow-up, milestone 7) is left of the reviewer#20 batch — it gates the Libraries.razor deletion (F9) and folds the deferrals above.

Recommended next (parallel-safe tracks, different surfaces):

  • arc item 3 #202 (media-source write API + SPA) — pre-review its API design via adversarial-reviewer#19 (slice 2) FIRST. ⚠️ has an SPA component; another session was churning web/src/App.tsx (#244) — check that's landed / coordinate before touching App.tsx.
  • #235 (async contract + build observability) — backend-heavy, folds the three deferrals above; but it explicitly touches web/src/App.tsx:3064-3109 and web/src/api/libraries.ts (status-code-breaking SPA changes) → coordinate with the App.tsx session; it's also the #91b Libraries.razor gate.
  • Or an unclaimed reviewer audit (#21/#22/#23 — read-only, parallel-safe with anything).

in-progress labels on #233/#234 removed; worktree cleaned; main pulled to 77272ed5.

## Session 2026-07-11 — arc item 2 remainder: troubleshooting + playout lock leaks (#233 → #234) **Closed**: #233 + #234 (PR #250, merged to main `77272ed5`; CI green, branch deleted) **Filed**: none **Triage**: no new issues. Review surfaced findings but all were either fixed in-PR or map to existing #235. **Reviews**: PR #250 got a fork adversarial pass **and** an independent **Codex** pass (`codex:rescue`) + a fork re-review of the follow-up refactor. Codex caught a **High the fork missed** — the troubleshooting handler's outer `catch` released the lock unconditionally → cross-release when an exception fired *before* this caller acquired (cancellation during validation / DB error). Fixed in-PR (`8cdc9636`) by centralizing lock ownership in `Handle` behind a `lockAcquired` flag (media-path acquisition moved out of `GetProcess`, which is now lock-free); non-vacuous regression test verified against a negative control. Deferred (with rationale, in the PR + issue close comments) → **#235**: worker-dispatch-failure lock leak, `BuildPlayoutHandler` silent-skip observability, TOCTOU race-loser 404-vs-409. adversarial-reviewer#20's #233/#234 items now done; **#235 remains** from that audit. Snapshot unchanged: reviewer **#21/#22/#23 still pending pickup**, #19 pre-review open, #20 open (its remaining item = #235). **Arc change**: none to ordering. Arc item 2 (`#231 → #232` + `#233/#234 remainder`) is now **fully DONE**. Only **#235** (the async-contract-normalization + observability follow-up, milestone 7) is left of the reviewer#20 batch — it gates the `Libraries.razor` deletion (F9) and folds the deferrals above. **Recommended next** (parallel-safe tracks, different surfaces): - **arc item 3 #202** (media-source write API + SPA) — pre-review its API design via adversarial-reviewer#19 (slice 2) FIRST. ⚠️ has an SPA component; another session was churning `web/src/App.tsx` (#244) — check that's landed / coordinate before touching App.tsx. - **#235** (async contract + build observability) — backend-heavy, folds the three deferrals above; **but** it explicitly touches `web/src/App.tsx:3064-3109` and `web/src/api/libraries.ts` (status-code-breaking SPA changes) → coordinate with the App.tsx session; it's also the #91b `Libraries.razor` gate. - Or an unclaimed reviewer audit (#21/#22/#23 — read-only, parallel-safe with anything). `in-progress` labels on #233/#234 removed; worktree cleaned; main pulled to `77272ed5`.
Author
Owner

Session 2026-07-11 — mutation-safety audit sweep (adversarial-reviewer #21/#22/#23)

Closed: none (these are read-only audits — deliverables posted, reviewer issues stay open awaiting your review, per the #20 pattern). Picked the reviewer-audit track (user's call) since arc item 3 #202 is claimed/in-flight and #235 has App.tsx-coordination risk with that same session.

Filed (all from the audit findings, CONFIRMED against main @ d93bca65, orchestrator-spot-checked):

  • #251 — Deco-template/deco edits don't invalidate dependent playouts; editor reports 200 but filler/break content stays stale indefinitely until a manual Reset (no BlockKey deco dimension + deco re-applies only on Reset + DailyRebuildTime null by default). bug/review/priority: high.
  • #252 — Schedule-items PUT cascade-deletes PlayoutScheduleItemFillGroupIndex; every save (even a no-op) silently resets fill-group/shuffle progression (OnDelete(Cascade) + RemoveRange+reinsert). bug/review/priority: high.
  • #253 — Replace-all PUTs have no optimistic concurrency (grep-confirmed zero ETag/rowversion/IsConcurrencyToken); stale two-tab overwrite returns 200 silently across 10 aggregate surfaces. Proposes a shared If-Match/int RowVersion→412 contract. api/review/priority: high.
  • #254 — Hardening cluster (low/medium tail): post-commit CancellationToken.None sweep (#22 F4/F3), orphan guide xml + static File.Delete bypass (#22 F2), reindex-not-reconciled (#22 F1), child-id instability (#23 F3), dirty-nav guard on only SchedulesScreen (#23 F6), Max()-on-empty latent trap (#23 F4→#172), empty-list semantics (#23 F7), block/template immediacy (#21 F2). review/priority: low.

Triage:

  • #251 → priority pickup (High-truthfulness: "save succeeds, content silently wrong" running-product bug; blast radius contained to filler/break, manual-Reset workaround exists → could also read as Medium, labeled high for the truthfulness angle). Backend-only scheduling handler — parallel-safe vs #202 (media-source) and #235.
  • #252 → priority pickup (High; running-product correctness — schedule edits silently reset shuffle progression). Discrete, independently fixable (reconcile-by-id in one handler). Backend-only — parallel-safe.
  • #253 → gate input for #197 + priority (High severity, but big cross-cutting change: 9 aggregates + ETag surface + SPA echo + dual-provider migration). Timing tension worth your call: the audit argues for locking the shared contract before more editors ossify (#202 media-source editors, #91b) — i.e. it may deserve to jump ahead of new-editor work rather than wait for the distant #197 gate. Not a hard gate-blocker for #91b removal itself.
  • #254 → backlog / #197-prep. Several items better absorbed into related trackers (F4→#172, F1/F7→#197, F6→#230/#238) than fixed standalone.

Reviews: picked up and delivered all three pending audits — #21 (invalidation), #22 (post-commit side effects), #23 (replace-all concurrency); full findings posted as comments on each. Updated pending-audit snapshot: #21/#22/#23 = DELIVERED (open, awaiting your review/dispatch decision, like #20). Still pending pickup: #24 (plan review of modularization epic #243–247). Still open: #19 (pre-review, next slice = #202 API), #20 (remaining item = #235). No new audits filed in the reviewer repo.

Arc change: none.

Recommended next (parallel-safe tracks, distinct surfaces):

  • #251 + #252 — both backend-only scheduling-handler fixes, small and discrete, no overlap with #202's media-source surface or #235's App.tsx/libraries.ts surface. Ideal next-free-session priority pickups; can even be one PR (both touch Scheduling/Commands replace handlers).
  • #253 — the concurrency contract; recommend a decision on timing (pre-ossification vs #197-gate) before building, and a Fable/[PLAN-MODE] design pass given its cross-cutting blast radius. Strong #197 input either way.
  • Arc item 3 #202 remains the parallel session's (claimed 13:58, PLAN-MODE). #235 still open with the App.tsx-coordination caveat.
  • No in-progress labels to remove (reviewer audits claimed by comment; no ersatztv issue was labeled this session).
## Session 2026-07-11 — mutation-safety audit sweep (adversarial-reviewer #21/#22/#23) **Closed**: none (these are read-only audits — deliverables posted, reviewer issues stay open awaiting your review, per the #20 pattern). Picked the reviewer-audit track (user's call) since arc item 3 #202 is claimed/in-flight and #235 has App.tsx-coordination risk with that same session. **Filed** (all from the audit findings, CONFIRMED against `main @ d93bca65`, orchestrator-spot-checked): - **#251** — Deco-template/deco edits don't invalidate dependent playouts; editor reports 200 but filler/break content stays stale **indefinitely** until a manual Reset (no `BlockKey` deco dimension + deco re-applies only on `Reset` + `DailyRebuildTime` null by default). `bug`/`review`/`priority: high`. - **#252** — Schedule-items PUT cascade-deletes `PlayoutScheduleItemFillGroupIndex`; **every** save (even a no-op) silently resets fill-group/shuffle progression (`OnDelete(Cascade)` + `RemoveRange`+reinsert). `bug`/`review`/`priority: high`. - **#253** — Replace-all PUTs have **no optimistic concurrency** (grep-confirmed zero ETag/rowversion/`IsConcurrencyToken`); stale two-tab overwrite returns 200 silently across 10 aggregate surfaces. Proposes a shared `If-Match`/`int RowVersion`→412 contract. `api`/`review`/`priority: high`. - **#254** — Hardening cluster (low/medium tail): post-commit `CancellationToken.None` sweep (#22 F4/F3), orphan guide xml + static `File.Delete` bypass (#22 F2), reindex-not-reconciled (#22 F1), child-id instability (#23 F3), dirty-nav guard on only SchedulesScreen (#23 F6), `Max()`-on-empty latent trap (#23 F4→#172), empty-list semantics (#23 F7), block/template immediacy (#21 F2). `review`/`priority: low`. **Triage**: - **#251** → **priority pickup** (High-truthfulness: "save succeeds, content silently wrong" running-product bug; blast radius contained to filler/break, manual-Reset workaround exists → could also read as Medium, labeled high for the truthfulness angle). Backend-only scheduling handler — **parallel-safe** vs #202 (media-source) and #235. - **#252** → **priority pickup** (High; running-product correctness — schedule edits silently reset shuffle progression). Discrete, independently fixable (reconcile-by-id in one handler). Backend-only — **parallel-safe**. - **#253** → **gate input for #197** + priority (High severity, but big cross-cutting change: 9 aggregates + ETag surface + SPA echo + dual-provider migration). Timing tension worth your call: the audit argues for locking the shared contract **before more editors ossify** (#202 media-source editors, #91b) — i.e. it may deserve to jump ahead of new-editor work rather than wait for the distant #197 gate. Not a hard gate-blocker for #91b removal itself. - **#254** → **backlog / #197-prep**. Several items better absorbed into related trackers (F4→#172, F1/F7→#197, F6→#230/#238) than fixed standalone. **Reviews**: picked up and **delivered all three pending audits** — #21 (invalidation), #22 (post-commit side effects), #23 (replace-all concurrency); full findings posted as comments on each. Updated pending-audit snapshot: **#21/#22/#23 = DELIVERED** (open, awaiting your review/dispatch decision, like #20). Still pending pickup: **#24** (plan review of modularization epic #243–247). Still open: #19 (pre-review, next slice = #202 API), #20 (remaining item = #235). No new audits filed in the reviewer repo. **Arc change**: none. **Recommended next** (parallel-safe tracks, distinct surfaces): - **#251 + #252** — both backend-only scheduling-handler fixes, small and discrete, no overlap with #202's media-source surface or #235's `App.tsx`/`libraries.ts` surface. Ideal next-free-session priority pickups; can even be one PR (both touch `Scheduling/Commands` replace handlers). - **#253** — the concurrency contract; recommend a decision on timing (pre-ossification vs #197-gate) before building, and a Fable/[PLAN-MODE] design pass given its cross-cutting blast radius. Strong #197 input either way. - Arc item 3 **#202** remains the parallel session's (claimed 13:58, PLAN-MODE). #235 still open with the App.tsx-coordination caveat. - No `in-progress` labels to remove (reviewer audits claimed by comment; no ersatztv issue was labeled this session).
Author
Owner

Arc change (amends the "Arc change: none" in the session comment above) — body edited on Timothy's call to integrate the audit findings into the plan at priority (no implementation this session; work will restart fresh from the handoff):

  • Added a "Priority pickups" block after the numbered arc: #251 + #252 (running-product scheduling correctness, backend-only, parallel-safe vs #202/#235) and #253 (replace-all concurrency contract — Fable/[PLAN-MODE] design pass, strong #197 input, "lock before more editors ossify"). Explicitly flagged that #252 is not the small fix I first triaged — its correct stable-identity reconcile shares the #253/F3 surface (polymorphic TPH item, no client-sent Id), so it should either fold into #253 or ship a scoped positional-reconcile; #251 ships regardless.
  • #197 (item 5) inputs now list #253's concurrency contract + the #254 tail.
  • Backlog line gains #254 (mutation-hardening cluster; several items fold into #172/#197/#230).
  • Pending-reviews snapshot refreshed: #21/#22/#23 = DELIVERED (→ #251–#254), #24 plan-review now the only PENDING pickup, #20's remaining item = #235.

in-progress labels I briefly added to #251/#252 have been removed; the exploratory worktree was discarded (nothing committed). No code changed.

**Arc change (amends the "Arc change: none" in the session comment above)** — body edited on Timothy's call to integrate the audit findings into the plan at priority (no implementation this session; work will restart fresh from the handoff): - Added a **"Priority pickups"** block after the numbered arc: **#251 + #252** (running-product scheduling correctness, backend-only, parallel-safe vs #202/#235) and **#253** (replace-all concurrency contract — Fable/[PLAN-MODE] design pass, strong #197 input, "lock before more editors ossify"). Explicitly flagged that **#252 is not the small fix I first triaged** — its correct stable-identity reconcile shares the #253/F3 surface (polymorphic TPH item, no client-sent Id), so it should either fold into #253 or ship a scoped positional-reconcile; #251 ships regardless. - **#197 (item 5) inputs** now list #253's concurrency contract + the #254 tail. - **Backlog** line gains **#254** (mutation-hardening cluster; several items fold into #172/#197/#230). - **Pending-reviews snapshot** refreshed: #21/#22/#23 = **DELIVERED** (→ #251–#254), #24 plan-review now the only PENDING pickup, #20's remaining item = #235. `in-progress` labels I briefly added to #251/#252 have been removed; the exploratory worktree was discarded (nothing committed). No code changed.
Author
Owner

Session 2026-07-11 — priority pickups #251 + #252 (mutation-safety fixes)

Closed: #251 + #252 (PR #258, merged 87060aa3). One backend-only PR — both are Scheduling/ProgramSchedules replace-handler correctness bugs from the audit sweep.

  • #252: ReplaceProgramScheduleItems no longer delete-and-reinserts on every save (which cascade-wiped PlayoutScheduleItemFillGroupIndex fill-group/shuffle state even on a no-op PUT). Now a positional in-place reconcile (SetValues reusing BuildItem; keep the item id → cascade never fires; subtype-change/surplus = delete+insert that slot). Scoped placeholder — true stable child identity is #253.
  • #251: deco / deco-template content edits + default-deco reassignment now enqueue BuildPlayout(Reset) for exactly the affected playouts (were enqueuing nothing → filler/break content stale until manual Reset). Whole post-commit invalidation runs on CancellationToken.None.

Filed: none.

Triage: no new issues this session (consumed the already-filed #251/#252).

Reviews: PR got a fork adversarial pass + independent Codex pass + a scoped re-review of the fix commit (the #242 discipline). Codex caught two Blocker-adjacent gaps the fork missed — a post-commit cancellation window in the deco queries (Medium) and an unguarded enqueue for a nonexistent playout in UpdateDefaultDeco (Low); both fixed in follow-up 51ac11e5 + re-reviewed clean. Codex also corrected a fact: schedule items are TPT, not TPH (reconcile safe either way). Pending-audit snapshot unchanged: #24 plan-review still PENDING pickup; #19 open; #20 remaining item = #235; #21/#22/#23 delivered (→ #251–#254). I picked up no reviewer-repo audit this session.

Arc change: none to ordering. Priority pickups #251 + #252 are now DONE. Remaining priority pickup = #253 (replace-all optimistic-concurrency contract — Fable/[PLAN-MODE] design; subsumes #252's stable-identity + #254 F3).

Recommended next (parallel-safe, distinct surfaces):

  • #253 — the concurrency-contract design pass (Fable/[PLAN-MODE]). ⚠️ It will build on / replace the positional reconcile I just added to ReplaceProgramScheduleItemsHandler — whoever takes #253 should treat that as the scoped placeholder for the stable-id reconcile, not re-derive it. Strong #197 input.
  • #235 — async-contract normalization + build observability (backend-heavy, but touches web/src/App.tsx + libraries.ts → coordinate with the #202 session churning App.tsx). #91b Libraries.razor gate.
  • #24 reviewer audit (plan review of the #243–#247 modularization epic) — read-only, parallel-safe.
  • Arc item #202 remains claimed/in-flight (worktrees 202-s1..s4 active).

CI note: PR head 51ac11e5 went green (Build & test ✓, EF migration ✓). The merge-commit run (87060aa3) shows cancelled/failure only because a docs-lore push (06c877b5) immediately superseded it — CI's concurrency guard cancelled the older run (superseded runs are ignorable, per lore). Current main 06c877b5 re-confirming green. in-progress labels removed from #251/#252; worktree cleaned; main pulled + lore appended (06c877b5).

## Session 2026-07-11 — priority pickups #251 + #252 (mutation-safety fixes) **Closed**: #251 + #252 (PR #258, merged `87060aa3`). One backend-only PR — both are `Scheduling`/`ProgramSchedules` replace-handler correctness bugs from the audit sweep. - **#252**: `ReplaceProgramScheduleItems` no longer delete-and-reinserts on every save (which cascade-wiped `PlayoutScheduleItemFillGroupIndex` fill-group/shuffle state even on a no-op PUT). Now a **positional in-place reconcile** (`SetValues` reusing `BuildItem`; keep the item id → cascade never fires; subtype-change/surplus = delete+insert that slot). Scoped placeholder — true stable child identity is **#253**. - **#251**: deco / deco-template content edits + default-deco reassignment now enqueue `BuildPlayout(Reset)` for exactly the affected playouts (were enqueuing nothing → filler/break content stale until manual Reset). Whole post-commit invalidation runs on `CancellationToken.None`. **Filed**: none. **Triage**: no new issues this session (consumed the already-filed #251/#252). **Reviews**: PR got a fork adversarial pass **+ independent Codex pass + a scoped re-review of the fix commit** (the #242 discipline). Codex caught **two Blocker-adjacent gaps the fork missed** — a post-commit cancellation window in the deco queries (Medium) and an unguarded enqueue for a nonexistent playout in `UpdateDefaultDeco` (Low); both fixed in follow-up `51ac11e5` + re-reviewed clean. Codex also corrected a fact: schedule items are **TPT, not TPH** (reconcile safe either way). **Pending-audit snapshot unchanged**: #24 plan-review still PENDING pickup; #19 open; #20 remaining item = #235; #21/#22/#23 delivered (→ #251–#254). I picked up no reviewer-repo audit this session. **Arc change**: none to ordering. **Priority pickups #251 + #252 are now DONE.** Remaining priority pickup = **#253** (replace-all optimistic-concurrency contract — Fable/[PLAN-MODE] design; subsumes #252's stable-identity + #254 F3). **Recommended next** (parallel-safe, distinct surfaces): - **#253** — the concurrency-contract design pass (Fable/[PLAN-MODE]). ⚠️ It will build on / replace the **positional reconcile I just added to `ReplaceProgramScheduleItemsHandler`** — whoever takes #253 should treat that as the scoped placeholder for the stable-id reconcile, not re-derive it. Strong #197 input. - **#235** — async-contract normalization + build observability (backend-heavy, but touches `web/src/App.tsx` + `libraries.ts` → coordinate with the #202 session churning App.tsx). #91b `Libraries.razor` gate. - **#24** reviewer audit (plan review of the #243–#247 modularization epic) — read-only, parallel-safe. - Arc item **#202** remains claimed/in-flight (worktrees `202-s1..s4` active). **CI note**: PR head `51ac11e5` went green (Build & test ✓, EF migration ✓). The merge-commit run (`87060aa3`) shows `cancelled`/`failure` **only because a docs-lore push (`06c877b5`) immediately superseded it** — CI's concurrency guard cancelled the older run (superseded runs are ignorable, per lore). Current main `06c877b5` re-confirming green. `in-progress` labels removed from #251/#252; worktree cleaned; main pulled + lore appended (`06c877b5`).
Author
Owner

Session 2026-07-11 — reviewer audit pickup: plan review of modularization epic (adversarial-reviewer#24)

Closed: none (read-only plan review — adversarial-reviewer#24 deliverable posted; that issue stays open for your dispatch, per the #20/#21-#23 pattern).

Filed: none new this session (worked an existing reviewer audit).

Triage (verdicts for the three bugs filed 15:06 by the #202 pre-review, not yet triaged in a tracker comment — the 15:30 #251/#252 session correctly reported "Filed: none" as they weren't its work):

  • #255 (High — path-replacement UPDATE SQL not source-scoped; cross-source overwrite via the new write API) → scoped INTO arc item 3 #202, not an independent pickup. Body says fix ships in #202 design doc §C4b. Latent under Blazor (single-source rows only); becomes a real cross-tenant mutation vector once #202's REST write API lands → must ship with #202, not after.
  • #256 (High — media-source sign-out/disconnect leak EntityLocker on exception, no finally → permanent 409) → scoped INTO #202 (§C6). Same lock-lifecycle class as #231–#235; #202's design has the controller acquire + handler release, so the missing finally is load-bearing there.
  • #257 (Medium — Plex library-preferences save releases library lock before the network scan runs; Unlock ordering) → scoped INTO #202 (§C7).
  • Net: all three are owned by the in-flight #202 (filed "for record"), so no independent pickup and not parallel-safe to touch — flagging so no future session double-works them. They raise #202's own review bar (the #202 close should verify §C4b/§C6/§C7 land with fault-injection + cross-source tests).

Reviews (pending-audit snapshot):

  • Picked up + DELIVERED: adversarial-reviewer#24 — plan review of the App.tsx modularization epic #243–#247. Verdict: PLAN SOUND, proceed. #244 (landed) passes the per-PR evidence bar on all counts (single-line App delta, self-contained colocated behavior tests incl. #212 regression, no compat shim → independently revertible). Delivered 5 must-honor preservation invariants (INV-1 route-object reference stability for the §2 Object.is bail; INV-2 keep navigate/popstate/currentPathRef/activeRoute in the composition root; INV-3 preserve Playouts mutatingRef+refresh()+409 coupling — the one §9 temporal hotspot; INV-4 PlayoutsRouteScreen sub-path unit; INV-5 Guide 60s now-clock is the lone physical timer in App.tsx), one timing flag, three plan gaps. #230 confirmed settled in-code (PR #242); the heavy temporal machinery already lives in api/*.ts hooks (de-risks #246/most of #245).
  • Snapshot now: #24 = DELIVERED (open, awaiting your dispatch). Still open: #19 (pre-review, next slice = #202 API), #20 (remaining item = #235). #21/#22/#23 delivered → #251–#254 (#251/#252 now closed via PR #258; #253/#254 open). No PENDING-pickup audits remain in adversarial-reviewer.

Arc change: none. (Two recommendations for the modularization epic — surfaced by the plan review, NOT arc changes, your call whether to fold in: (a) split #246 so Dashboard+Guide are now-eligible but Libraries is explicitly gated behind #202's merge — they collide on the same App.tsx byte range ~1487-1710 and #202 has already started, violating #246's own "not concurrently"; (b) promote #238 to a hard blocking prerequisite of #247 — #247's CustomEvent removal is correctly ordered after #238, but #238 is open/unowned/backlog, so #247 currently depends on unscheduled work.)

Recommended next (parallel-safe tracks, distinct surfaces):

  • #253 — replace-all optimistic-concurrency contract (Fable/[PLAN-MODE] design pass). Top open priority-pickup; design-only, parallel-safe; strong #197 input. Builds on / replaces the positional reconcile in ReplaceProgramScheduleItemsHandler (PR #258) — treat that as the scoped placeholder, don't re-derive.
  • #235 — async-contract normalization + build observability. Backend-heavy but touches web/src/App.tsx:3064-3109 + libraries.ts → coordinate with the #202 session (and note it's the #91b Libraries.razor gate). Also collides with a future #246-Libraries extraction.
  • arc item 3 #202 remains claimed/in-flight (owns #255/#256/#257).
  • Modularization epic phases #245/#246-Dashboard+Guide are eligible if that track is resumed (rebase off 533c6c1e+); hold #246-Libraries per the flag above.

No in-progress labels to remove (reviewer audit claimed by comment only; no ersatztv issue was labeled this session).

## Session 2026-07-11 — reviewer audit pickup: plan review of modularization epic (adversarial-reviewer#24) **Closed**: none (read-only plan review — adversarial-reviewer#24 deliverable posted; that issue stays open for your dispatch, per the #20/#21-#23 pattern). **Filed**: none new this session (worked an existing reviewer audit). **Triage** (verdicts for the three bugs filed 15:06 by the #202 pre-review, not yet triaged in a tracker comment — the 15:30 #251/#252 session correctly reported "Filed: none" as they weren't its work): - **#255** (High — path-replacement UPDATE SQL not source-scoped; cross-source overwrite via the new write API) → **scoped INTO arc item 3 #202**, not an independent pickup. Body says fix ships in #202 design doc §C4b. Latent under Blazor (single-source rows only); becomes a real cross-tenant mutation vector once #202's REST write API lands → must ship *with* #202, not after. - **#256** (High — media-source sign-out/disconnect leak EntityLocker on exception, no `finally` → permanent 409) → **scoped INTO #202** (§C6). Same lock-lifecycle class as #231–#235; #202's design has the controller acquire + handler release, so the missing `finally` is load-bearing there. - **#257** (Medium — Plex library-preferences save releases library lock before the network scan runs; Unlock ordering) → **scoped INTO #202** (§C7). - Net: all three are **owned by the in-flight #202** (filed "for record"), so no independent pickup and **not parallel-safe to touch** — flagging so no future session double-works them. They raise #202's own review bar (the #202 close should verify §C4b/§C6/§C7 land with fault-injection + cross-source tests). **Reviews** (pending-audit snapshot): - **Picked up + DELIVERED: adversarial-reviewer#24** — plan review of the App.tsx modularization epic #243–#247. **Verdict: PLAN SOUND, proceed.** #244 (landed) passes the per-PR evidence bar on all counts (single-line App delta, self-contained colocated behavior tests incl. #212 regression, no compat shim → independently revertible). Delivered 5 must-honor preservation invariants (INV-1 route-object reference stability for the §2 Object.is bail; INV-2 keep navigate/popstate/currentPathRef/activeRoute in the composition root; INV-3 preserve Playouts `mutatingRef`+`refresh()`+409 coupling — the one §9 temporal hotspot; INV-4 PlayoutsRouteScreen sub-path unit; INV-5 Guide 60s now-clock is the lone physical timer in App.tsx), one timing flag, three plan gaps. #230 confirmed settled in-code (PR #242); the heavy temporal machinery already lives in `api/*.ts` hooks (de-risks #246/most of #245). - Snapshot now: **#24 = DELIVERED** (open, awaiting your dispatch). Still open: **#19** (pre-review, next slice = #202 API), **#20** (remaining item = #235). **#21/#22/#23** delivered → #251–#254 (#251/#252 now closed via PR #258; #253/#254 open). **No PENDING-pickup audits remain** in adversarial-reviewer. **Arc change**: none. (Two recommendations for the modularization epic — surfaced by the plan review, NOT arc changes, your call whether to fold in: (a) **split #246** so Dashboard+Guide are now-eligible but **Libraries is explicitly gated behind #202's merge** — they collide on the same App.tsx byte range ~1487-1710 and #202 has already started, violating #246's own "not concurrently"; (b) **promote #238 to a hard blocking prerequisite of #247** — #247's CustomEvent removal is correctly ordered after #238, but #238 is open/unowned/backlog, so #247 currently depends on unscheduled work.) **Recommended next** (parallel-safe tracks, distinct surfaces): - **#253** — replace-all optimistic-concurrency contract (Fable/[PLAN-MODE] design pass). Top open priority-pickup; design-only, parallel-safe; strong #197 input. Builds on / replaces the positional reconcile in `ReplaceProgramScheduleItemsHandler` (PR #258) — treat that as the scoped placeholder, don't re-derive. - **#235** — async-contract normalization + build observability. Backend-heavy but touches `web/src/App.tsx:3064-3109` + `libraries.ts` → **coordinate with the #202 session** (and note it's the #91b `Libraries.razor` gate). Also collides with a future #246-Libraries extraction. - **arc item 3 #202** remains claimed/in-flight (owns #255/#256/#257). - Modularization epic phases #245/#246-Dashboard+Guide are eligible if that track is resumed (rebase off `533c6c1e`+); hold #246-Libraries per the flag above. No `in-progress` labels to remove (reviewer audit claimed by comment only; no ersatztv issue was labeled this session).
Author
Owner

Follow-up to the session comment above — user approved both plan-review recommendations; actioned (modularization milestone only, no arc impact):

  • #246 body edited → split into Batch A (Dashboard+Guide, now-eligible) / Batch B (Libraries, hard-gated behind #202's merge); AC now forbids landing Libraries concurrently with #202.
  • #238 promoted to a hard prerequisite of #247 (added to the "ChicoryTV modularization" milestone; comment explains the gate). #247 body strengthened: #238 = hard prerequisite, plus the two silent-break invariants (stable route-object array for the §2 Object.is bail; keep navigate/popstate/currentPathRef/activeRoute in the composition root) added to scope + AC.
  • #243 epic comment records both refinements + the suggested order (#245 → #246-A → #246-B post-#202 → #238 → #247).

The numbered arc (the prize) is untouched — these are internal to the parallel modularization milestone. #253 (Fable/[PLAN-MODE] concurrency-contract design) remains the top open priority-pickup.

**Follow-up to the session comment above** — user approved both plan-review recommendations; actioned (modularization milestone only, no arc impact): - **#246** body edited → split into Batch A (`Dashboard`+`Guide`, now-eligible) / Batch B (`Libraries`, **hard-gated behind #202's merge**); AC now forbids landing Libraries concurrently with #202. - **#238** promoted to a **hard prerequisite of #247** (added to the "ChicoryTV modularization" milestone; comment explains the gate). #247 body strengthened: #238 = hard prerequisite, plus the two silent-break invariants (stable route-object array for the §2 Object.is bail; keep navigate/popstate/currentPathRef/activeRoute in the composition root) added to scope + AC. - **#243** epic comment records both refinements + the suggested order (#245 → #246-A → #246-B post-#202 → #238 → #247). The numbered arc (the prize) is untouched — these are internal to the parallel modularization milestone. #253 (Fable/[PLAN-MODE] concurrency-contract design) remains the top open priority-pickup.
Author
Owner

Session 2026-07-11 — #253 optimistic-concurrency contract (Fable/[PLAN-MODE] design pass)

Closed: none — #253 was a design-only pass. Design DELIVERED as the canonical spec (issue comment #253#issuecomment-8472); #253 stays open as the PR1→PR4 implementation arc. in-progress removed.

Filed: #259 — replace-all reconcile: content-aware stable child identity ("moved item inherits the wrong slot's fill-group/shuffle state"), split OUT of #253 per the ratified scope boundary.

Triage:

  • #259 → backlog / #197-prep (review/content/priority: medium). Correctness but contained — the acute state-wipe-on-every-save is already fixed by #258's positional reconcile; #259 is the subtler moved-item residual (#254 F3 + the deep half of #252). Not a #91b gate. Needs a DTO-level child-id contract (SPA round-trip) — own design pass when picked up.

Reviews:

  • Codex quota-exhausted this session (usage limit; retry ~16:39). The independent adversarial pass on the #253 design was done via a fresh cold subagent instead (documented fallback). It confirmed the core EF mechanic SOUND and surfaced 4 must-fix recipe refinements (H1 catch-swallow→422 in the two Playout handlers; H2 the concrete ToEither().BindAsync shape; H3 bulk-ExecuteUpdate/Delete writers bypass the bump → config-only boundary; M1 premise reword) — all folded into the delivered design §9.
  • No adversarial-reviewer audits picked up/filed. Snapshot unchanged: no PENDING-pickup audits remain (#24 delivered last session); #19 open (next slice = #202 API pre-review), #20 remaining item = #235. #21/#22/#23 delivered → #251–#254 (#251/#252 closed via #258; #253 design done this session; #254 open).

Arc change: none to ordering. Two framing corrections (noted here, not body-stored per protocol): (1) the body's priority-pickup line "#253 subsumes #252's stable-identity + #254's F3" is superseded — #253 is now scoped to the concurrency contract ONLY; child stable-identity is its own issue #259. (2) #253's status: design delivered + ratified + independently reviewed; the remaining work is the PR1→PR4 arc in the design comment.

Ratified this session (design judgment calls, Timothy): child-identity OUT (→#259); all mutating handlers of a versioned aggregate's editor-visible config state bump Version (config-only boundary — build output excluded); Phase-2 = 428 after SPA-echo + release soak; uniform int Version; shared save-guard helper (not interceptor); header-only ETag; PreconditionFailedError→412 (409 stays the lock guard).

Recommended next (parallel-safe tracks, distinct surfaces):

  • #253 PR1 (infra + Block reference aggregate) is now the top implementable priority-pickup — the design comment is the spec; PR1 MUST bake in the §9 H1/H2/H3/M1 refinements (it's the #197 input, review it hardest). ⚠️ Coordination: PR1 touches ApiResults.cs, web/src/api/client.ts, TvContext/dual-provider migration, and the Blocks editor → overlaps the #202 session (media-source, churning App.tsx/client) and #235 (App.tsx/libraries.ts). Sequence the shared-file touches or expect merge-main rounds.
  • #235 — async-contract normalization + build observability (backend-heavy; App.tsx/libraries.ts; #91b Libraries.razor gate). Coordinate with #202.
  • arc item #202 remains claimed/in-flight (owns #255/#256/#257).
  • Modularization epic #245/#246-A eligible if that track resumes (rebase off 533c6c1e+; hold #246-B Libraries behind #202).
## Session 2026-07-11 — #253 optimistic-concurrency contract (Fable/[PLAN-MODE] design pass) **Closed**: none — #253 was a **design-only** pass. Design DELIVERED as the canonical spec (issue comment [#253#issuecomment-8472](http://192.168.1.95:3000/timothy/ersatztv/issues/253#issuecomment-8472)); #253 stays **open** as the PR1→PR4 implementation arc. `in-progress` removed. **Filed**: **#259** — replace-all reconcile: content-aware stable child identity ("moved item inherits the wrong slot's fill-group/shuffle state"), split OUT of #253 per the ratified scope boundary. **Triage**: - **#259** → **backlog / #197-prep** (`review`/`content`/priority: medium). Correctness but contained — the acute state-wipe-on-every-save is already fixed by #258's positional reconcile; #259 is the subtler moved-item residual (#254 F3 + the deep half of #252). Not a #91b gate. Needs a DTO-level child-id contract (SPA round-trip) — own design pass when picked up. **Reviews**: - **Codex quota-exhausted** this session (usage limit; retry ~16:39). The independent adversarial pass on the #253 design was done via a **fresh cold subagent** instead (documented fallback). It confirmed the core EF mechanic SOUND and surfaced 4 must-fix recipe refinements (H1 catch-swallow→422 in the two Playout handlers; H2 the concrete `ToEither().BindAsync` shape; H3 bulk-`ExecuteUpdate/Delete` writers bypass the bump → config-only boundary; M1 premise reword) — all folded into the delivered design §9. - No adversarial-reviewer audits picked up/filed. Snapshot unchanged: **no PENDING-pickup audits remain** (#24 delivered last session); **#19** open (next slice = #202 API pre-review), **#20** remaining item = **#235**. #21/#22/#23 delivered → #251–#254 (#251/#252 closed via #258; #253 design done this session; #254 open). **Arc change**: none to ordering. Two framing corrections (noted here, not body-stored per protocol): (1) the body's priority-pickup line "**#253 subsumes #252's stable-identity + #254's F3**" is **superseded** — #253 is now scoped to the concurrency contract ONLY; child stable-identity is its own issue **#259**. (2) #253's status: design delivered + ratified + independently reviewed; the remaining work is the PR1→PR4 arc in the design comment. **Ratified this session** (design judgment calls, Timothy): child-identity OUT (→#259); all mutating handlers of a versioned aggregate's *editor-visible config state* bump `Version` (config-only boundary — build output excluded); Phase-2 = 428 after SPA-echo + release soak; uniform `int Version`; shared save-guard helper (not interceptor); header-only ETag; `PreconditionFailedError`→412 (409 stays the lock guard). **Recommended next** (parallel-safe tracks, distinct surfaces): - **#253 PR1** (infra + Block reference aggregate) is now the top *implementable* priority-pickup — the design comment is the spec; PR1 MUST bake in the §9 H1/H2/H3/M1 refinements (it's the #197 input, review it hardest). ⚠️ Coordination: PR1 touches `ApiResults.cs`, `web/src/api/client.ts`, `TvContext`/dual-provider migration, and the Blocks editor → overlaps the **#202** session (media-source, churning App.tsx/client) and **#235** (App.tsx/`libraries.ts`). Sequence the shared-file touches or expect merge-main rounds. - **#235** — async-contract normalization + build observability (backend-heavy; App.tsx/`libraries.ts`; #91b `Libraries.razor` gate). Coordinate with #202. - **arc item #202** remains claimed/in-flight (owns #255/#256/#257). - Modularization epic #245/#246-A eligible if that track resumes (rebase off `533c6c1e`+; hold #246-B Libraries behind #202).
Author
Owner

Session 2026-07-11 (evening) — #204 pattern-based legacy→SPA redirect matcher (#91b removal-plan prerequisite)

Closed: #204 (PR #260, merged to main 495e450e fast-forward; branch deleted). in-progress removed.

Filed: none.

Triage: no new issues of my own. For record — #259 (filed 16:16 by the in-flight #253 session: "content-aware stable child identity", split from #253) → backlog, #253-owned (content/priority:medium; it's the deferred true-stable-child-identity follow-up the #252 positional reconcile placeholdered — the #253 session owns its triage/scheduling). Not an arc item.

Reviews: no adversarial-reviewer audit picked up this session (none pending pickup — #24 was delivered earlier today; #19 open, next slice #202 API; #20 remaining item #235). PR #260 got a fork adversarial pass = MERGEABLE (no Blocker/High/Medium; collision-freedom, query-merge, empty-segment, strict-{id}, Startup-wiring all held under tracing) + a live E2E (14/14 curl probes on the running host — the real net for the Startup wiring). Fork's one nit (guard meta-test covered only Tier-1 Map keys, not Tier-2 templates — and that invariant is load-bearing for the un-prefix-guarded /api|/artwork|/docs|/openapi surface) fixed in follow-up commit before merge. Did not spend a Codex pass — pure-logic change, fully covered by unit + live E2E + fork; held Codex quota. [PLAN-MODE] design escalated to Fable per the kickoff (mechanism/query-collision/id-validation/ordering); decision recorded in docs/decisions.md (2026-07-11, #204).

Arc change: none to ordering. This is progress inside arc item 4 (#91 phase b removal PR), not a reorder: #204 was the redirect-mechanism MUST-FIX blocker — the matcher can now express every parameterized legacy route (two-tier: 52 exact + 36 segment-template rules; strict positive-int {id}; AppendQueryString &-merge). Every PARITY-OK route now redirects (blazor-route-parity.md §2 emptied → §1). Removal PR remaining gates unchanged: #235's F9 (deep-scan + external-collections-scan have no API → Libraries.razor can't be deleted), #204's deferred catch-all fallback (replaces MapFallbackToPage("/_Host") — belongs in the removal PR itself, can't land while Blazor serves un-migrated routes), plus #204(redirects)+#205(rollback tag)+#206(auth posture) folded into that PR.

Recommended next (parallel-safe, distinct surfaces — #202 and #253 remain the other two sessions' in-flight claims):

  • #205 + #206 — the other two #91b removal-prep issues (pre-removal rollback tag; auth-posture step). Planning-heavy, disjoint from all in-flight code surfaces. Natural continuation of the removal-prep track #204 just advanced.
  • #235 — async-contract normalization + build observability (the last reviewer#20 item; also a Libraries.razor F9 gate) — but it edits web/src/App.tsx:3064-3109 + libraries.ts, so coordinate with the #202 session churning App.tsx (and any resumed #246-Libraries extraction).
  • A reviewer audit if any new one appears (none pending now).

Main build+smoke CI on 495e450e confirming green post-merge (PR test job already green + local build 0-err + live E2E booted the host).

## Session 2026-07-11 (evening) — #204 pattern-based legacy→SPA redirect matcher (#91b removal-plan prerequisite) **Closed**: #204 (PR #260, merged to main `495e450e` fast-forward; branch deleted). `in-progress` removed. **Filed**: none. **Triage**: no new issues of my own. For record — **#259** (filed 16:16 by the in-flight #253 session: "content-aware stable child identity", split from #253) → **backlog, #253-owned** (content/priority:medium; it's the deferred true-stable-child-identity follow-up the #252 positional reconcile placeholdered — the #253 session owns its triage/scheduling). Not an arc item. **Reviews**: no adversarial-reviewer audit picked up this session (none pending pickup — #24 was delivered earlier today; #19 open, next slice #202 API; #20 remaining item #235). PR #260 got a **fork adversarial pass = MERGEABLE** (no Blocker/High/Medium; collision-freedom, query-merge, empty-segment, strict-`{id}`, Startup-wiring all held under tracing) + a **live E2E** (14/14 curl probes on the running host — the real net for the Startup wiring). Fork's one nit (guard meta-test covered only Tier-1 Map keys, not Tier-2 templates — and that invariant is load-bearing for the un-prefix-guarded /api|/artwork|/docs|/openapi surface) fixed in follow-up commit before merge. Did **not** spend a Codex pass — pure-logic change, fully covered by unit + live E2E + fork; held Codex quota. `[PLAN-MODE]` design escalated to **Fable** per the kickoff (mechanism/query-collision/id-validation/ordering); decision recorded in `docs/decisions.md` (2026-07-11, #204). **Arc change**: none to ordering. This is **progress inside arc item 4** (#91 phase b removal PR), not a reorder: #204 was the redirect-mechanism MUST-FIX blocker — the matcher can now express every parameterized legacy route (two-tier: 52 exact + 36 segment-template rules; strict positive-int `{id}`; `AppendQueryString` `&`-merge). Every PARITY-OK route now redirects (`blazor-route-parity.md` §2 emptied → §1). **Removal PR remaining gates unchanged**: #235's F9 (deep-scan + external-collections-scan have no API → `Libraries.razor` can't be deleted), #204's deferred **catch-all fallback** (replaces `MapFallbackToPage("/_Host")` — belongs in the removal PR itself, can't land while Blazor serves un-migrated routes), plus #204(redirects)+#205(rollback tag)+#206(auth posture) folded into that PR. **Recommended next** (parallel-safe, distinct surfaces — #202 and #253 remain the other two sessions' in-flight claims): - **#205 + #206** — the other two #91b removal-prep issues (pre-removal rollback tag; auth-posture step). Planning-heavy, disjoint from all in-flight code surfaces. Natural continuation of the removal-prep track #204 just advanced. - **#235** — async-contract normalization + build observability (the last reviewer#20 item; also a `Libraries.razor` F9 gate) — but it edits `web/src/App.tsx:3064-3109` + `libraries.ts`, so **coordinate with the #202 session** churning App.tsx (and any resumed #246-Libraries extraction). - A reviewer audit if any new one appears (none pending now). Main build+smoke CI on `495e450e` confirming green post-merge (PR `test` job already green + local build 0-err + live E2E booted the host).
Author
Owner

Session 2026-07-11 — #91b removal-PR prep: auth-posture sign-off + rollback tag (#205 + #206)

Closed: #205 + #206 (PR #261, merged 7608ccce; CI green, FF, branch deleted). Docs-only. Picked this pair (user's call from a 4-option menu) because the top open arc item #235 collides with the in-flight #202 (both own the Libraries/App.tsx+libraries.ts surface) and #205/#206 are removal-PR groundwork that touches no active session's code.

Filed: none.

Triage: no new issues this session.

Reviews: no adversarial-reviewer audits picked up/filed. Snapshot unchanged: no PENDING-pickup audits remain; #19 open (next slice = #202 API pre-review), #20 remaining item = #235, #21/#22/#23 delivered → #251–#254, #24 delivered. PR #261's auth-posture claim got an independent cold-reviewer pass (a fresh zero-context subagent, per the #91-gate discipline) — verified every wiring claim + line number against source, verdict SIGN-OFF SOUND, no counterexample.

Arc change: none to ordering. Two notes:

  • #204 merged during this window (8a2238b6/495e450e — id-carrying pattern redirects). PR #261 was FF-rejected, rebased over it, and I updated the new runbook so #204 is no longer listed as an open removal gate (its catch-all fallback folds into removal Step 2). Redirect-map restructure is #204's; my change is docs-append + that correction.
  • New durable artifact: docs/blazor-route-parity.md §5 "Removal execution runbook" — the removal PR (arc item 4) now has an ordered, in-doc checklist (rollback tag → surgical Blazor delete + catch-all fallback → auth sign-off → verify/docs), replacing the plan that lived only in #91 comments. Both #205/#206 sign-offs live in docs/decisions.md (2026-07-11).

Removal-gate status after this session: milestone 7's #205/#206 closed; the removal PR itself remains gated on #202 (media-source write API+SPA), #235 F9 (deep-scan/external-collections-scan API — Libraries.razor blocker), and the mandatory cold adversarial pass. #204 done.

Recommended next (parallel-safe, distinct surfaces — active sessions as of now: #202 implementing in worktrees 202-s6a-local/202-s6b-remote = App.tsx/libraries live; #253 PR1 in 253-pr1):

  • #235 — async-contract + build observability + F9 (the last removal gate besides #202). ⚠️ Still collides with #202 on App.tsx:~3064/libraries.ts — hold until #202 lands or coordinate tightly.
  • #254 — backend mutation-hardening cluster (low-sev tail, backend-isolated, CancellationToken.None sweep etc.). Fully parallel-safe vs all three active sessions; good #197-prep.
  • #259 — content-aware stable child identity (backend). Note it overlaps ReplaceProgramScheduleItemsHandler, which #253 PR1 is touching — sequence after #253 PR1 lands.
  • Removal PR (arc item 4) is not yet startable — gated on #202 + #235 F9 + cold pass.

in-progress labels removed from #205/#206; worktree cleaned; main pulled to 7608ccce.

## Session 2026-07-11 — #91b removal-PR prep: auth-posture sign-off + rollback tag (#205 + #206) **Closed**: #205 + #206 (PR #261, merged `7608ccce`; CI green, FF, branch deleted). Docs-only. Picked this pair (user's call from a 4-option menu) because the top open arc item **#235 collides with the in-flight #202** (both own the Libraries/`App.tsx`+`libraries.ts` surface) and #205/#206 are removal-PR groundwork that touches no active session's code. **Filed**: none. **Triage**: no new issues this session. **Reviews**: no adversarial-reviewer audits picked up/filed. Snapshot **unchanged**: no PENDING-pickup audits remain; #19 open (next slice = #202 API pre-review), #20 remaining item = #235, #21/#22/#23 delivered → #251–#254, #24 delivered. PR #261's auth-posture claim got an **independent cold-reviewer pass** (a fresh zero-context subagent, per the #91-gate discipline) — verified every wiring claim + line number against source, verdict SIGN-OFF SOUND, no counterexample. **Arc change**: none to ordering. Two notes: - **#204 merged during this window** (`8a2238b6`/`495e450e` — id-carrying pattern redirects). PR #261 was FF-rejected, rebased over it, and I updated the new runbook so #204 is no longer listed as an open removal gate (its catch-all fallback folds into removal Step 2). Redirect-map restructure is #204's; my change is docs-append + that correction. - **New durable artifact**: `docs/blazor-route-parity.md` **§5 "Removal execution runbook"** — the removal PR (arc item 4) now has an ordered, in-doc checklist (rollback tag → surgical Blazor delete + catch-all fallback → auth sign-off → verify/docs), replacing the plan that lived only in #91 comments. Both #205/#206 sign-offs live in `docs/decisions.md` (2026-07-11). **Removal-gate status after this session**: milestone 7's #205/#206 **closed**; the removal PR itself remains gated on **#202** (media-source write API+SPA), **#235 F9** (deep-scan/external-collections-scan API — `Libraries.razor` blocker), and the **mandatory cold adversarial pass**. #204 done. **Recommended next** (parallel-safe, distinct surfaces — active sessions as of now: **#202** implementing in worktrees `202-s6a-local`/`202-s6b-remote` = App.tsx/libraries live; **#253 PR1** in `253-pr1`): - **#235** — async-contract + build observability + F9 (the last removal gate besides #202). ⚠️ Still collides with #202 on `App.tsx:~3064`/`libraries.ts` — **hold until #202 lands** or coordinate tightly. - **#254** — backend mutation-hardening cluster (low-sev tail, backend-isolated, `CancellationToken.None` sweep etc.). Fully parallel-safe vs all three active sessions; good #197-prep. - **#259** — content-aware stable child identity (backend). Note it overlaps `ReplaceProgramScheduleItemsHandler`, which **#253 PR1 is touching** — sequence after #253 PR1 lands. - Removal PR (arc item 4) is **not yet startable** — gated on #202 + #235 F9 + cold pass. `in-progress` labels removed from #205/#206; worktree cleaned; main pulled to `7608ccce`.
Author
Owner

Session 2026-07-11 — #202 media-source management + SPA (arc item 3)

Closed: #202 (PR #262, merged 8e2f9954) · #255 + #256 + #257 (owned pre-existing bug fixes shipped within #262)
Filed: #264 — local libraries show "Never scanned" (local scanner sets path-level LastScan, not library.LastScan; read API reads the latter)
Triage: #264 → backlog (Low, pre-existing scanner display bug surfaced by the SPA hub; NOT a #202 regression — item counts update correctly). #255/#256 were High (cross-source mutation + lock-wedge) but reachable only via the new REST surface #202 introduced, so fixed in-PR rather than slotted separately.
Reviews: adversarial-reviewer#19 slice 2 DELIVERED — #202 API design pre-reviewed (Codex: CHANGES REQUIRED, 4 blockers/6 highs, all reconciled; then a fresh independent re-review → MERGEABLE). Slices 3/4 (#91b removal, #197) remain queued on #19. Pending audits unchanged: #21/#22/#23 still PENDING pickup; #20 open (its #233/#234/#235 remainder). Also ran an adversarial diff review (SHIP, 2 findings fixed) + live local E2E before merge.
Arc change: none to ordering. Arc item 3 (#202) is DONE. The media-source cluster no longer blocks #91b (item 4); #91b's remaining gates are unchanged (#235 F9 deep-scan/external-collections-scan API + the mandatory cold adversarial pass). Redirects for the 14 /media/sources/* routes deferred to the #91b removal PR (parity doc Section 2).
Recommended next: with #202 done, the arc's forward edge is #91b prep (item 4 — now only #235 F9 + cold adversarial pass remain; #204/#205/#206 redirect-matcher/auth-posture/rollback-tag landed on main this session via a parallel session) and #197 (item 5, HARD GATE). Reviewer audits #21/#22/#23 are unclaimed and parallel-safe. Cross-session note: main advanced twice during this session (#251/#252 then #204/#205/#206) — merge main into any open branch; the parity doc Section 2 needed a prose reconciliation after the #204 merge (its "section empty" framing vs #202's media-source rows).

Manual E2E remaining (needs the homelab, handed to Timothy): real Jellyfin/Emby server connect (sync-prefs + path-replacements + the secure key affordance) and the Plex interactive pin-flow (poll-until-lock-releases / "finalizing" state). Checklist relayed in-conversation. Local-library CRUD + scan + dirty-guard + secure-GET-shape were E2E-verified automatically. in-progress removed from #202.

## Session 2026-07-11 — #202 media-source management + SPA (arc item 3) **Closed**: #202 (PR #262, merged `8e2f9954`) · #255 + #256 + #257 (owned pre-existing bug fixes shipped within #262) **Filed**: #264 — local libraries show "Never scanned" (local scanner sets path-level `LastScan`, not `library.LastScan`; read API reads the latter) **Triage**: #264 → **backlog** (Low, pre-existing scanner display bug surfaced by the SPA hub; NOT a #202 regression — item counts update correctly). #255/#256 were **High** (cross-source mutation + lock-wedge) but reachable only via the new REST surface #202 introduced, so fixed in-PR rather than slotted separately. **Reviews**: **adversarial-reviewer#19 slice 2 DELIVERED** — #202 API design pre-reviewed (Codex: CHANGES REQUIRED, 4 blockers/6 highs, all reconciled; then a fresh independent re-review → MERGEABLE). Slices 3/4 (#91b removal, #197) remain queued on #19. Pending audits unchanged: **#21/#22/#23 still PENDING pickup**; #20 open (its #233/#234/#235 remainder). Also ran an adversarial diff review (SHIP, 2 findings fixed) + live local E2E before merge. **Arc change**: none to ordering. **Arc item 3 (#202) is DONE.** The media-source cluster no longer blocks #91b (item 4); #91b's remaining gates are unchanged (#235 F9 deep-scan/external-collections-scan API + the mandatory cold adversarial pass). Redirects for the 14 `/media/sources/*` routes deferred to the #91b removal PR (parity doc Section 2). **Recommended next**: with #202 done, the arc's forward edge is **#91b prep** (item 4 — now only #235 F9 + cold adversarial pass remain; #204/#205/#206 redirect-matcher/auth-posture/rollback-tag landed on main this session via a parallel session) and **#197** (item 5, HARD GATE). Reviewer audits **#21/#22/#23** are unclaimed and parallel-safe. **Cross-session note**: main advanced twice during this session (#251/#252 then #204/#205/#206) — merge main into any open branch; the parity doc Section 2 needed a prose reconciliation after the #204 merge (its "section empty" framing vs #202's media-source rows). **Manual E2E remaining** (needs the homelab, handed to Timothy): real Jellyfin/Emby server connect (sync-prefs + path-replacements + the secure key affordance) and the Plex interactive pin-flow (poll-until-lock-releases / "finalizing" state). Checklist relayed in-conversation. Local-library CRUD + scan + dirty-guard + secure-GET-shape were E2E-verified automatically. `in-progress` removed from #202.
Author
Owner

Session 2026-07-11 — #253 PR1: optimistic-concurrency contract (infra + Block reference)

Closed: none outright — #253 PR1 landed (PR #263, merged to main 2de091ea); #253 stays open as the PR2→PR4 fan-out arc. in-progress removed from #253.

Filed: #265 — If-Match RFC 7232 412-vs-400 status semantics (non-canonical/weak/list tags → 400 today; stricter valid-but-non-matching → 412 refinement deferred).

Triage:

  • #265 → backlog / #197-prep (Medium, api/review). Fail-safe today (a non-canonical If-Match is rejected 400, never a silent overwrite); the correct fix needs RFC 7232 evaluation ordering + weak-tag comparison + list support → belongs in the #197 cold contract pass, which #253 already feeds. Not a gate.

Reviews: no adversarial-reviewer audits picked up this session (none pending — all delivered). PR #263 got an independent fork adversarial pass (SHIP) + two Codex passes (codex:rescue): the initial pass found a High (client load-TOCTOU: root+items read concurrently → stale-root/current-ETag silent overwrite) + Medium (non-canonical If-Match coerced to a version); both fixed; the fix-commit re-review (the #242 discipline) confirmed both resolved and raised the one new Medium → deferred as #265. Snapshot unchanged otherwise.

Arc change: none to ordering. The top implementable priority-pickup (#253 PR1) is DONE; #253's remaining PR2–PR4 are the fan-out. During this session main advanced to include #204 + #205/#206 (another session) — merged main into the PR branch (only docs/decisions.md conflicted, both append-only; reconciled + regenerated-artifact check clean).

Recommended next (parallel-safe, distinct surfaces):

  • #253 PR2 (RR + Reconcile fan-out: Template/DecoTemplate/Playlist/schedule-items) — mechanical copies of the Block recipe now frozen in api-conventions.md §7a; each aggregate's sibling config writers must bump Version (incl. bulk ExecuteUpdate/Delete via .SetProperty). PR3 = Diff+Scalar (incl. §9 H1 catch-rethrow in the two Playout handlers). ⚠️ Coordinates with #202 (media-source, still in-progress) + #235 on client.ts/generated artifacts/migrations — merge main + regenerate.
  • #235 (async-contract normalization; App.tsx/libraries.ts; #91b Libraries.razor gate) — coordinate with #202.
  • #202 remains claimed/in-flight (owns #255/#256/#257).
  • Modularization #245/#246-A eligible if that track resumes (hold #246-B Libraries behind #202).
## Session 2026-07-11 — #253 PR1: optimistic-concurrency contract (infra + Block reference) **Closed**: none outright — **#253 PR1 landed** (PR #263, merged to main `2de091ea`); #253 stays open as the PR2→PR4 fan-out arc. `in-progress` removed from #253. **Filed**: #265 — If-Match RFC 7232 412-vs-400 status semantics (non-canonical/weak/list tags → 400 today; stricter valid-but-non-matching → 412 refinement deferred). **Triage**: - **#265** → **backlog / #197-prep** (Medium, `api`/`review`). Fail-safe today (a non-canonical If-Match is *rejected* 400, never a silent overwrite); the correct fix needs RFC 7232 evaluation ordering + weak-tag comparison + list support → belongs in the #197 cold contract pass, which #253 already feeds. Not a gate. **Reviews**: no adversarial-reviewer audits picked up this session (none pending — all delivered). PR #263 got an independent **fork** adversarial pass (SHIP) **+ two Codex passes** (`codex:rescue`): the initial pass found a **High** (client load-TOCTOU: root+items read concurrently → stale-root/current-ETag silent overwrite) + **Medium** (non-canonical If-Match coerced to a version); both fixed; the **fix-commit re-review** (the #242 discipline) confirmed both resolved and raised the one new Medium → deferred as #265. Snapshot unchanged otherwise. **Arc change**: none to ordering. The top *implementable* priority-pickup (#253 PR1) is **DONE**; #253's remaining PR2–PR4 are the fan-out. During this session main advanced to include #204 + #205/#206 (another session) — merged main into the PR branch (only `docs/decisions.md` conflicted, both append-only; reconciled + regenerated-artifact check clean). **Recommended next** (parallel-safe, distinct surfaces): - **#253 PR2** (RR + Reconcile fan-out: Template/DecoTemplate/Playlist/schedule-items) — mechanical copies of the Block recipe now frozen in `api-conventions.md` §7a; each aggregate's sibling config writers must bump `Version` (incl. bulk `ExecuteUpdate/Delete` via `.SetProperty`). PR3 = Diff+Scalar (incl. §9 H1 catch-rethrow in the two Playout handlers). ⚠️ Coordinates with **#202** (media-source, still `in-progress`) + **#235** on `client.ts`/generated artifacts/migrations — merge main + regenerate. - **#235** (async-contract normalization; App.tsx/`libraries.ts`; #91b `Libraries.razor` gate) — coordinate with #202. - **#202** remains claimed/in-flight (owns #255/#256/#257). - Modularization #245/#246-A eligible if that track resumes (hold #246-B Libraries behind #202).
Author
Owner

Session 2026-07-11 (evening) — #254 backend mutation-hardening (post-commit token sweep + guide-xml + empty-list)

Closed: none outright — #254's backend-isolated core landed (PR #266, merged FF to main 13cd00c8; CI green on the merge head). #254 stays open for its low-severity residual (config/FFmpeg multi-upsert token case + F1 reindex reconcile; the rest of its named items have homes: F3→#259, F6→#230/#238/#248, F7+#21-F2→#197). Picked #254 as the recommended parallel-safe pickup — both gate items (#202, #253 PR1) were claimed/in-flight at session start.

Filed: none.

Triage: no new issues of my own. For record — #265 (If-Match 412-vs-400, filed 18:07 by the #253 PR1 session) was already triaged there → backlog/#197-prep; I confirmed my new api-conventions.md §7b and #253's §7a both cross-reference it, so it's doc-tracked. No change.

Reviews: no adversarial-reviewer audits picked up (none pending). PR #266 got an independent cold adversarial pass (fork, scoped "review only") = MERGEABLE-WITH-NITS: it caught a Medium — Create/DeleteSmartCollectionHandler had the same missed post-commit Refresh(cancellationToken) as the (already-swept) UpdateSmartCollectionHandler — folded into the PR (6055bd36) + completeness-grep-confirmed no other sites; and a Low (empty-list test asserts the guard, not the raw Max() path) left as-is with reasoning (non-vacuous, guard sits before the foreach). Did not spend a Codex pass — pure mechanical token sweep, fully covered by the diff-verify + 249/34 local tests + the fork; held Codex quota.

Arc change: none to ordering. Not an arc item — backend #197-prep hardening, parallel to the gates. Forward-edge note: with #202 (arc item 3) + #253 PR1 both landed on main during this session, the arc's forward edge is now #91b prep (item 4 — remaining gates: #235 F9 + the mandatory cold adversarial pass) and #197 (item 5, HARD GATE).

Parallel-merge handled: main advanced under me (#202 8e2f9954 + #253 PR1 2de091ea). Merged main in — only two conflicts, both doc-appends. Section-NUMBER collision: #253 and I both added an api-conventions.md §7a — renumbered mine to §7b (post-commit token) after #253's §7a (concurrency), and fixed my decisions.md prose cross-ref §7a→§7b (the "re-read prose cross-refs after a parallel merge" lesson, now extended: section identifiers collide too, not just prose). Main didn't touch any swept handler (0 lines) — code merged clean; re-verified 131 tests green on the merged tree.

Recommended next (parallel-safe, distinct surfaces — #253 PR2–PR4 fan-out is the other likely active track):

  • #235 — async-contract + build observability + F9 (last #91b gate besides the cold pass). Now that #202 landed, its App.tsx/libraries.ts coordination is against main (not an in-flight branch) — cleaner to pick up. ⚠️ still coordinate with a resumed #246-Libraries and #253 PR2's client.ts touches.
  • #253 PR2–PR4 — RR/Reconcile/Diff fan-out; each versioned aggregate's config writers bump Version. Heads-up: several of those handlers now carry my post-commit CancellationToken.None convention (see decisions.md coordination note) — merge main, expect it present, add Version++ pre-commit alongside.
  • #254 residual (config/FFmpeg multi-upsert token; F1) — backend-isolated, low-pri #197-prep.
  • #259 — content-aware stable child identity (backend; overlaps ReplaceProgramScheduleItemsHandler).

Infra note (server-management-owned, user is investigating): bumblebee (192.168.1.99) went host-down mid-session (ping loss + 443/80/22/8409 all closed) → public gitea.tblindustries.be unreachable (Gitea backend .95 fine) and CI capacity degraded to the VM runner only. Consequence: the concurrent main push 8e2f9954 (run 464) failed on the runner dying mid-build — not a code break; main's CI wants a rerun once bumblebee/runners recover. My PR runs drained fine on the VM runner. in-progress removed from #254; worktree retained pending main-CI confirmation.

## Session 2026-07-11 (evening) — #254 backend mutation-hardening (post-commit token sweep + guide-xml + empty-list) **Closed**: none outright — **#254's backend-isolated core landed** (PR #266, merged FF to main `13cd00c8`; CI green on the merge head). #254 stays **open** for its low-severity residual (config/FFmpeg multi-upsert token case + F1 reindex reconcile; the rest of its named items have homes: F3→#259, F6→#230/#238/#248, F7+#21-F2→#197). Picked #254 as the recommended parallel-safe pickup — both gate items (#202, #253 PR1) were claimed/in-flight at session start. **Filed**: none. **Triage**: no new issues of my own. For record — **#265** (If-Match 412-vs-400, filed 18:07 by the #253 PR1 session) was **already triaged there** → backlog/#197-prep; I confirmed my new `api-conventions.md` §7b and #253's §7a both cross-reference it, so it's doc-tracked. No change. **Reviews**: no adversarial-reviewer audits picked up (none pending). PR #266 got an independent **cold adversarial pass** (fork, scoped "review only") = **MERGEABLE-WITH-NITS**: it caught a **Medium** — `Create/DeleteSmartCollectionHandler` had the same missed post-commit `Refresh(cancellationToken)` as the (already-swept) `UpdateSmartCollectionHandler` — **folded into the PR** (`6055bd36`) + completeness-grep-confirmed no other sites; and a **Low** (empty-list test asserts the guard, not the raw `Max()` path) left as-is with reasoning (non-vacuous, guard sits before the foreach). Did **not** spend a Codex pass — pure mechanical token sweep, fully covered by the diff-verify + 249/34 local tests + the fork; held Codex quota. **Arc change**: none to ordering. Not an arc item — backend #197-prep hardening, parallel to the gates. **Forward-edge note**: with **#202 (arc item 3) + #253 PR1 both landed on main during this session**, the arc's forward edge is now **#91b prep (item 4** — remaining gates: #235 F9 + the mandatory cold adversarial pass) and **#197 (item 5, HARD GATE)**. **Parallel-merge handled**: main advanced under me (#202 `8e2f9954` + #253 PR1 `2de091ea`). Merged main in — only two conflicts, both doc-appends. **Section-NUMBER collision**: #253 and I *both* added an `api-conventions.md` §7a — renumbered mine to **§7b** (post-commit token) after #253's §7a (concurrency), and fixed my `decisions.md` prose cross-ref §7a→§7b (the "re-read prose cross-refs after a parallel merge" lesson, now extended: section *identifiers* collide too, not just prose). Main didn't touch any swept handler (0 lines) — code merged clean; re-verified 131 tests green on the merged tree. **Recommended next** (parallel-safe, distinct surfaces — #253 PR2–PR4 fan-out is the other likely active track): - **#235** — async-contract + build observability + F9 (last #91b gate besides the cold pass). Now that **#202 landed**, its `App.tsx`/`libraries.ts` coordination is against *main* (not an in-flight branch) — cleaner to pick up. ⚠️ still coordinate with a resumed #246-Libraries and #253 PR2's `client.ts` touches. - **#253 PR2–PR4** — RR/Reconcile/Diff fan-out; each versioned aggregate's config writers bump `Version`. **Heads-up**: several of those handlers now carry my post-commit `CancellationToken.None` convention (see `decisions.md` coordination note) — merge main, expect it present, add `Version++` pre-commit alongside. - **#254 residual** (config/FFmpeg multi-upsert token; F1) — backend-isolated, low-pri #197-prep. - **#259** — content-aware stable child identity (backend; overlaps `ReplaceProgramScheduleItemsHandler`). **Infra note (server-management-owned, user is investigating)**: **bumblebee (192.168.1.99) went host-down mid-session** (ping loss + 443/80/22/8409 all closed) → public `gitea.tblindustries.be` unreachable (Gitea backend `.95` fine) and **CI capacity degraded** to the VM runner only. Consequence: the concurrent **main push `8e2f9954` (run 464) failed on the runner dying mid-build** — not a code break; **main's CI wants a rerun** once bumblebee/runners recover. My PR runs drained fine on the VM runner. `in-progress` removed from #254; worktree retained pending main-CI confirmation.
Author
Owner

Session 2026-07-11 (evening) — #235 async-op contract + F9 scan endpoints (reviewer#20 last item)

Closed: #235 (PR #267, merged 4aeabada fast-forward; CI green, branch deleted). The last reviewer#20 batch item.

Filed: none.

Triage: no new issues.

Reviews: adversarial-reviewer#20 is now fully DELIVERED — its final open item #235 is done; #231/#232/#233/#234 already closed. No pending-pickup audits remain (#19 open, next slice = #91b/#197; #21/#22/#23 delivered → #251–#254; #24 delivered). PR #267 got a cold Claude fork pass (MERGEABLE) + independent Codex pass which caught a High the fork missed — a scan-collections lock cross-release (#250 class): SchedulerService enqueued periodic collection scans unlocked while ScannerService released the per-provider lock unconditionally-if-held, so a scheduled scan could cross-release an API scan's lock (spurious 202 vs 409; no data corruption — single-reader channel). Fork-vs-Codex disagreement → escalated to Fable per protocol; Fable ratified Codex + caught that the naive per-source fix would starve the 2nd+ media source, prescribing the §C7 lock-once-per-provider batch (Unlock: isLast). Fixed in-PR + Codex re-review of the fix commit = FIX CORRECT (the #242 discipline). Codex Low (make the QueueShowScanResult switch total) also folded in.

Arc change: none to ordering. Reviewer#20 batch fully DONE. Progress inside arc item 4 (#91b removal): the F9 API gate is now CLOSED — deep-scan (?deep=) + external-collections-scan (POST /api/media-sources/{plex|jellyfin|emby}/{id}/scan-collections) endpoints exist, so Libraries.razor is no longer API-blocked. Removal PR's remaining parity work is now just the SPA deep-scan / collections buttons (wiring the shipped libraries.ts clients into LibrariesScreen) + the mandatory cold adversarial pass (parity doc §5). #202, #204, #205, #206 all landed earlier today.

Removal-gate status: #91b removal PR (arc item 4) is now gated only on (a) SPA deep/collections buttons, (b) #204's catch-all fallback (belongs in the removal PR), (c) the mandatory cold adversarial pass. No remaining API gate.

Recommended next (parallel-safe, distinct surfaces):

  • arc item 4 (#91b removal PR) is now much closer to startable — the last API gate (F9) is closed. Remaining pre-work is SPA-side (deep/collections buttons) + the cold pass; the catch-all fallback lands in the removal PR itself.
  • #254 was merged to main this session by a parallel session (a1bd303c/6055bd36/13cd00c8) — I merged it into #267 before landing; its in-progress may still be set, check with that session.
  • #253 PR2+ (concurrency contract fan-out) and #259 (content-aware child identity) remain open on the scheduling surface.
  • #265 (If-Match 412-not-400, RFC 7232) — deferred #197 input, still open/unclaimed.

Cross-session notes: main advanced twice during this session (#253 PR1 #263, then #254). #267 merged both in and regenerated artifacts each time (no drift). Two new files on main: ErsatzTV.Tests/Services/{WorkerServiceTests,ScannerServiceCollectionLockTests,SchedulerServiceCollectionLockTests}.cs. in-progress removed from #235; feature + slice worktrees cleaned; main pulled to 4aeabada.

## Session 2026-07-11 (evening) — #235 async-op contract + F9 scan endpoints (reviewer#20 last item) **Closed**: #235 (PR #267, merged `4aeabada` fast-forward; CI green, branch deleted). The last reviewer#20 batch item. **Filed**: none. **Triage**: no new issues. **Reviews**: **adversarial-reviewer#20 is now fully DELIVERED** — its final open item #235 is done; #231/#232/#233/#234 already closed. No pending-pickup audits remain (#19 open, next slice = #91b/#197; #21/#22/#23 delivered → #251–#254; #24 delivered). PR #267 got a cold Claude fork pass (MERGEABLE) **+ independent Codex pass** which caught a **High the fork missed** — a scan-collections lock cross-release (#250 class): `SchedulerService` enqueued periodic collection scans *unlocked* while `ScannerService` released the per-provider lock unconditionally-if-held, so a scheduled scan could cross-release an API scan's lock (spurious 202 vs 409; no data corruption — single-reader channel). Fork-vs-Codex disagreement → **escalated to Fable** per protocol; Fable ratified Codex + caught that the naive per-source fix would starve the 2nd+ media source, prescribing the §C7 lock-once-per-provider batch (`Unlock: isLast`). Fixed in-PR + **Codex re-review of the fix commit = FIX CORRECT** (the #242 discipline). Codex Low (make the `QueueShowScanResult` switch total) also folded in. **Arc change**: none to ordering. **Reviewer#20 batch fully DONE.** Progress inside arc item 4 (#91b removal): the **F9 API gate is now CLOSED** — deep-scan (`?deep=`) + external-collections-scan (`POST /api/media-sources/{plex|jellyfin|emby}/{id}/scan-collections`) endpoints exist, so `Libraries.razor` is no longer API-blocked. Removal PR's remaining parity work is now just the **SPA deep-scan / collections buttons** (wiring the shipped `libraries.ts` clients into `LibrariesScreen`) + the mandatory cold adversarial pass (parity doc §5). #202, #204, #205, #206 all landed earlier today. **Removal-gate status**: #91b removal PR (arc item 4) is now gated only on (a) SPA deep/collections buttons, (b) #204's catch-all fallback (belongs in the removal PR), (c) the mandatory cold adversarial pass. No remaining *API* gate. **Recommended next** (parallel-safe, distinct surfaces): - **arc item 4 (#91b removal PR)** is now much closer to startable — the last API gate (F9) is closed. Remaining pre-work is SPA-side (deep/collections buttons) + the cold pass; the catch-all fallback lands in the removal PR itself. - **#254** was merged to main this session by a parallel session (`a1bd303c`/`6055bd36`/`13cd00c8`) — I merged it into #267 before landing; its `in-progress` may still be set, check with that session. - **#253 PR2+** (concurrency contract fan-out) and **#259** (content-aware child identity) remain open on the scheduling surface. - **#265** (If-Match 412-not-400, RFC 7232) — deferred #197 input, still open/unclaimed. **Cross-session notes**: main advanced twice during this session (#253 PR1 #263, then #254). #267 merged both in and regenerated artifacts each time (no drift). Two new files on main: `ErsatzTV.Tests/Services/{WorkerServiceTests,ScannerServiceCollectionLockTests,SchedulerServiceCollectionLockTests}.cs`. `in-progress` removed from #235; feature + slice worktrees cleaned; main pulled to `4aeabada`.
Author
Owner

Session 2026-07-11 (late evening) — #91b SPA pre-work: deep-scan + External Collections buttons (arc item 4)

Closed: none outright — PR #272 landed (bb7b1899, FF; branch deleted). Progress inside arc item 4 (#91b removal); #91 stays open for the removal PR proper. in-progress removed from #91.

Filed: #271 — external-collections scan has no status/progress REST surface (SPA collections buttons are optimistic + timeout-bounded).

Triage: #271 → backlog (Low, frontend/review; capability parity is met, only the live-progress affordance is approximate — not a gate). Also captured the fork's Low-1 (backend collections lock is family-global vs the SPA's per-source optimistic keying → benign 409 on a 2nd same-family source) as a note on #271.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged (#19 open, next slice = #91b/#197; #20/#21/#22/#23/#24 all delivered). PR #272 got a fork adversarial pass = MERGEABLE (verified the derivation-equivalence claim against GetAllMediaSourcesForApiHandler.ShouldIncludeLibrary, the useCollectionsScan leak/wedge-freedom, deep-threading, popstate contract; 2 Low, both non-blocking) + a live headless E2E = PASS against a real DB/handlers with a negative control (sync-disabled library excluded; all four button URLs + ?deep= observed at 202). Held Codex — pure-SPA change, fully covered by 674 component tests + fork + live E2E; consistent with the reserve-quota discipline.

Arc change: none to ordering. SPA Libraries.razor parity is DONE (blazor-route-parity.md §5 updated). The #91b removal PR (arc item 4) is now gated only on the mandatory cold adversarial pass (+ #204's catch-all fallback, which lands in the removal PR itself). No API gate, no SPA gate remains.

Recommended next (parallel-safe, distinct surfaces):

  • arc item 4 — the #91b removal PR itself is now startable. Remaining: cold adversarial pass + Fable go/no-go, then Steps 1–4 of blazor-route-parity.md §5 (rollback tag → surgical Blazor delete + catch-all fallback → auth cleanup → verify). ⚠️ It edits Startup.cs (736-795 MapWhen branch) and deletes ErsatzTV/Pages/** — coordinate so no session is mid-flight on a Blazor page.
  • #253 PR3/PR4 (Diff+Scalar / remaining aggregates) — another session owns #253 PR2 (#268 open in review); check that before touching the versioned-aggregate handlers.
  • #197 prep (arc item 5, HARD GATE) — #253's contract + #271 are inputs.

Cross-session note: main advanced to bb7b1899 this session (only PR #272; FF). No conflicts introduced.

## Session 2026-07-11 (late evening) — #91b SPA pre-work: deep-scan + External Collections buttons (arc item 4) **Closed**: none outright — **PR #272 landed** (`bb7b1899`, FF; branch deleted). Progress *inside* arc item 4 (#91b removal); #91 stays open for the removal PR proper. `in-progress` removed from #91. **Filed**: #271 — external-collections scan has no status/progress REST surface (SPA collections buttons are optimistic + timeout-bounded). **Triage**: #271 → **backlog** (Low, `frontend`/`review`; capability parity is met, only the live-progress affordance is approximate — not a gate). Also captured the fork's Low-1 (backend collections lock is family-global vs the SPA's per-source optimistic keying → benign 409 on a 2nd same-family source) as a note on #271. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot **unchanged** (#19 open, next slice = #91b/#197; #20/#21/#22/#23/#24 all delivered). PR #272 got a **fork adversarial pass = MERGEABLE** (verified the derivation-equivalence claim against `GetAllMediaSourcesForApiHandler.ShouldIncludeLibrary`, the `useCollectionsScan` leak/wedge-freedom, deep-threading, popstate contract; 2 Low, both non-blocking) **+ a live headless E2E = PASS** against a real DB/handlers with a negative control (sync-disabled library excluded; all four button URLs + `?deep=` observed at 202). **Held Codex** — pure-SPA change, fully covered by 674 component tests + fork + live E2E; consistent with the reserve-quota discipline. **Arc change**: none to ordering. **SPA `Libraries.razor` parity is DONE** (`blazor-route-parity.md §5` updated). The #91b removal PR (arc item 4) is now gated only on the **mandatory cold adversarial pass** (+ #204's catch-all fallback, which lands in the removal PR itself). No API gate, no SPA gate remains. **Recommended next** (parallel-safe, distinct surfaces): - **arc item 4 — the #91b removal PR itself is now startable.** Remaining: cold adversarial pass + Fable go/no-go, then Steps 1–4 of `blazor-route-parity.md §5` (rollback tag → surgical Blazor delete + catch-all fallback → auth cleanup → verify). ⚠️ It edits `Startup.cs` (736-795 MapWhen branch) and deletes `ErsatzTV/Pages/**` — coordinate so no session is mid-flight on a Blazor page. - **#253 PR3/PR4** (Diff+Scalar / remaining aggregates) — another session owns #253 PR2 (#268 open in review); check that before touching the versioned-aggregate handlers. - **#197** prep (arc item 5, HARD GATE) — #253's contract + #271 are inputs. Cross-session note: main advanced to `bb7b1899` this session (only PR #272; FF). No conflicts introduced.
Author
Owner

Session 2026-07-11 (evening) — #253 PR2: optimistic-concurrency RR + Reconcile fan-out

Closed: none outright — #253 PR2 landed (PR #268, merged to main 550d7ac9); #253 stays open as the PR3→PR4 arc. in-progress removed from #253.

Filed: none.

Triage: no new issues. Review findings were all fixed in-PR (see Reviews).

Reviews: PR #268 got a fork adversarial pass (SHIP) + an independent Codex pass that caught 2 Blockers the fork missed (replace-PUT ETag/items inconsistency → silent overwrite; Playlist Promise.all load race) + 2 Mediums — all fixed. The fix-commit re-review (the #242 discipline) came back MERGEABLE clean — done via a fork because the Codex re-review runtime was interrupted mid-turn (the companion left a stale "running" flag; the reliable signal is the rollout transcript advancing / task_complete, not the status list — worth remembering). Pending-audit snapshot unchanged (none pending pickup; #19 open next-slice #91b/#197, #20 remaining item #235).

Arc change: none to ordering. #253 PR2 done → remaining #253 = PR3 (Diff+Scalar: Collection, Playout ×2, MultiCollection, RerunCollection, incl. §9 H1 catch-rethrow) + PR4 (Phase-2 428 flip).

Notable: reconciled the PR branch with a twice-advanced main (#251/#252 earlier, then #235/#254/#91b during this session) — merged main in, verified Version++ survived in the shared ProgramSchedules handlers, regenerated OpenAPI artifacts (no drift). Live HTTP contract E2E run on all four aggregates via a locally mutex-bypassed build (reverted, uncommitted) so it ran alongside the user's instance on ports 8410/8411.

Also this session (separate, at user request): setting up husky + lint-staged autolint on feat/autolint-precommit (its own PR, NOT an arc item) — pre-commit eslint+typecheck on staged web files, plus (user picked all) pre-commit dotnet format on staged .cs, pre-push check:api+build+lint, and a commit-msg Co-Authored-By trailer check. Base hook verified (block/pass); the 3 extra guardrails in progress. Prompted by this session's CI miss: a lint error reached CI because I ran typecheck/build/vitest but not npm run lint after an SPA edit — the pre-commit/pre-push hooks close that gap.

Recommended next (parallel-safe, distinct surfaces):

  • #253 PR3 (Diff + Scalar fan-out) — copies the now-proven Block/PR2 recipe frozen in api-conventions §7a; the two Playout handlers need the §9 H1 catch(Exception)→422 rethrow so the guard's 412 survives. Coordinate with any live #235/#254 Playout work.
  • #235 async-contract remainder if still open (Libraries F9 gate).
  • Merge-to-main CI on 550d7ac9 re-confirming green (PR head was green; merge commit runs its own build+smoke).
## Session 2026-07-11 (evening) — #253 PR2: optimistic-concurrency RR + Reconcile fan-out **Closed**: none outright — **#253 PR2 landed** (PR #268, merged to main `550d7ac9`); #253 stays open as the PR3→PR4 arc. `in-progress` removed from #253. **Filed**: none. **Triage**: no new issues. Review findings were all fixed in-PR (see Reviews). **Reviews**: PR #268 got a **fork** adversarial pass (SHIP) + an **independent Codex** pass that caught **2 Blockers the fork missed** (replace-PUT ETag/items inconsistency → silent overwrite; Playlist `Promise.all` load race) + 2 Mediums — all fixed. The **fix-commit re-review** (the #242 discipline) came back **MERGEABLE clean** — done via a **fork** because the Codex re-review runtime was `interrupted` mid-turn (the companion left a stale "running" flag; the reliable signal is the rollout transcript advancing / `task_complete`, not the status list — worth remembering). Pending-audit snapshot unchanged (none pending pickup; #19 open next-slice #91b/#197, #20 remaining item #235). **Arc change**: none to ordering. #253 PR2 done → remaining #253 = PR3 (Diff+Scalar: Collection, Playout ×2, MultiCollection, RerunCollection, incl. §9 H1 catch-rethrow) + PR4 (Phase-2 428 flip). **Notable**: reconciled the PR branch with a twice-advanced main (#251/#252 earlier, then #235/#254/#91b during this session) — merged main in, verified `Version++` survived in the shared `ProgramSchedules` handlers, regenerated OpenAPI artifacts (no drift). Live HTTP contract E2E run on all four aggregates via a **locally mutex-bypassed build** (reverted, uncommitted) so it ran alongside the user's instance on ports 8410/8411. **Also this session (separate, at user request)**: setting up **husky + lint-staged autolint** on `feat/autolint-precommit` (its own PR, NOT an arc item) — pre-commit eslint+typecheck on staged web files, plus (user picked all) pre-commit `dotnet format` on staged `.cs`, pre-push `check:api`+build+lint, and a commit-msg `Co-Authored-By` trailer check. Base hook verified (block/pass); the 3 extra guardrails in progress. Prompted by this session's CI miss: a lint error reached CI because I ran typecheck/build/vitest but not `npm run lint` after an SPA edit — the pre-commit/pre-push hooks close that gap. **Recommended next** (parallel-safe, distinct surfaces): - **#253 PR3** (Diff + Scalar fan-out) — copies the now-proven Block/PR2 recipe frozen in api-conventions §7a; the two Playout handlers need the §9 H1 `catch(Exception)`→422 rethrow so the guard's 412 survives. Coordinate with any live #235/#254 Playout work. - **#235** async-contract remainder if still open (Libraries F9 gate). - Merge-to-main CI on `550d7ac9` re-confirming green (PR head was green; merge commit runs its own build+smoke).
Author
Owner

Session 2026-07-11 — #253 PR3: Diff + Scalar concurrency fan-out

Closed: none outright — #253 PR3 landed (PR #270, FF-merged to main 22440cc1). #253 stays open for PR4 (Phase-2 428 flip), the only remaining arc item on it now that PR1+PR2+PR3 cover all 10 replace-all PUTs. Claimed PR3 by comment (parallel to PR2); in-progress on #253 was already removed by the PR2 session.

Filed: #269 — same-root config writers under an active Version token. (Filed mid-session as an ETag-rotation deferral, then corrected + re-scoped after review found it was actually a 500 exposure — see Reviews.)

Triage:

  • #269 → priority pickup / #197-prep (api/review, was HIGH as a 500 but the realistic UPDATE writers are now fixed in PR #270; the residual is DELETE handlers + repo Add* writers + cross-editor ETag-rotation completeness). Not a #91b gate.

Reviews: no adversarial-reviewer audits picked up (none pending). PR #270 got a fork adversarial pass on the diff = MERGEABLE-WITH-NITS which caught a HIGH I'd missed: activating Version as an IsConcurrencyToken makes EF guard every root UPDATE, so non-participating plain-SaveChangesAsync root writers (playout settings/schedule-file/checkpoint, collection name) 500 on a concurrent bump. Fixed in-PR (shared SaveChangesForcingVersion force-write-on-conflict, 6 handlers + non-vacuous test) + a fix-commit re-review (the #242 discipline) = MERGEABLE. Codex pass was dispatched but never returned output (companion/quota) → relied on the two fork reviews (documented fallback). Also live HTTP E2E on two endpoints + full suites green (1389/493/678).

Arc change: none to ordering. PR2 (#268) + PR3 both landed today → the arc's forward edge is #91b prep (item 4) and #197 (item 5, HARD GATE); PR4 (428 flip) is the #253 tail. Cross-session: main advanced twice under me — PR2 #268 and the #91b LibrariesScreen deep-scan/external-collections wiring (bb7b1899, which also filed #271 collections scan-status endpoint). Merged main in (only decisions.md conflicted, both append-only; regenerated OpenAPI artifacts per the lore).

Recommended next (parallel-safe):

  • #253 PR4 (Phase-2: flip missing If-Match → 428 once editors soak) — small, but wait for a release soak per the design.
  • #91b removal PR (item 4) — #202 done; the #91b scan-button wiring landed (bb7b1899); check whether #235 F9 is now fully satisfied. The mandatory cold adversarial pass still applies.
  • #197 (item 5) — inputs now include the full #253 contract (PR1–PR3) + #265 + #269.
  • ⚠️ #197/#269/PR4 note: PR2's ProgramSchedule root likely carries the same token-activation 500 exposure (UpdateProgramScheduleHandler is a separate root-scalar writer from the bumping ReplaceProgramScheduleItems) — audit it with the SaveChangesForcingVersion lens.
## Session 2026-07-11 — #253 PR3: Diff + Scalar concurrency fan-out **Closed**: none outright — **#253 PR3 landed** (PR #270, FF-merged to main `22440cc1`). #253 stays **open** for PR4 (Phase-2 428 flip), the only remaining arc item on it now that PR1+PR2+PR3 cover all 10 replace-all PUTs. Claimed PR3 by comment (parallel to PR2); `in-progress` on #253 was already removed by the PR2 session. **Filed**: **#269** — same-root config writers under an active `Version` token. (Filed mid-session as an ETag-rotation deferral, then **corrected + re-scoped** after review found it was actually a 500 exposure — see Reviews.) **Triage**: - **#269** → **priority pickup / #197-prep** (`api`/`review`, was HIGH as a 500 but the realistic UPDATE writers are now fixed in PR #270; the residual is DELETE handlers + repo `Add*` writers + cross-editor ETag-rotation completeness). Not a #91b gate. **Reviews**: no adversarial-reviewer audits picked up (none pending). PR #270 got a **fork adversarial pass** on the diff = MERGEABLE-WITH-NITS which caught a **HIGH I'd missed**: activating `Version` as an `IsConcurrencyToken` makes EF guard *every* root UPDATE, so non-participating plain-`SaveChangesAsync` root writers (playout settings/schedule-file/checkpoint, collection name) 500 on a concurrent bump. **Fixed in-PR** (shared `SaveChangesForcingVersion` force-write-on-conflict, 6 handlers + non-vacuous test) + a **fix-commit re-review** (the #242 discipline) = **MERGEABLE**. Codex pass was dispatched but never returned output (companion/quota) → relied on the two fork reviews (documented fallback). Also live HTTP E2E on two endpoints + full suites green (1389/493/678). **Arc change**: none to ordering. PR2 (#268) + PR3 both landed today → the arc's forward edge is **#91b prep (item 4)** and **#197 (item 5, HARD GATE)**; PR4 (428 flip) is the #253 tail. **Cross-session**: main advanced twice under me — PR2 `#268` and the #91b LibrariesScreen deep-scan/external-collections wiring (`bb7b1899`, which also filed **#271** collections scan-status endpoint). Merged main in (only `decisions.md` conflicted, both append-only; regenerated OpenAPI artifacts per the lore). **Recommended next** (parallel-safe): - **#253 PR4** (Phase-2: flip missing If-Match → 428 once editors soak) — small, but wait for a release soak per the design. - **#91b removal PR (item 4)** — #202 done; the #91b scan-button wiring landed (`bb7b1899`); check whether **#235 F9** is now fully satisfied. The mandatory cold adversarial pass still applies. - **#197 (item 5)** — inputs now include the full #253 contract (PR1–PR3) + #265 + #269. - ⚠️ **#197/#269/PR4 note**: PR2's `ProgramSchedule` root likely carries the same token-activation 500 exposure (`UpdateProgramScheduleHandler` is a separate root-scalar writer from the bumping `ReplaceProgramScheduleItems`) — audit it with the `SaveChangesForcingVersion` lens.
Author
Owner

Session 2026-07-11 (night) — #91b Blazor removal PR (arc item 4 DONE)

Closed: #91 (PR #274, merged FF to main 78262ea4; CI green, branch deleted). The milestone "Blazor removal (#91 phase b)" is complete (0 open). Legacy Blazor Server/MudBlazor UI deleted — SPA is the only UI; root / + every legacy route reach /app via LegacyUiRedirects + a new Startup catch-all.

Filed: none.

Triage: no new issues.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged (#19 open, next slice = #197; #20/#21/#22/#23/#24 all delivered). The removal cleared its mandatory gates: a cold zero-context fork review + an independent Codex pass (both MERGEABLE-WITH-NITS, no disagreement; the fork caught an orphaned NavigationManagerExtensions.cs, Codex caught 2 dead-residue nits — all 3 fixed as follow-up commits), a live-E2E redirect sweep (20/20: legacy exact/pattern/media-sources 302s correct, catch-all→/app, /api 404 & /artwork 400 never redirected, no 500s), and a Fable GO go/no-go.

Arc change: none to ordering. Arc item 4 (#91b removal) is DONE. The arc's forward edge is now item 5 — #197 (cold API contract + security review, HARD GATE before remote exposure). Item 6 (go-live) follows #197.

  • Rollback tag blazor-final = bb310f6b (pre-deletion tip; not a v* tag, no release build). Restore = checkout tag or revert the merge.
  • Removal folded in: #204 catch-all fallback, the 14 /media/sources/*→/app/libraries/* redirects (+ lifted the #204 prefix ban), #206 auth posture (Blazor-attached OIDC removed; service wiring kept for #197), pruned 9 dead packages + the #25 razor-Sonar NoWarn.

Recommended next (parallel-safe, distinct surfaces):

  • #197 (arc item 5, HARD GATE) — cold API contract + security review. Inputs now include the completed Blazor removal (auth posture is now "SPA/API open until #197" with no Blazor OIDC page-gate), #253's replace-all concurrency contract (PR2 #268 landed; PR3/PR4 in flight in another session), #265 (If-Match 412), #271 (collections scan-status), #254 tail, backlog nits.
  • #253 PR3/PR4 fan-out continues (another session owns feat/253-pr3-diff-scalar).
  • Cross-session note: main advanced to 78262ea4; merged origin/main (incl. #268 #253-PR2 + #273 husky) into the removal branch before landing — clean auto-merge, no conflicts, full suite re-verified green (ErsatzTV.Tests 1404). Husky pre-commit/pre-push hooks are now live on main (#273): staged dotnet format + web check:api/lint/typecheck/build on push.

in-progress removed from #91; worktree retained locally (branch merged+deleted on origin).

## Session 2026-07-11 (night) — #91b Blazor removal PR (arc item 4 DONE) **Closed**: #91 (PR #274, merged FF to main `78262ea4`; CI green, branch deleted). The **milestone "Blazor removal (#91 phase b)" is complete** (0 open). Legacy Blazor Server/MudBlazor UI deleted — SPA is the only UI; root `/` + every legacy route reach `/app` via `LegacyUiRedirects` + a new Startup catch-all. **Filed**: none. **Triage**: no new issues. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot **unchanged** (#19 open, next slice = #197; #20/#21/#22/#23/#24 all delivered). The removal cleared its mandatory gates: a cold zero-context fork review **+ an independent Codex pass** (both MERGEABLE-WITH-NITS, **no disagreement**; the fork caught an orphaned `NavigationManagerExtensions.cs`, Codex caught 2 dead-residue nits — all 3 fixed as follow-up commits), a **live-E2E redirect sweep** (20/20: legacy exact/pattern/media-sources 302s correct, catch-all→/app, `/api` 404 & `/artwork` 400 never redirected, no 500s), and a **Fable GO** go/no-go. **Arc change**: none to ordering. **Arc item 4 (#91b removal) is DONE.** The arc's forward edge is now **item 5 — #197** (cold API contract + security review, HARD GATE before remote exposure). Item 6 (go-live) follows #197. - Rollback tag **`blazor-final` = `bb310f6b`** (pre-deletion tip; not a `v*` tag, no release build). Restore = checkout tag or revert the merge. - Removal folded in: #204 catch-all fallback, the 14 `/media/sources/*`→`/app/libraries/*` redirects (+ lifted the #204 prefix ban), #206 auth posture (Blazor-attached OIDC removed; service wiring kept for #197), pruned 9 dead packages + the #25 razor-Sonar NoWarn. **Recommended next** (parallel-safe, distinct surfaces): - **#197** (arc item 5, HARD GATE) — cold API contract + security review. Inputs now include the completed Blazor removal (auth posture is now "SPA/API open until #197" with no Blazor OIDC page-gate), #253's replace-all concurrency contract (PR2 #268 landed; **PR3/PR4 in flight** in another session), #265 (If-Match 412), #271 (collections scan-status), #254 tail, backlog nits. - **#253 PR3/PR4** fan-out continues (another session owns `feat/253-pr3-diff-scalar`). - **Cross-session note**: main advanced to `78262ea4`; merged origin/main (incl. #268 #253-PR2 + #273 husky) into the removal branch before landing — clean auto-merge, no conflicts, full suite re-verified green (ErsatzTV.Tests 1404). Husky pre-commit/pre-push hooks are now live on `main` (#273): staged `dotnet format` + web `check:api`/lint/typecheck/build on push. `in-progress` removed from #91; worktree retained locally (branch merged+deleted on origin).
Author
Owner

Session 2026-07-11 (night) — #259 content-aware stable child identity (schedule items)

Closed: #259 (PR #276, merged to main 5e5a5931; branch deleted; main merge-commit CI running). Priority-pickup on the scheduling surface, parallel to the gates.

Filed: none.

Triage: no new issues of my own. The cold review's one Low (handler trusts the controller for 0→null normalization — defense-in-depth) was left as-is: unreachable (the only production constructor is ToReplaceCommand, which normalizes) and fail-safe (a stray 0/negative id → 422, never corruption). Not worth an issue; noted in the close comment.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged (#19 open, next slice = #91b/#197; #20–#24 delivered). PR #276 got a cold fork adversarial pass = MERGEABLE-WITH-NITS (attacked every payload shape by hand: positional fallback byte-identical to main, Version++ in both modes, guards persist-nothing, 412-before-422, no 500 on unknown id, no unique-index collision on reordered Index, SPA round-trip complete/no id fabrication) — its one Medium (id-mode subtype-change untested at handler level) fixed in-PR with a new test (id-matched subtype-change delete+insert alongside a same-type sibling whose state survives). Held Codex: this reuses the existing §7a guard machinery unchanged (no new lock/ownership primitive — the class where Codex earned its keep in #235/#250/#253), and is covered by the reconcile matrix + a proven-non-vacuous negative control + the integration round-trip + fork + a live headless E2E = PASS 8/8 (reorder→200 carrying ids, two consecutive saves no-422 proving response re-seed, new item id:null, bogus id→422). Consistent with the reserve-quota discipline (cf. #272).

Arc change: none to ordering. #259 was a parallel priority-pickup, not an arc gate.

#253 PR4 clarification (relevant to arc item 5 / #197 inputs): while scoping tonight's pickup I established that #253 PR4 (the Phase-2 If-Match→428 flip) is not blocked by a soak timer — it's blocked because Phase-1 has never reached prod. Prod = v26.5.0 (2026-07-07); the latest tag v26.6.0 (2026-07-08) predates all of #253 (git merge-base confirms PR1–PR3 are NOT in v26.6.0). So 'wait for one release to soak' means 'ship Phase-1 to prod and observe it', and doing PR4 now would ship 428 having never run Phase-1 in prod. Also worth noting for #197: the two-phase force-write→428 dance is arguably over-engineered here — we own the only client (the lockstep-shipped SPA, already echoing If-Match on every editor, E2E-verified), so the soak's only residual value is catching a bug in our own SPA echo. Reasonable to collapse the phasing (flip to 428 in the same release Phase-1 first ships, once verified) rather than mandate a separate later release. Flagged as a #197 contract-decision input, not acted on.

Recommended next (parallel-safe, distinct surfaces):

  • arc item 4 (#91b removal) is in flight in another session (PR #274) — do not touch Startup.cs/Pages/**.
  • #269 (same-root config sibling writers / DELETE-handler token exposure) — priority-pickup, fold into #197 per the issue's own recommendation (don't ship standalone).
  • #197 (arc item 5, HARD GATE) — start once #91b lands and freezes the auth posture; inputs now include the full #253 contract, #265, #269, #271, and the #253-PR4-phasing decision above.
  • #253 PR4 — do NOT ship until Phase-1 (PR1–PR3) actually reaches prod (see clarification above).

Cross-session: main advanced to 5e5a5931 (my merge). #259 used a backend-first feature worktree + a parallel SPA sub-worktree (merged back), both cleaned up.

## Session 2026-07-11 (night) — #259 content-aware stable child identity (schedule items) **Closed**: #259 (PR #276, merged to main `5e5a5931`; branch deleted; main merge-commit CI running). Priority-pickup on the scheduling surface, parallel to the gates. **Filed**: none. **Triage**: no new issues of my own. The cold review's one Low (handler trusts the controller for `0→null` normalization — defense-in-depth) was left as-is: unreachable (the only production constructor is `ToReplaceCommand`, which normalizes) and fail-safe (a stray `0`/negative id → 422, never corruption). Not worth an issue; noted in the close comment. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot **unchanged** (#19 open, next slice = #91b/#197; #20–#24 delivered). PR #276 got a **cold fork adversarial pass = MERGEABLE-WITH-NITS** (attacked every payload shape by hand: positional fallback byte-identical to main, `Version++` in both modes, guards persist-nothing, 412-before-422, no 500 on unknown id, no unique-index collision on reordered `Index`, SPA round-trip complete/no id fabrication) — its one Medium (id-mode subtype-change untested at handler level) **fixed in-PR** with a new test (id-matched subtype-change delete+insert alongside a same-type sibling whose state survives). **Held Codex**: this reuses the existing §7a guard machinery unchanged (no new lock/ownership primitive — the class where Codex earned its keep in #235/#250/#253), and is covered by the reconcile matrix + a proven-non-vacuous negative control + the integration round-trip + fork + a live headless **E2E = PASS 8/8** (reorder→200 carrying ids, two consecutive saves no-422 proving response re-seed, new item id:null, bogus id→422). Consistent with the reserve-quota discipline (cf. #272). **Arc change**: none to ordering. #259 was a parallel priority-pickup, not an arc gate. **#253 PR4 clarification (relevant to arc item 5 / #197 inputs)**: while scoping tonight's pickup I established that #253 PR4 (the Phase-2 `If-Match`→428 flip) is **not blocked by a soak *timer* — it's blocked because Phase-1 has never reached prod**. Prod = v26.5.0 (2026-07-07); the latest tag v26.6.0 (2026-07-08) predates all of #253 (`git merge-base` confirms PR1–PR3 are NOT in v26.6.0). So 'wait for one release to soak' means 'ship Phase-1 to prod and observe it', and doing PR4 now would ship 428 having never run Phase-1 in prod. Also worth noting for #197: the two-phase force-write→428 dance is arguably over-engineered here — we own the only client (the lockstep-shipped SPA, already echoing `If-Match` on every editor, E2E-verified), so the soak's only residual value is catching a bug in *our own* SPA echo. Reasonable to **collapse the phasing** (flip to 428 in the same release Phase-1 first ships, once verified) rather than mandate a separate later release. Flagged as a #197 contract-decision input, not acted on. **Recommended next** (parallel-safe, distinct surfaces): - **arc item 4 (#91b removal)** is in flight in another session (PR #274) — do not touch `Startup.cs`/`Pages/**`. - **#269** (same-root config sibling writers / DELETE-handler token exposure) — priority-pickup, **fold into #197** per the issue's own recommendation (don't ship standalone). - **#197** (arc item 5, HARD GATE) — start once #91b lands and freezes the auth posture; inputs now include the full #253 contract, #265, #269, #271, and the #253-PR4-phasing decision above. - **#253 PR4** — do NOT ship until Phase-1 (PR1–PR3) actually reaches prod (see clarification above). **Cross-session**: main advanced to `5e5a5931` (my merge). #259 used a backend-first feature worktree + a parallel SPA sub-worktree (merged back), both cleaned up.
Author
Owner

Session 2026-07-11 (night) — #197 cold API review + Phase-0 hardening

Closed: #66 (superseded by #283). PR #279 merged FF to main (c55a7fda, CI green, diff-reviewed MERGEABLE) — Phase-0 safe hardening (constant-time key compare, playout paging clamps, nosniff/X-Frame/Referrer headers). #197 stays open as the umbrella gate (arc item 5), now decomposed into #280–#289.

Filed: #280 S1 fail-open writes (BLOCKER) · #281 S2 non-/api mutation bypass, SortController (HIGH) · #282 S3/S5 unauth sensitive-read disclosure (HIGH) · #283 S4/S9 content-type stored-XSS (HIGH, supersedes #66) · #284 S6 CORS AllowAll (HIGH/MED) · #285 S7/S10 residue (MED) · #286 C1 no versioning (BLOCKER) · #287 C2/4/5/6 OpenAPI security/error/operationId/DayOfWeek (BLOCKER/HIGH) · #288 C3/C7 DTO nullability + raw VMs (HIGH) · #289 MCP hardening (HIGH, PR#76).

Triage: all #280–#289 are gate items under #197 (arc item 5 — HARD GATE before remote exposure). Owner set exposure = remote via reverse proxy + Authelia, so S1–S6 + C1/C2 are hard blockers. Phase-1 (auth + XSS) blocks exposure; Phase-2 (contract) blocks the freeze; #289 blocks MCP go-live. Bundling map posted on #197.

Reviews: adversarial-reviewer#19 slice 4/4 (the #197 review) DELIVERED — all four #19 slices now complete → recommend closing #19. #24 (App.tsx modularization plan review) still pending pickup. Method: 4 cold independent passes (3 Claude + 1 Codex — Codex caught 2 unique HIGHs) + a Fable [PLAN-MODE] posture design + a scoped fork review of the #279 diff before merge. Codex quota was available and used.

Arc change: none to ordering. Arc item 5 (#197) is now: cold review done + phased roadmap + Phase-0 landed; the gate is decomposed into #280–#289. Forward edge stays #197 (via its sub-issues) → go-live (item 6).

Recommended next (parallel-safe by disjoint surface — see the #197 bundling map):

  • Bundle A (auth posture: #280/#281/#282/#284/#285) — ONE worktree/PR, shared ApiKeyAuthorizationFilter+Startup.cs+SPA client.ts+ApiControllerSecurityTests surface; do NOT parallelize internally. Write the decisions.md security-posture entry here (incl. the PR#279 headers baseline).
  • Bundle B (#283 content-type XSS) and #289 (MCP, separate repo) — mutually parallel-safe with A → up to 3 concurrent sessions.
  • Then Bundle C (#287/#288 contract cleanup, one regen), then #286 (versioning) LAST.
  • Owner decisions still open (flagged on #197): read/write key split, channel {id} vs {channelNumber} re-key, /api/*→/api/v1 sunset window, OIDC-registration removal, Api:RequireKeyForReads default.
  • Note: the user's main checkout should git pull (main advanced to c55a7fda).

in-progress removed from #197.

## Session 2026-07-11 (night) — #197 cold API review + Phase-0 hardening **Closed**: #66 (superseded by #283). **PR #279 merged FF to main** (`c55a7fda`, CI green, diff-reviewed MERGEABLE) — Phase-0 safe hardening (constant-time key compare, playout paging clamps, nosniff/X-Frame/Referrer headers). **#197 stays open** as the umbrella gate (arc item 5), now decomposed into #280–#289. **Filed**: #280 S1 fail-open writes (BLOCKER) · #281 S2 non-/api mutation bypass, SortController (HIGH) · #282 S3/S5 unauth sensitive-read disclosure (HIGH) · #283 S4/S9 content-type stored-XSS (HIGH, supersedes #66) · #284 S6 CORS AllowAll (HIGH/MED) · #285 S7/S10 residue (MED) · #286 C1 no versioning (BLOCKER) · #287 C2/4/5/6 OpenAPI security/error/operationId/DayOfWeek (BLOCKER/HIGH) · #288 C3/C7 DTO nullability + raw VMs (HIGH) · #289 MCP hardening (HIGH, PR#76). **Triage**: all #280–#289 are gate items under **#197 (arc item 5 — HARD GATE before remote exposure)**. Owner set exposure = **remote via reverse proxy + Authelia**, so S1–S6 + C1/C2 are hard blockers. Phase-1 (auth + XSS) blocks exposure; Phase-2 (contract) blocks the freeze; #289 blocks MCP go-live. Bundling map posted on #197. **Reviews**: **adversarial-reviewer#19 slice 4/4 (the #197 review) DELIVERED** — all four #19 slices now complete → recommend closing #19. #24 (App.tsx modularization plan review) still pending pickup. Method: 4 cold independent passes (3 Claude + 1 Codex — Codex caught 2 unique HIGHs) + a Fable [PLAN-MODE] posture design + a scoped fork review of the #279 diff before merge. Codex quota was available and used. **Arc change**: none to ordering. Arc item 5 (#197) is now: cold review done + phased roadmap + Phase-0 landed; the gate is decomposed into #280–#289. Forward edge stays #197 (via its sub-issues) → go-live (item 6). **Recommended next** (parallel-safe by disjoint surface — see the #197 bundling map): - **Bundle A** (auth posture: #280/#281/#282/#284/#285) — ONE worktree/PR, shared `ApiKeyAuthorizationFilter`+`Startup.cs`+SPA `client.ts`+`ApiControllerSecurityTests` surface; **do NOT parallelize internally**. Write the decisions.md security-posture entry here (incl. the PR#279 headers baseline). - **Bundle B** (#283 content-type XSS) and **#289** (MCP, separate repo) — mutually parallel-safe with A → up to 3 concurrent sessions. - Then **Bundle C** (#287/#288 contract cleanup, one regen), then **#286** (versioning) LAST. - **Owner decisions still open** (flagged on #197): read/write key split, channel `{id}` vs `{channelNumber}` re-key, `/api/*`→`/api/v1` sunset window, OIDC-registration removal, `Api:RequireKeyForReads` default. - Note: the user's main checkout should `git pull` (main advanced to `c55a7fda`). `in-progress` removed from #197.
Author
Owner

Session 2026-07-12 — #197 Bundle A (auth posture)

Closed: #280 (S1 fail-closed writes, BLOCKER) · #281 (S2 delete SortController/AccountController) · #282 (S3/S5 sensitive-read tier) · #284 (S6 CORS) · #285 (S7/S10 residue) — all via PR #292 (FF-merged to main ee6be81c, main build+smoke running). Arc item 5's Phase-1 auth cluster.

Owner decisions (the two the Fable design pass left open): single API key (no read/write split); Api:RequireKeyForReads defaults true (whole /api gated). Confirmed live that this does NOT touch Jellyfin — /iptv/* + /artwork/* are outside the filter scope.

Filed: #293 — page/cap search/all-items (DoS hardening, deferred from #285 because it feeds the SPA add-all flow; unauth exposure already closed by read-gating).

Triage:

  • #293 → backlog (api/security/low). Not a gate; Phase-3 / #197 follow-up.

Reviews: no adversarial-reviewer audits picked up. PR #292 got a cold zero-context fork review (MERGEABLE-WITH-NITS, no BLOCKER/HIGH — verified no fail-open path, /apixyz non-match, empty-key reject, no CORS residue, clean deletions, no SPA raw-fetch bypass) + a fix-commit re-review (the #242 discipline — also MERGEABLE-WITH-NITS). All findings resolved in-PR: M1 (ApiKeyProviderTests, never-empty invariant), L2 (atomic 0600 create), L3 (rethrow-not-clobber on unreadable key + delete-race tolerance, now tested). Codex not used this session (reserved). Live headless E2E = full auth flow PASS incl. /iptv open with no key.

Arc change: none to ordering. Arc item 5 (#197) Phase-1 auth = Bundle A DONE. Forward edge of the Phase-1 remote-exposure gate is now Bundle B (#283 content-type XSS) + #289 (MCP) — both open, disjoint surfaces, parallel-safe. Phase-2 freeze = Bundle C (#287/#288) then #286 (versioning, LAST).

Recommended next (parallel-safe by disjoint surface):

  • Bundle B — #283 (content-type XSS; artwork upload/serve/ImageCache) — completes the Phase-1 remote-exposure gate alongside Bundle A.
  • #289 (MCP hardening, separate issue58-mcp/PR #76 repo) — independent.
  • Then Bundle C (#287/#288) → #286 last. Coordination note: #286 (versioning) rewrites every route + regenerates all artifacts → must be sequenced AFTER Bundle C and any other in-flight /api work.

in-progress removed from #280/#281/#282/#284/#285.

## Session 2026-07-12 — #197 Bundle A (auth posture) **Closed**: #280 (S1 fail-closed writes, BLOCKER) · #281 (S2 delete SortController/AccountController) · #282 (S3/S5 sensitive-read tier) · #284 (S6 CORS) · #285 (S7/S10 residue) — all via **PR #292** (FF-merged to main `ee6be81c`, main build+smoke running). Arc item 5's Phase-1 auth cluster. **Owner decisions** (the two the Fable design pass left open): **single API key** (no read/write split); **`Api:RequireKeyForReads` defaults true** (whole `/api` gated). Confirmed live that this does NOT touch Jellyfin — `/iptv/*` + `/artwork/*` are outside the filter scope. **Filed**: **#293** — page/cap `search/all-items` (DoS hardening, deferred from #285 because it feeds the SPA add-all flow; unauth exposure already closed by read-gating). **Triage**: - **#293** → **backlog** (`api`/`security`/low). Not a gate; Phase-3 / #197 follow-up. **Reviews**: no adversarial-reviewer audits picked up. PR #292 got a **cold zero-context fork review** (MERGEABLE-WITH-NITS, no BLOCKER/HIGH — verified no fail-open path, `/apixyz` non-match, empty-key reject, no CORS residue, clean deletions, no SPA raw-fetch bypass) **+ a fix-commit re-review** (the #242 discipline — also MERGEABLE-WITH-NITS). All findings resolved in-PR: M1 (`ApiKeyProviderTests`, never-empty invariant), L2 (atomic 0600 create), L3 (rethrow-not-clobber on unreadable key + delete-race tolerance, now tested). Codex not used this session (reserved). Live headless E2E = full auth flow PASS incl. `/iptv` open with no key. **Arc change**: none to ordering. **Arc item 5 (#197) Phase-1 auth = Bundle A DONE.** Forward edge of the Phase-1 remote-exposure gate is now **Bundle B (#283 content-type XSS)** + **#289 (MCP)** — both open, disjoint surfaces, parallel-safe. Phase-2 freeze = Bundle C (#287/#288) then #286 (versioning, LAST). **Recommended next** (parallel-safe by disjoint surface): - **Bundle B — #283** (content-type XSS; artwork upload/serve/ImageCache) — completes the Phase-1 remote-exposure gate alongside Bundle A. - **#289** (MCP hardening, separate `issue58-mcp`/PR #76 repo) — independent. - Then **Bundle C (#287/#288)** → **#286** last. **Coordination note**: #286 (versioning) rewrites every route + regenerates all artifacts → must be sequenced AFTER Bundle C and any other in-flight `/api` work. `in-progress` removed from #280/#281/#282/#284/#285.
Author
Owner

Session 2026-07-12 — #283 artwork content-type stored-XSS (Bundle B, #197 gate)

Closed: #283 (PR #291, merged FF to main f8fd9084; branch deleted). Bundle B of the #197 Phase-1 remote-exposure gate is done — artwork upload sniffs true format from bytes (SkiaSharp), serve-side ?contentType= reflection removed structurally + clamped to the image allow-list, DTO Sanitized() defense-in-depth, Kestrel MaxRequestBodySize for S9. Also merged PR #294 (CI: MySQL migration-job resilience).

Filed: none.

Triage: no new issues.

Reviews: one cold fork adversarial pass = MERGEABLE (verified sniff correctness, every serve sink, Sanitized() coverage, no M3U/SPA regression, non-vacuous tests); its one actionable LOW (serve path could still render a legacy-poisoned/polyglot file) folded in as the image-allow-list clamp. Held Codex — the fix is self-contained (no new lock/ownership primitive), covered by 1421+14+20 tests, a 6/6 live E2E, and the fork pass. No adversarial-reviewer audits picked up/filed (snapshot unchanged: #19 all slices delivered, #24 still pending pickup).

CI-infra finding (diagnosed + fixed this session): the EF migration integrity job's MySQL apply-to-fresh-DB step was flaking — blocked #291 twice. Root cause = concurrent migration jobs on one runner host (each spins its own mysql:8.4 service → starvation), NOT a model/code problem (has-pending-model-changes passed; identical tree passed on a quieter host; both runners passed and failed). Fixed in PR #294 (DefaultCommandTimeout=300 + a 3× resume-retry on the apply; documented in docs/ci-cd.md). De-flakes the job for all in-flight PRs — worth knowing for any session that sees a lone MySQL-apply red.

Arc change: none to ordering. #283 is one of the #197 Phase-1 sub-issues (arc item 5); Bundle A (#280/281/282/284/285) landed in parallel via PR #292 (another session). Remaining #197: Bundle C (#287/#288), #286 versioning, #289 MCP, #283-adjacent nits.

Recommended next (parallel-safe): Bundle C (#287/#288 contract cleanup, one OpenAPI regen) then #286 (/api/v1 versioning, LAST); #289 (MCP hardening, separate repo/PR #76). Confirm current in-progress labels on the #197 sub-issues before claiming — multiple sessions were active tonight.

## Session 2026-07-12 — #283 artwork content-type stored-XSS (Bundle B, #197 gate) **Closed**: #283 (PR #291, merged FF to main `f8fd9084`; branch deleted). Bundle B of the #197 Phase-1 remote-exposure gate is done — artwork upload sniffs true format from bytes (SkiaSharp), serve-side `?contentType=` reflection removed structurally + clamped to the image allow-list, DTO `Sanitized()` defense-in-depth, Kestrel `MaxRequestBodySize` for S9. Also **merged PR #294** (CI: MySQL migration-job resilience). **Filed**: none. **Triage**: no new issues. **Reviews**: one **cold fork adversarial pass = MERGEABLE** (verified sniff correctness, every serve sink, `Sanitized()` coverage, no M3U/SPA regression, non-vacuous tests); its one actionable LOW (serve path could still render a legacy-poisoned/polyglot file) **folded in** as the image-allow-list clamp. **Held Codex** — the fix is self-contained (no new lock/ownership primitive), covered by 1421+14+20 tests, a 6/6 live E2E, and the fork pass. No adversarial-reviewer audits picked up/filed (snapshot unchanged: #19 all slices delivered, #24 still pending pickup). **CI-infra finding (diagnosed + fixed this session)**: the `EF migration integrity` job's MySQL apply-to-fresh-DB step was flaking — blocked #291 twice. **Root cause = concurrent migration jobs on one runner host** (each spins its own `mysql:8.4` service → starvation), NOT a model/code problem (`has-pending-model-changes` passed; identical tree passed on a quieter host; both runners passed *and* failed). Fixed in **PR #294** (`DefaultCommandTimeout=300` + a 3× resume-retry on the apply; documented in `docs/ci-cd.md`). De-flakes the job for **all** in-flight PRs — worth knowing for any session that sees a lone MySQL-apply red. **Arc change**: none to ordering. #283 is one of the #197 Phase-1 sub-issues (arc item 5); Bundle A (#280/281/282/284/285) landed in parallel via **PR #292** (another session). Remaining #197: Bundle C (#287/#288), #286 versioning, #289 MCP, #283-adjacent nits. **Recommended next** (parallel-safe): **Bundle C** (#287/#288 contract cleanup, one OpenAPI regen) then **#286** (`/api/v1` versioning, LAST); **#289** (MCP hardening, separate repo/PR #76). Confirm current `in-progress` labels on the #197 sub-issues before claiming — multiple sessions were active tonight.
Author
Owner

Session 2026-07-12 — #289 MCP server hardening

Closed: #289 (landed on issue58-mcp / PR #76, FF d1c04030..65d88b51; not main). The #197 MCP-consumer security findings are fixed on the MCP branch.

Filed: none.

Triage: no new issues. Review findings on this PR were all fixed in-PR; the one LOW theoretical (adjacent-param path-segment concat) is unreachable with the current catalog → documented in a code comment, not filed.

Reviews: no adversarial-reviewer audits picked up — snapshot unchanged (#19 open [its #197 slice incl. this MCP finding was delivered last session], #24 pending pickup, #20 remaining = #235). This PR got a full independent loop: 2 cold adversarial forks + 1 Codex pass on the initial diff (genuinely complementary — the fork caught a HIGH body-read-timeout regression Codex missed; Codex caught an env-overflow trio + a .. path collapse the fork missed) → all folded in → 1 cold re-review of the fix commit = MERGEABLE. Codex quota was available and used.

Arc change: none. #289 is a Phase-1 sub-item of arc item 5 (#197) — the MCP go-live gate (#197 roadmap listed it as PR#76 hardening) — worked parallel to Bundle B, disjoint surface.

Scope decision (owner-directed this session): the MCP's final design is not read-only (#58 targets read + write/operate). So the verb-guard was implemented as a runtime read-only posture (ERSATZTV_ALLOW_WRITES, default false), not a hardcoded GET-clamp — write tools slot in behind the opt-in later. MCP expansion (write/workflow tools) is deferred until after the contract freeze (Bundle C #287/#288 → #286 versioning): the tool catalog binds to operationIds/schemas that #287 shows are still unstable (93/243 ops lack operationIds), so expanding now would churn. Don't pick up MCP expansion before the freeze.

Recommended next (parallel-safe by disjoint surface):

  • Bundle B — #283 (content-type XSS): user indicated it's landing this session in another session — do not double-pick.
  • Bundle C (#287/#288) — OpenAPI contract freeze; sequence AFTER Bundle B lands (both regenerate v1.json/endpoint-index), then #286 (versioning) LAST (rewrites every route + all generated artifacts → after all other in-flight /api work).
  • PR #76 (MCP) is now security-clean but not on the critical arc: before it can merge to main it needs a main-merge (it's ~5 days behind, pre-Bundle-A/B/#91b) + generated-artifact regen + full-solution CI; go-live also gated on #197. Leave it parked until the arc reaches go-live, or land the main-merge opportunistically.

in-progress removed from #289.

## Session 2026-07-12 — #289 MCP server hardening **Closed**: #289 (landed on `issue58-mcp` / **PR #76**, FF `d1c04030..65d88b51`; not `main`). The #197 MCP-consumer security findings are fixed on the MCP branch. **Filed**: none. **Triage**: no new issues. Review findings on this PR were all fixed in-PR; the one LOW theoretical (adjacent-param path-segment concat) is unreachable with the current catalog → documented in a code comment, not filed. **Reviews**: no adversarial-reviewer audits picked up — snapshot **unchanged** (#19 open [its #197 slice incl. this MCP finding was delivered last session], #24 pending pickup, #20 remaining = #235). This PR got a full independent loop: **2 cold adversarial forks + 1 Codex pass** on the initial diff (genuinely complementary — the fork caught a HIGH body-read-timeout regression Codex missed; Codex caught an env-overflow trio + a `..` path collapse the fork missed) → all folded in → **1 cold re-review of the fix commit = MERGEABLE**. Codex quota was available and used. **Arc change**: none. #289 is a Phase-1 sub-item of **arc item 5 (#197)** — the MCP go-live gate (#197 roadmap listed it as PR#76 hardening) — worked parallel to Bundle B, disjoint surface. **Scope decision (owner-directed this session)**: the MCP's *final* design is **not** read-only (#58 targets read + write/operate). So the verb-guard was implemented as a **runtime read-only posture** (`ERSATZTV_ALLOW_WRITES`, default false), not a hardcoded GET-clamp — write tools slot in behind the opt-in later. **MCP *expansion* (write/workflow tools) is deferred until after the contract freeze** (Bundle C #287/#288 → #286 versioning): the tool catalog binds to operationIds/schemas that #287 shows are still unstable (93/243 ops lack operationIds), so expanding now would churn. Don't pick up MCP expansion before the freeze. **Recommended next** (parallel-safe by disjoint surface): - **Bundle B — #283** (content-type XSS): user indicated it's landing this session in another session — do not double-pick. - **Bundle C (#287/#288)** — OpenAPI contract freeze; **sequence AFTER Bundle B lands** (both regenerate `v1.json`/`endpoint-index`), then **#286 (versioning) LAST** (rewrites every route + all generated artifacts → after all other in-flight `/api` work). - **PR #76 (MCP)** is now security-clean but **not on the critical arc**: before it can merge to `main` it needs a `main`-merge (it's ~5 days behind, pre-Bundle-A/B/#91b) + generated-artifact regen + full-solution CI; go-live also gated on #197. Leave it parked until the arc reaches go-live, or land the main-merge opportunistically. `in-progress` removed from #289.
Author
Owner

Session 2026-07-12 (midday) — #295 [PLAN-MODE] browser-SPA auth design pass

Closed: none (design-only pass — no code).
Design ratified: #295 — browser SPA auth → OIDC + local-login session, API key demoted to machine/MCP-only. Full implementable spec posted on #295 (Fable [PLAN-MODE] pass, owner-ruled all 4 forks = recommended defaults: setup-claim first-run / ConfigElement single-admin no-migration / app-owned OIDC no-header-trust / ApiKey-only OpenAPI). Architecture: evolve the global ApiKeyAuthorizationFilter → session-OR-key gate (fail-closed by omission preserved), revive the retained-but-inert OIDC middleware in the legacy MapWhen branch, ConfigElement-backed PBKDF2 local admin + security-stamp revocation, SameSite=Lax cookie + X-CSRF custom-header CSRF defense, first-run setup-claim screen with env-seed/machine-key recovery. 3-PR phasing (server → SPA → key rotation), no DB migration in PR1, backward-compatible transition.

Filed: none.

Triage: #295 = gate-blocker — it's a #197 Phase-1 remote-exposure sub-item (revises Bundle A's human-facing key-paste gap), priority: high + security. Already slotted under arc item 5 by its filer; this pass produced its buildable design. Not a new arc position.

Reviews: no adversarial-reviewer audits picked up. Snapshot unchanged (#19 all slices delivered, #24 App.tsx-modularization plan review still pending pickup, #20 remaining = #235). Design pass was single-model (Fable) per owner's model-routing choice this session; no independent adversarial pass needed on a design-only artifact — the implementation PRs get the usual per-PR review loop.

Arc change: none. Arc item 5 (#197) frontier unchanged: Bundle C (#287/#288) in flight → #286 versioning LAST; #295 is a parallel Phase-1 item whose design is now unblocked for implementation.

Recommended next (dependency-noted):

  • #295 implementation (PR1 server-only) — ready to pick up, but branch off main AFTER Bundle C (#287/#288) lands: Bundle C extracted ApiKeyAuthorizationFilter.EndpointRequiresKey(...) (the combined gate must compose it, not re-derive) + the ApiKey OpenAPI scheme + ChannelController/Startup edits — branching earlier conflicts. PR1 has no migration and keeps the SPA working via its stored key (backward compatible).
  • Bundle C (#287/#288) — still in-progress (claimed ~01:36, another session); do not double-pick.
  • #286 (versioning) — LAST, after all in-flight /api work.
  • Parallel-safe pickups meanwhile: adversarial-reviewer #24 (App.tsx plan review, read-only) or #235 (#20 async-lock remaining item).

in-progress removed from #295 (design done; implementation re-claimed when picked up).

## Session 2026-07-12 (midday) — #295 [PLAN-MODE] browser-SPA auth design pass **Closed**: none (design-only pass — no code). **Design ratified**: **#295** — browser SPA auth → **OIDC + local-login session**, API key **demoted to machine/MCP-only**. Full implementable spec posted on #295 (Fable [PLAN-MODE] pass, owner-ruled all 4 forks = recommended defaults: setup-claim first-run / `ConfigElement` single-admin no-migration / app-owned OIDC no-header-trust / `ApiKey`-only OpenAPI). Architecture: evolve the global `ApiKeyAuthorizationFilter` → session-OR-key gate (fail-closed by omission preserved), revive the retained-but-inert OIDC middleware in the legacy `MapWhen` branch, `ConfigElement`-backed PBKDF2 local admin + security-stamp revocation, `SameSite=Lax` cookie + `X-CSRF` custom-header CSRF defense, first-run setup-claim screen with env-seed/machine-key recovery. 3-PR phasing (server → SPA → key rotation), no DB migration in PR1, backward-compatible transition. **Filed**: none. **Triage**: **#295 = gate-blocker** — it's a **#197 Phase-1 remote-exposure sub-item** (revises Bundle A's human-facing key-paste gap), `priority: high` + security. Already slotted under arc item 5 by its filer; this pass produced its buildable design. Not a new arc position. **Reviews**: no adversarial-reviewer audits picked up. Snapshot **unchanged** (#19 all slices delivered, #24 App.tsx-modularization plan review still pending pickup, #20 remaining = #235). Design pass was single-model (Fable) per owner's model-routing choice this session; no independent adversarial pass needed on a design-only artifact — the implementation PRs get the usual per-PR review loop. **Arc change**: none. Arc item 5 (#197) frontier unchanged: Bundle C (#287/#288) in flight → #286 versioning LAST; #295 is a parallel Phase-1 item whose **design is now unblocked for implementation**. **Recommended next** (dependency-noted): - **#295 implementation (PR1 server-only)** — ready to pick up, but **branch off `main` AFTER Bundle C (#287/#288) lands**: Bundle C extracted `ApiKeyAuthorizationFilter.EndpointRequiresKey(...)` (the combined gate must compose it, not re-derive) + the `ApiKey` OpenAPI scheme + `ChannelController`/`Startup` edits — branching earlier conflicts. PR1 has no migration and keeps the SPA working via its stored key (backward compatible). - **Bundle C (#287/#288)** — still `in-progress` (claimed ~01:36, another session); do not double-pick. - **#286 (versioning)** — LAST, after all in-flight `/api` work. - Parallel-safe pickups meanwhile: **adversarial-reviewer #24** (App.tsx plan review, read-only) or **#235** (#20 async-lock remaining item). `in-progress` removed from #295 (design done; implementation re-claimed when picked up).
Author
Owner

Session 2026-07-12 (midday) — adversarial-reviewer#24 independent plan re-review

Closed: none (read-only review pass — no code).

Delivered: adversarial-reviewer#24 independent second-opinion pass on the App.tsx modularization plan (#243–#247), code-verified against main @ 9e2d1608. Verdict RATIFY — plan still sound; the prior 2026-07-11 pre-implementation review's structural claims all re-confirmed at current line numbers. No blocking findings. Note: the substantive pre-review was already done last session; this pass adds current-state corrections + one new invariant the pre-#202 review couldn't have seen. Folded the actionable outputs into the child issues.

Filed: none.

Triage: no new issues. Findings are recommendations to implementers (not code defects):

  • #246 → re-scope to Batch A (Dashboard+Guide) only; Batch B (Libraries) is already DONE — extracted by #202 itself (slice S4 99fab8e0). The review's headline "#246 collides with in-flight #202" flag is resolved by fact.
  • #245 (Playouts) → unblocked (its #231–#235 gate is satisfied; all closed). INV-3 (mutatingRef+409-refresh) re-confirmed; #212-class empty-state test to carry.
  • #247 → added INV-2b (App-owned guarded-popstate for LibrariesRouteScreen, a new #202-era silent-break contract), INV-4 update (two self-owning wrappers now — PlayoutsRouteScreen + new MediaRouteScreen), G-4 non-goal (don't unify wrapper vs. key-remount sub-path patterns), a PlaceholderScreen-fallthrough exhaustiveness test, and refreshed INV-1/INV-2 line numbers. G-1 re-confirmed: #238 still a hard #247 prerequisite, still open+unowned.

Reviews: picked up + delivered adversarial-reviewer#24 (independent pass). Snapshot change: #24 pre-implementation review is complete and re-ratified; it stays open only for per-PR spot-checks of #245/#246A/#247 and the final post-#247 review — no new extraction PR exists to spot-check yet. #24's original claim (2026-07-11 15:37) was the same author; this was a genuinely independent re-pass against current main, not a duplicate. Other audits unchanged (#19 all slices delivered; #20 remaining = none-open; #21–#23 delivered).

Arc change: none. The modularization epic (milestone "ChicoryTV modularization") is not an arc gate — parallel to the #197 gate. Arc frontier unchanged: Bundle C (#287/#288) in flight → #286 versioning LAST; #295 impl after Bundle C; #289/PR#76 parked.

Recommended next (parallel-safe by disjoint surface):

  • Arc-critical stays claimed/blocked: Bundle C (#287/#288, in-progress ~01:36) → then #286 (versioning, LAST) and #295 impl. Don't double-pick Bundle C.
  • Modularization epic (non-gate, now well-specified): #245 Playouts (eligible, largest) or #246 Batch A (Dashboard+Guide) (eligible, low-risk) are both pickup-ready and disjoint from the /api arc work. #247 waits on #238 (still unowned — worth scheduling #238 as it gates the epic's final phase).
  • Note: the modularization frontend work and the #197 /api arc work touch different surfaces → safe to run concurrently with a Bundle-C session.

No in-progress labels to remove — reviewer-repo audits are claimed by comment only.

## Session 2026-07-12 (midday) — adversarial-reviewer#24 independent plan re-review **Closed**: none (read-only review pass — no code). **Delivered**: **adversarial-reviewer#24** independent second-opinion pass on the App.tsx modularization plan (#243–#247), code-verified against `main @ 9e2d1608`. Verdict **RATIFY — plan still sound**; the prior 2026-07-11 pre-implementation review's structural claims all re-confirmed at current line numbers. No blocking findings. Note: the substantive pre-review was already done last session; this pass adds current-state corrections + one new invariant the pre-#202 review couldn't have seen. Folded the actionable outputs into the child issues. **Filed**: none. **Triage**: no new issues. Findings are recommendations to implementers (not code defects): - **#246** → re-scope to **Batch A (Dashboard+Guide) only**; **Batch B (Libraries) is already DONE** — extracted by #202 itself (slice S4 `99fab8e0`). The review's headline "#246 collides with in-flight #202" flag is resolved by fact. - **#245 (Playouts)** → **unblocked** (its #231–#235 gate is satisfied; all closed). INV-3 (`mutatingRef`+409-refresh) re-confirmed; #212-class empty-state test to carry. - **#247** → added **INV-2b** (App-owned guarded-popstate for `LibrariesRouteScreen`, a new #202-era silent-break contract), **INV-4 update** (two self-owning wrappers now — `PlayoutsRouteScreen` + new `MediaRouteScreen`), **G-4** non-goal (don't unify wrapper vs. `key`-remount sub-path patterns), a PlaceholderScreen-fallthrough exhaustiveness test, and refreshed INV-1/INV-2 line numbers. **G-1 re-confirmed: #238 still a hard #247 prerequisite, still open+unowned.** **Reviews**: **picked up + delivered adversarial-reviewer#24** (independent pass). Snapshot change: #24 pre-implementation review is complete and re-ratified; it stays open only for per-PR spot-checks of #245/#246A/#247 and the final post-#247 review — **no new extraction PR exists to spot-check yet**. #24's original claim (2026-07-11 15:37) was the same author; this was a genuinely independent re-pass against current main, not a duplicate. Other audits unchanged (#19 all slices delivered; #20 remaining = none-open; #21–#23 delivered). **Arc change**: none. The modularization epic (milestone "ChicoryTV modularization") is **not** an arc gate — parallel to the #197 gate. Arc frontier unchanged: Bundle C (#287/#288) in flight → #286 versioning LAST; #295 impl after Bundle C; #289/PR#76 parked. **Recommended next** (parallel-safe by disjoint surface): - **Arc-critical** stays claimed/blocked: Bundle C (#287/#288, `in-progress` ~01:36) → then #286 (versioning, LAST) and #295 impl. Don't double-pick Bundle C. - **Modularization epic (non-gate, now well-specified)**: **#245 Playouts** (eligible, largest) or **#246 Batch A (Dashboard+Guide)** (eligible, low-risk) are both pickup-ready and disjoint from the `/api` arc work. **#247** waits on **#238** (still unowned — worth scheduling #238 as it gates the epic's final phase). - Note: the modularization frontend work and the #197 `/api` arc work touch different surfaces → safe to run concurrently with a Bundle-C session. *No `in-progress` labels to remove — reviewer-repo audits are claimed by comment only.*
Author
Owner

Session 2026-07-12 — #197 Bundle C (contract-freeze honesty) + channel re-key

Closed: #287 + #288 (PR #296, FF-merged to main c40e78d8; PR CI green — Build&test + EF migration integrity both ✅; branch deleted). Main post-merge image build (:latest) in progress. This is arc item 5 (#197) Phase-2 "Bundle C".

Filed: #295 — browser SPA auth = OIDC + local login (session), demote API key to external/MCP-only. · #297 — add channelId to PlayoutListItemResponseModel so the SPA reset button keys directly (nit from the re-key).

Triage:

  • #295 → gate-blocker (priority: high) — slots into #197 Phase-1 (remote-exposure): Bundle A gated /api behind the key and forces humans to hand-paste it (no bootstrap UI). Owner-directed target: /api/* accepts a session principal (OIDC/local login) OR X-Api-Key. Already being designed by another session / Fable now — do not double-pick. Partially revises Bundle A (#280/281/282/284/285). [PLAN-MODE].
  • #297 → backlog (priority: low) — SPA ergonomics; not a gate.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged (#19 all four slices delivered → still recommend closing #19; #24 App.tsx-modularization plan review still pending pickup; #20 remaining item = #235). PR #296 got the full independent loop: cold zero-context fork review (MERGEABLE-WITH-NITS) + Codex pass (MERGEABLE) — Codex caught an operationId visitation-order-stability LOW the fork missed; cold caught a MediaSources nullable-honesty LOW — both fixed in c40e78d8 + a fix-commit re-review (MERGEABLE) (the #242 discipline). Live headless E2E 5/5 (API-key gate, channel-editor round-trip, reset→202, DayOfWeek round-trip, search). Codex quota available and used.

Arc change: none to ordering. Arc item 5 (#197) Phase-2 Bundle C is DONE. Remaining #197: #286 (/api/v1 versioning, LAST) + #295 (new Phase-1 auth redesign, in design) + #289 (MCP, already landed on PR#76 branch, awaiting main-merge at go-live). docs/decisions.md got 3 entries (2026-07-12): #287 by-construction OpenAPI honesty, #288 VM-wrap + reverse the §7a Version-leak, and channel routes key on Id not Number (playout/reset re-keyed {channelNumber}→{id:int}).

Recommended next (parallel-safe by disjoint surface):

  • #286 (/api/v1 versioning) — now unblocked (Bundle C was the last /api contract change before versioning). It rewrites every route + regenerates all artifacts, so sequence it AFTER any other in-flight /api work and coordinate with #295's /api-filter changes (branch #286 off main after #295's filter edits land, or vice-versa — whoever's second merges main first).
  • #295 is in design (other session/Fable) — its implementation should build on this PR's shared ApiKeyAuthorizationFilter.EndpointRequiresKey predicate (clean seam for "session OR key").
  • Note: user's main checkout pulled to c40e78d8.

in-progress removed from #287/#288.

## Session 2026-07-12 — #197 Bundle C (contract-freeze honesty) + channel re-key **Closed**: #287 + #288 (PR #296, FF-merged to `main` `c40e78d8`; PR CI green — Build&test + EF migration integrity both ✅; branch deleted). Main post-merge image build (`:latest`) in progress. This is **arc item 5 (#197) Phase-2 "Bundle C"**. **Filed**: #295 — browser SPA auth = OIDC + local login (session), demote API key to external/MCP-only. · #297 — add `channelId` to `PlayoutListItemResponseModel` so the SPA reset button keys directly (nit from the re-key). **Triage**: - **#295 → gate-blocker** (priority: high) — slots into **#197 Phase-1** (remote-exposure): Bundle A gated `/api` behind the key and forces humans to hand-paste it (no bootstrap UI). Owner-directed target: `/api/*` accepts a session principal (OIDC/local login) OR `X-Api-Key`. **Already being designed by another session / Fable now** — do not double-pick. Partially revises Bundle A (#280/281/282/284/285). `[PLAN-MODE]`. - **#297 → backlog** (priority: low) — SPA ergonomics; not a gate. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot **unchanged** (#19 all four slices delivered → still recommend closing #19; #24 App.tsx-modularization plan review still pending pickup; #20 remaining item = #235). PR #296 got the full independent loop: cold zero-context fork review (MERGEABLE-WITH-NITS) + **Codex pass (MERGEABLE)** — Codex caught an operationId visitation-order-stability LOW the fork missed; cold caught a MediaSources nullable-honesty LOW — both fixed in `c40e78d8` + a **fix-commit re-review (MERGEABLE)** (the #242 discipline). Live headless **E2E 5/5** (API-key gate, channel-editor round-trip, reset→202, DayOfWeek round-trip, search). Codex quota available and used. **Arc change**: none to ordering. Arc item 5 (#197) Phase-2 **Bundle C is DONE**. Remaining #197: **#286** (`/api/v1` versioning, LAST) + **#295** (new Phase-1 auth redesign, in design) + #289 (MCP, already landed on PR#76 branch, awaiting main-merge at go-live). `docs/decisions.md` got 3 entries (2026-07-12): #287 by-construction OpenAPI honesty, #288 VM-wrap + reverse the §7a Version-leak, and **channel routes key on `Id` not `Number`** (playout/reset re-keyed `{channelNumber}`→`{id:int}`). **Recommended next** (parallel-safe by disjoint surface): - **#286** (`/api/v1` versioning) — **now unblocked** (Bundle C was the last `/api` contract change before versioning). It rewrites every route + regenerates all artifacts, so sequence it AFTER any other in-flight `/api` work and coordinate with #295's `/api`-filter changes (branch #286 off main *after* #295's filter edits land, or vice-versa — whoever's second merges main first). - **#295** is in design (other session/Fable) — its implementation should build on this PR's shared `ApiKeyAuthorizationFilter.EndpointRequiresKey` predicate (clean seam for "session OR key"). - Note: user's main checkout pulled to `c40e78d8`. `in-progress` removed from #287/#288.
Author
Owner

Session 2026-07-12 (afternoon) — #271 external-collections scan-status surface

Closed: #271 (PR #298, FF-merged to main 4112413c; PR CI green — Build & test + EF migration integrity + Docs reminder ✅; branch deleted). Main post-merge image build in progress.

What/why: Added GET /api/media-sources/collections-scan-status (family-global lock state from IEntityLocker.Are{X}CollectionsLocked()) + rewired useCollectionsScan to poll+reconcile authoritatively (seeds on mount, drops COLLECTIONS_PENDING_TIMEOUT_MS). This removes the last F9 API gap — arc item 4 (#91b) is now unblocked: Libraries.razor's collections-scan affordance has full authoritative parity (deep-scan already landed via #235). Reused the library hook's grace-tick reconcile (generalized over key type); a cold-review nit fix disables all rows of a family on click (family-global lock parity with Blazor).

Filed: none.

Triage: no new issues.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged (#19 all slices delivered [recommend closing]; #24 App.tsx-modularization plan re-ratified last session, open only for per-PR spot-checks; #20 remaining = none open). PR #298 got the per-PR loop: cold zero-context fork review (MERGEABLE-WITH-NITS; both nits — family-global row-disable + sibling-POST guard — fixed in-PR) + a fix-commit re-review (MERGEABLE; #242 discipline). Codex not used (self-contained SPA-mostly change, fully fork-covered). Live headless smoke: 401 without key, 200 + [] with key.

Arc change: none to ordering. Arc item 4 (#91b) is now free of its #271 F9 gate; its other gates (#204/#205/#206 removal-plan items) landed 2026-07-11. Arc item 5 (#197) frontier unchanged: #286 (/api/v1 versioning, LAST) + #295 (browser-SPA auth, design ratified, impl unclaimed) remain; #289/PR#76 parked for go-live.

Recommended next (parallel-safe by disjoint surface):

  • #91b Blazor-removal PR (arc item 4) — now unblocked; large, needs the mandatory cold adversarial go/no-go (Fable per escalation rules) + full removal recon from the 2026-07-09 notes on #91.
  • #295 impl PR1 (server auth) — branch off current main (Bundle C landed); coordinate /api-filter edits with #286.
  • #286 (/api/v1 versioning) — LAST; after #295's filter edits and any other in-flight /api work.
  • Note: user's main checkout should pull to 4112413c.

in-progress removed from #271.

## Session 2026-07-12 (afternoon) — #271 external-collections scan-status surface **Closed**: #271 (PR #298, FF-merged to `main` `4112413c`; PR CI green — Build & test + EF migration integrity + Docs reminder ✅; branch deleted). Main post-merge image build in progress. **What/why**: Added `GET /api/media-sources/collections-scan-status` (family-global lock state from `IEntityLocker.Are{X}CollectionsLocked()`) + rewired `useCollectionsScan` to poll+reconcile authoritatively (seeds on mount, drops `COLLECTIONS_PENDING_TIMEOUT_MS`). This removes the last F9 API gap — **arc item 4 (#91b) is now unblocked**: `Libraries.razor`'s collections-scan affordance has full authoritative parity (deep-scan already landed via #235). Reused the library hook's grace-tick reconcile (generalized over key type); a cold-review nit fix disables all rows of a family on click (family-global lock parity with Blazor). **Filed**: none. **Triage**: no new issues. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot **unchanged** (#19 all slices delivered [recommend closing]; #24 App.tsx-modularization plan re-ratified last session, open only for per-PR spot-checks; #20 remaining = none open). PR #298 got the per-PR loop: cold zero-context fork review (MERGEABLE-WITH-NITS; both nits — family-global row-disable + sibling-POST guard — fixed in-PR) + a fix-commit re-review (MERGEABLE; #242 discipline). Codex not used (self-contained SPA-mostly change, fully fork-covered). Live headless smoke: 401 without key, 200 + `[]` with key. **Arc change**: none to ordering. Arc item 4 (#91b) is now free of its #271 F9 gate; its other gates (#204/#205/#206 removal-plan items) landed 2026-07-11. Arc item 5 (#197) frontier unchanged: **#286** (`/api/v1` versioning, LAST) + **#295** (browser-SPA auth, design ratified, impl unclaimed) remain; #289/PR#76 parked for go-live. **Recommended next** (parallel-safe by disjoint surface): - **#91b Blazor-removal PR** (arc item 4) — now unblocked; large, needs the mandatory cold adversarial go/no-go (Fable per escalation rules) + full removal recon from the 2026-07-09 notes on #91. - **#295 impl PR1 (server auth)** — branch off current `main` (Bundle C landed); coordinate `/api`-filter edits with #286. - **#286 (`/api/v1` versioning)** — LAST; after #295's filter edits and any other in-flight `/api` work. - Note: user's main checkout should pull to `4112413c`. `in-progress` removed from #271.
Author
Owner

Session 2026-07-12 (afternoon) — arc + pickup-list correction: stale "done" items

Closed: none (bookkeeping-only — no code).

Correction 1 — arc item 4 (#91b Blazor removal) was already DONE. The arc body + the two most-recent session comments (12:27, 15:09 today) still framed it as "now unblocked / recommended next," but it had already merged 2026-07-11. Verified against main @ 4112413c:

  • Removal landed via branch feat/91b-blazor-removal → 408b0deb feat(91b): remove legacy Blazor Server UI (#91 phase b) (+ cleanup 4407dd53, 3c5d0590, merge 78262ea4).
  • Tree check: 0 .razor files; no Pages/Shared/Components dirs; no AddServerSideBlazor / MapBlazorHub / AddMudServices / MudBlazor package refs (only historical comments mention Blazor). Retired routes 302 via LegacyUiRedirects.cs + Startup catch-all.
  • #91 CLOSED 2026-07-11 21:31; milestone "Blazor removal (#91 phase b)" CLOSED (0 open / 9 closed).
  • CLAUDE.md + docs/README.md were already accurate — only this tracker was stale. Root cause: #271 (external-collections scan-status, landed today) was narrated as "the last F9 gate unblocking #91b," but the removal PR had already merged the prior evening — #271 was a follow-up parity gap, not a blocker.

Correction 2 — the Priority-pickups list was stale too. #251 + #252 are DONE (PR #258, merged 87060aa3, 2026-07-11) and #254 is DONE (a1bd303c). Only #253 (replace-all optimistic concurrency, + open follow-up #269 and nit #265) genuinely remains as a priority pickup. My first pass at this correction preserved the stale #251/#252 text; fixed in a second body edit.

Filed: none.

Triage: no new issues.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot refreshed to 2026-07-12 in-body (#21/#22/#23 findings #251/#252/#254 now CLOSED, #253 open; #19 recommend closing; #24 open for per-PR spot-checks; #20 no open item).

Arc change: body edited (twice) — items 1–4 marked DONE (item 4 with removal evidence + stale-framing note); item 5 (#197) marked the current arc frontier; Priority-pickups corrected (#251/#252 DONE, #253+#269/#265 the remaining one); Backlog #254 marked DONE; added a "verify the issue is actually OPEN before claiming" note to the pickup protocol.

Recommended next (arc frontier = #197, parallel-safe by disjoint surface):

  • Arc-critical (#197): #295 impl PR1 (server auth — appears actively claimed, updated ~15:40, still in-progress; do not double-pick) → #286 (/api/v1 versioning, LAST). #289/PR#76 parked for go-live main-merge.
  • Priority pickup: #253 concurrency contract (Fable/[PLAN-MODE] for remaining scope; verify exact remaining scope vs #269/#265 on the issue first).
  • Non-gate: modularization #245 (Playouts) / #246 Batch A eligible; #238 gates #247 and is still unowned.
## Session 2026-07-12 (afternoon) — arc + pickup-list correction: stale "done" items **Closed**: none (bookkeeping-only — no code). **Correction 1 — arc item 4 (#91b Blazor removal) was already DONE.** The arc body + the two most-recent session comments (12:27, 15:09 today) still framed it as "now unblocked / recommended next," but it had already merged **2026-07-11**. Verified against `main @ 4112413c`: - Removal landed via branch `feat/91b-blazor-removal` → `408b0deb feat(91b): remove legacy Blazor Server UI (#91 phase b)` (+ cleanup `4407dd53`, `3c5d0590`, merge `78262ea4`). - Tree check: **0 `.razor` files**; no `Pages/Shared/Components` dirs; no `AddServerSideBlazor` / `MapBlazorHub` / `AddMudServices` / `MudBlazor` package refs (only historical *comments* mention Blazor). Retired routes 302 via `LegacyUiRedirects.cs` + Startup catch-all. - **#91 CLOSED** 2026-07-11 21:31; **milestone "Blazor removal (#91 phase b)" CLOSED** (0 open / 9 closed). - CLAUDE.md + `docs/README.md` were already accurate — only this tracker was stale. Root cause: #271 (external-collections scan-status, landed today) was narrated as "the last F9 gate unblocking #91b," but the removal PR had already merged the prior evening — #271 was a follow-up parity gap, not a blocker. **Correction 2 — the Priority-pickups list was stale too.** #251 + #252 are **DONE** (PR #258, merged `87060aa3`, 2026-07-11) and #254 is **DONE** (`a1bd303c`). Only **#253** (replace-all optimistic concurrency, + open follow-up **#269** and nit **#265**) genuinely remains as a priority pickup. My first pass at this correction preserved the stale #251/#252 text; fixed in a second body edit. **Filed**: none. **Triage**: no new issues. **Reviews**: no adversarial-reviewer audits picked up/filed — snapshot refreshed to 2026-07-12 in-body (#21/#22/#23 findings #251/#252/#254 now CLOSED, #253 open; #19 recommend closing; #24 open for per-PR spot-checks; #20 no open item). **Arc change**: **body edited (twice)** — items 1–4 marked DONE (item 4 with removal evidence + stale-framing note); item 5 (#197) marked the current arc frontier; Priority-pickups corrected (#251/#252 DONE, #253+#269/#265 the remaining one); Backlog #254 marked DONE; added a "verify the issue is actually OPEN before claiming" note to the pickup protocol. **Recommended next** (arc frontier = #197, parallel-safe by disjoint surface): - **Arc-critical (#197)**: **#295** impl PR1 (server auth — appears actively claimed, updated ~15:40, still `in-progress`; do not double-pick) → **#286** (`/api/v1` versioning, LAST). #289/PR#76 parked for go-live main-merge. - **Priority pickup**: **#253** concurrency contract (Fable/[PLAN-MODE] for remaining scope; verify exact remaining scope vs #269/#265 on the issue first). - **Non-gate**: modularization **#245 (Playouts)** / **#246 Batch A** eligible; **#238** gates #247 and is still unowned.
Author
Owner

Session 2026-07-12 — queue-drift structural fix (bookkeeping, no code)

Closed: none.

What/why: audit prompted by repeated pickups re-doing finished work (thought closed issues were still open). The 16:00 correction had patched the symptoms (#91b, #251/#252) but not the cause. Root cause = status embedded in append-only prose (arc inline DONE markers + each comment's "Recommended next") which lags live issue state, worst across parallel sessions narrating each other's merges. A Fable review of the first fix caught that my kickoff edit re-planted the bug (hardcoded "#197 cluster" / "#91b milestone CLOSED" inside the "never trust prose-status" rule) + a label-vs-milestone mechanism ambiguity.

Arc change — body edited (per body-edit protocol, noting here):

  • Arc pruned of status → closed items 1–4 + closed priority-pickups (#251/#252/#254) + shipped #197 bundles moved to a new "# Done (history)" section. The arc now lists only OPEN work (frontier #197 sub-items #286/#295/#289; #253 priority pickup; backlog). Status now lives ONLY in live Gitea state → can't drift.
  • Source-of-truth blockquote de-hardcoded; gate/frontier defined structurally (lowest-numbered open arc item).
  • Pickup step 2: added label-ranking (arc order → gate via review label/milestone → priority:) + explicit pick-order across arc-frontier / priority-pickup / audit pools.
  • Reverted the "Candidate next" rename back to "Recommended next" (matches every real comment) but kept the "hint, re-verify OPEN" reframing.
  • Session-end rule now: "move a closed item to Done (history) in the same session".

Companion fixes to the kickoff file docs/handoffs/chicorytv-issue-queue.md pushed to main (aed4b5c0): same source-of-truth + structural-frontier + ranking rules, bounded comment read (~6 newest), MCP-down REST fallback, #237 is canonical pointer, new drift-root-cause lore bullet.

Filed: none. Triage: n/a. Reviews: one Fable quality/effectiveness review of the handoff system (verdict: acute failure fixed; must-fixes = de-hardcode + prune arc status + unify protocol — all now applied). Recommended next: no arc change; deferred nice-to-haves = tier/prune the ~40-bullet lore into "standing rules + incident archive" and promote settled one-time recipes into docs/ (owner's editorial call).

## Session 2026-07-12 — queue-drift structural fix (bookkeeping, no code) **Closed**: none. **What/why**: audit prompted by repeated pickups re-doing finished work (thought closed issues were still open). The 16:00 correction had patched the *symptoms* (#91b, #251/#252) but not the cause. **Root cause = status embedded in append-only prose** (arc inline `DONE` markers + each comment's "Recommended next") which lags live issue state, worst across parallel sessions narrating each other's merges. A Fable review of the first fix caught that my kickoff edit *re-planted* the bug (hardcoded "#197 cluster" / "#91b milestone CLOSED" inside the "never trust prose-status" rule) + a label-vs-milestone mechanism ambiguity. **Arc change — body edited** (per body-edit protocol, noting here): - **Arc pruned of status** → closed items 1–4 + closed priority-pickups (#251/#252/#254) + shipped #197 bundles moved to a new **"# Done (history)"** section. The arc now lists only OPEN work (frontier #197 sub-items #286/#295/#289; #253 priority pickup; backlog). Status now lives ONLY in live Gitea state → can't drift. - **Source-of-truth blockquote** de-hardcoded; gate/frontier defined structurally (lowest-numbered open arc item). - **Pickup step 2**: added label-ranking (arc order → gate via `review` label/milestone → `priority:`) + explicit pick-order across arc-frontier / priority-pickup / audit pools. - Reverted the "Candidate next" rename back to **"Recommended next"** (matches every real comment) but kept the "hint, re-verify OPEN" reframing. - Session-end rule now: "move a closed item to Done (history) in the same session". Companion fixes to the kickoff file `docs/handoffs/chicorytv-issue-queue.md` pushed to `main` (`aed4b5c0`): same source-of-truth + structural-frontier + ranking rules, bounded comment read (~6 newest), MCP-down REST fallback, `#237 is canonical` pointer, new drift-root-cause lore bullet. **Filed**: none. **Triage**: n/a. **Reviews**: one Fable quality/effectiveness review of the handoff system (verdict: acute failure fixed; must-fixes = de-hardcode + prune arc status + unify protocol — all now applied). **Recommended next**: no arc change; deferred nice-to-haves = tier/prune the ~40-bullet lore into "standing rules + incident archive" and promote settled one-time recipes into `docs/` (owner's editorial call).
Author
Owner

Session 2026-07-12 (evening) — #295 PR1 browser-SPA session auth

Closed: none. PR #300 merged to main 48d256f8 (FF; branch deleted) — but #295 stays OPEN (PR1 of 3). #295 is a #197 Phase-1 sub-item (arc item 5), the human-facing "paste the machine key" gap from Bundle A. PR1 = server-only session-OR-key gate, backward compatible (SPA still uses the machine key).

Filed:

  • #299 — CI: codify the accumulated manual live-E2E flows into an automated functional-E2E harness (only IPTV smoke is in CI today).
  • #301 — side-effecting [RequiresAuthentication] GETs (troubleshoot playback/archive) aren't CSRF-covered once sessions are a normal SPA credential.

Triage (verify against live Gitea state at pickup, not this line):

  • #301 → gate-blocker for #295 PR2 (security + priority: high). Latent in PR1 (no session client yet); must close before PR2 makes the cookie the SPA's normal credential.
  • #299 → priority pickup (ci-cd + priority: high). Not an arc gate — regression insurance for shipped functional flows; parallel-safe.

Reviews: no adversarial-reviewer audits picked up/filed — snapshot unchanged. PR #300 ran the full per-PR loop: 1 cold zero-context fork (MERGEABLE-WITH-NITS) + 3 independent Codex passes (BLOCKED → BLOCKED → MERGEABLE) as fix-commit re-reviews (#242 discipline). Codex earned its keep — it caught concurrency + startup-race defects the fork missed (login-during-password-change survival, non-atomic first-claim, env-seed-vs-setup race, client-side-only logout); all folded in as follow-up commits. Codex quota available and used.

Arc change: none to ordering, and no body edit (per the 2026-07-12 live-state-is-truth protocol — closed items live in Gitea, not inline markers). Arc item 5 (#197) Phase-1: #295 PR1 done; PR2 (SPA)/PR3 (key rotation) remain.

Recommended next (candidates — re-derive the frontier from live state at pickup; lowest-numbered OPEN arc item + its review/priority children):

  • #286 (/api/v1 versioning) — the LAST /api contract change; coordinate its filter edits with #295's ApiAuthorizationFilter.
  • #295 PR2 (SPA session cutover) — now unblocked by PR1; gated on #301 (fix the side-effecting-GET CSRF as part of it) + needs the spa-conventions §5e update.
  • #299 (CI E2E harness) — parallel-safe, high-value, disjoint from the arc.
  • #289/PR#76 (MCP) still parked for go-live; needs a main-merge when picked up.
  • Deployment note carried to go-live: set ForwardedHeaders:KnownProxies/:KnownNetworks behind the proxy (correct client-IP/HTTPS + rate-limiter integrity).

in-progress removed from #295.

## Session 2026-07-12 (evening) — #295 PR1 browser-SPA session auth **Closed**: none. **PR #300 merged** to main `48d256f8` (FF; branch deleted) — but **#295 stays OPEN** (PR1 of 3). #295 is a **#197 Phase-1** sub-item (arc item 5), the human-facing "paste the machine key" gap from Bundle A. PR1 = server-only session-OR-key gate, backward compatible (SPA still uses the machine key). **Filed**: - **#299** — CI: codify the accumulated manual live-E2E flows into an automated functional-E2E harness (only IPTV smoke is in CI today). - **#301** — side-effecting `[RequiresAuthentication]` GETs (troubleshoot playback/archive) aren't CSRF-covered once sessions are a normal SPA credential. **Triage** (verify against live Gitea state at pickup, not this line): - **#301 → gate-blocker for #295 PR2** (`security` + `priority: high`). Latent in PR1 (no session client yet); must close before PR2 makes the cookie the SPA's normal credential. - **#299 → priority pickup** (`ci-cd` + `priority: high`). Not an arc gate — regression insurance for shipped functional flows; parallel-safe. **Reviews**: no `adversarial-reviewer` audits picked up/filed — snapshot unchanged. PR #300 ran the full per-PR loop: 1 cold zero-context fork (MERGEABLE-WITH-NITS) + **3 independent Codex passes** (BLOCKED → BLOCKED → MERGEABLE) as fix-commit re-reviews (#242 discipline). Codex earned its keep — it caught concurrency + startup-race defects the fork missed (login-during-password-change survival, non-atomic first-claim, env-seed-vs-setup race, client-side-only logout); all folded in as follow-up commits. Codex quota available and used. **Arc change**: none to ordering, and **no body edit** (per the 2026-07-12 live-state-is-truth protocol — closed items live in Gitea, not inline markers). Arc item 5 (#197) Phase-1: #295 PR1 done; PR2 (SPA)/PR3 (key rotation) remain. **Recommended next** (candidates — re-derive the frontier from live state at pickup; lowest-numbered OPEN arc item + its `review`/`priority` children): - **#286** (`/api/v1` versioning) — the LAST `/api` contract change; coordinate its filter edits with #295's `ApiAuthorizationFilter`. - **#295 PR2** (SPA session cutover) — now unblocked by PR1; **gated on #301** (fix the side-effecting-GET CSRF as part of it) + needs the `spa-conventions §5e` update. - **#299** (CI E2E harness) — parallel-safe, high-value, disjoint from the arc. - #289/PR#76 (MCP) still parked for go-live; needs a main-merge when picked up. - **Deployment note carried to go-live**: set `ForwardedHeaders:KnownProxies`/`:KnownNetworks` behind the proxy (correct client-IP/HTTPS + rate-limiter integrity). `in-progress` removed from #295.
Author
Owner

Session 2026-07-12 (evening) — #269 500-exposure slice

Closed: none. Merged PR #302 — the DbUpdateConcurrencyException→500 slice of #269 (18 non-If-Match root writers force-write past a concurrent Version bump). #269 kept OPEN, re-scoped to its remaining cross-editor ETag-rotation completeness (→ #197 input). in-progress removed.

Filed: none.

Triage: no new issues.

Reviews: no adversarial-reviewer audits picked up. PR #302 got the full independent loop — 2 cold forks (initial diff + fix-commit re-review, #242 discipline) + 1 Codex pass on the initial diff; genuinely complementary (Codex + fork converged on the ErasePlayoutHistory HIGH the first sweep missed; the re-review caught an accidental web/node_modules symlink I'd committed). Codex quota available, used.

Arc change: none (body already carries the 2026-07-12 structural cure; #269 stays an open priority-pickup for its rotation remainder — status lives in live Gitea state, not here).

Dependency/conflict warnings: #269's rotation remainder + #265 (If-Match RFC 7232 semantics) + #253's contract all feed #197 — sequence the ETag-rotation/Add*ToCollection design call with #197's contract pass. PR #302 touched Scheduling/MediaCollections/Playouts/ProgramSchedules replace+delete handlers — coordinate if another session edits those. Main moved fast this session (PR base advanced 3× mid-merge, incl. the #295 auth PR1 landing on main); expect frequent re-merges on any longer-lived branch.

## Session 2026-07-12 (evening) — #269 500-exposure slice **Closed**: none. **Merged PR #302** — the `DbUpdateConcurrencyException`→500 slice of #269 (18 non-If-Match root writers force-write past a concurrent `Version` bump). **#269 kept OPEN**, re-scoped to its remaining cross-editor ETag-rotation completeness (→ #197 input). `in-progress` removed. **Filed**: none. **Triage**: no new issues. **Reviews**: no adversarial-reviewer audits picked up. PR #302 got the full independent loop — **2 cold forks** (initial diff + fix-commit re-review, #242 discipline) + **1 Codex pass** on the initial diff; genuinely complementary (Codex + fork converged on the `ErasePlayoutHistory` HIGH the first sweep missed; the re-review caught an accidental `web/node_modules` symlink I'd committed). Codex quota available, used. **Arc change**: none (body already carries the 2026-07-12 structural cure; #269 stays an open priority-pickup for its rotation remainder — status lives in live Gitea state, not here). **Dependency/conflict warnings**: #269's rotation remainder + **#265** (If-Match RFC 7232 semantics) + **#253**'s contract all feed **#197** — sequence the ETag-rotation/`Add*ToCollection` design call with #197's contract pass. PR #302 touched `Scheduling`/`MediaCollections`/`Playouts`/`ProgramSchedules` replace+delete handlers — coordinate if another session edits those. Main moved fast this session (PR base advanced 3× mid-merge, incl. the #295 auth PR1 landing on `main`); expect frequent re-merges on any longer-lived branch.
Author
Owner

Session 2026-07-12 — #303 Wave 3 process hooks (H3 + H9)

Closed: none — #303 is backlog (process/tooling), not an arc item; it stays OPEN for H10 + follow-ups. PR #305 merged → main 0123e589.
Filed: none.
Triage: no new issues.
Reviews: one adversarial fork review of PR #305 (verdict MERGEABLE, LOW/nit only, folded in). No adversarial-reviewer-repo audits picked up or filed.
Arc change: none (arc body untouched; #303 is off-arc).

Dependency/conflict warnings (cross-cutting — affects every session):

  • docs/decisions.md is now append-only by enforcement (H9). A commit that deletes or modifies an existing line is blocked by the Husky commit-msg hook and the blocking decisions-guard CI job. Normal new entries (TOC line + appended block = pure insertions) pass untouched. To edit a past entry (factual fix, or supersession), put [decisions-edit] in the commit message. Supersession convention: new dated entry + a > **Superseded …** banner on the old entry + a (superseded) TOC tag. Consolidation happens at each release; CI warns (non-blocking) once the file passes 1800 lines.
  • Root-level *.png staged files are now rejected by the Husky pre-commit (H3) — curl download endpoints / keep screenshots out of the repo root (already lore).
  • Requires a /hooks reload or fresh session for the Claude-side PreToolUse hooks to be live; the committed Husky hooks are live on any checkout after a web/ npm install.
## Session 2026-07-12 — #303 Wave 3 process hooks (H3 + H9) **Closed**: none — #303 is backlog (process/tooling), not an arc item; it stays OPEN for H10 + follow-ups. PR #305 merged → main `0123e589`. **Filed**: none. **Triage**: no new issues. **Reviews**: one adversarial fork review of PR #305 (verdict MERGEABLE, LOW/nit only, folded in). No adversarial-reviewer-repo audits picked up or filed. **Arc change**: none (arc body untouched; #303 is off-arc). **Dependency/conflict warnings (cross-cutting — affects every session):** - **`docs/decisions.md` is now append-only by enforcement (H9).** A commit that *deletes or modifies* an existing line is blocked by the Husky `commit-msg` hook and the blocking `decisions-guard` CI job. Normal new entries (TOC line + appended block = pure insertions) pass untouched. To edit a past entry (factual fix, or supersession), put **`[decisions-edit]`** in the commit message. Supersession convention: new dated entry + a `> **Superseded …**` banner on the old entry + a `(superseded)` TOC tag. Consolidation happens at each release; CI warns (non-blocking) once the file passes **1800 lines**. - **Root-level `*.png` staged files are now rejected** by the Husky `pre-commit` (H3) — curl download endpoints / keep screenshots out of the repo root (already lore). - Requires a `/hooks` reload or fresh session for the Claude-side PreToolUse hooks to be live; the committed Husky hooks are live on any checkout after a `web/` `npm install`.
Author
Owner

Session 2026-07-12 — #303 H10 (review-verdict merge-gate)

Closed: none — PR #306 merged → main 2879c012, but #303 stays OPEN (H10 was its last hook; four non-hook process follow-ups remain). Not an arc item.
Filed: none.
Triage: no new issues.
Reviews: 3 in-session adversarial fork passes on PR #306 (one per fix commit, dogfooding H10) — not adversarial-reviewer-repo audits; none filed there.
Arc change: none (this body untouched; #303 is backlog, parallel to the #197 arc).
Dependency/conflict warnings:

  • The H10 merge-gate is now LIVE on main (pretooluse-merge-consent.sh condition c). Any code PR you merge from now needs a PR comment line starting with Review-verdict: <MERGEABLE|APPROVED|BLOCKED> @ <current-head-sha> or the gate asks/denies — a stale verdict for an older commit is denied (the #242 mechanization). Docs/process-only PRs stay exempt (files under docs/|.claude/|.husky/|.gitea/|*.md). Post the adversarial/Codex verdict in that format after review; re-post referencing the new head after any fix commit. Convention + tiers: docs/decisions.md H10 entry, CLAUDE.md Task Completion Protocol.
  • The four #303 follow-ups (Codex-skip rubric, E2E-seeding-recipe promotion, out-of-ecosystem security scan, migration-on-prod-copy smoke) are independent, lighter, parallel-safe pickups if a session wants process work over the #197 frontier.
## Session 2026-07-12 — #303 H10 (review-verdict merge-gate) **Closed**: none — PR #306 merged → main `2879c012`, but #303 stays OPEN (H10 was its last hook; four non-hook process follow-ups remain). Not an arc item. **Filed**: none. **Triage**: no new issues. **Reviews**: 3 in-session adversarial fork passes on PR #306 (one per fix commit, dogfooding H10) — not adversarial-reviewer-repo audits; none filed there. **Arc change**: none (this body untouched; #303 is backlog, parallel to the #197 arc). **Dependency/conflict warnings**: - **The H10 merge-gate is now LIVE on main** (`pretooluse-merge-consent.sh` condition c). Any *code* PR you merge from now needs a PR comment line starting with `Review-verdict: <MERGEABLE|APPROVED|BLOCKED> @ <current-head-sha>` **or the gate asks/denies** — a stale verdict for an older commit is denied (the #242 mechanization). **Docs/process-only PRs stay exempt** (files under `docs/|.claude/|.husky/|.gitea/|*.md`). Post the adversarial/Codex verdict in that format after review; re-post referencing the new head after any fix commit. Convention + tiers: `docs/decisions.md` H10 entry, CLAUDE.md Task Completion Protocol. - The four #303 follow-ups (Codex-skip rubric, E2E-seeding-recipe promotion, out-of-ecosystem security scan, migration-on-prod-copy smoke) are independent, lighter, parallel-safe pickups if a session wants process work over the #197 frontier.
Author
Owner

Session 2026-07-12 (evening) — #269 cross-editor ETag rotation

Closed: #269 (PR #307, 7a9b30de, fast-forward to main). Arc: a #253 priority-pickup follow-up.

Filed: #308 — two concurrent same-item Add*ToCollection can both pass the membership check and the loser 500s on the composite-PK violation (SaveChangesForcingVersion catches only DbUpdateConcurrencyException).

Triage: #308 → backlog (bug, priority: low) — pre-existing, narrow (two identical requests racing), no corruption (PK guarantees ≤1 row).

Reviews: no adversarial-reviewer audits picked up. PR #307 ran the full independent loop: cold zero-context fork (MERGEABLE) + Codex (BLOCKED — F1 real: SaveChangesForcingVersion dropped the pending Version++ under a race, silently losing rotation) → Fable reconciled the fork-vs-Codex disagreement (Codex right on all 3; applied the delta-rebase fix + strengthened the race tests, negative-controlled by reverting the fix) → Codex re-review of the fix commit (F1 CLOSED, no new findings). The #303 H6 (## Done-when) + H10 (Review-verdict @ head) merge gates were both satisfied at merge.

Arc change: body edited — #269 moved to Done (history); the #253 priority-pickup line updated (#269 DONE, #265 now the remaining un-in-progress piece); #308 added to backlog.

Dependency/conflict warnings:

  • #265 (RFC-7232 If-Match parser reshape) is the remaining #253 piece — Fable-ratified design + full recon captured this session (10 controllers short-circuit Malformed→400 before the handler; reshape ExpectedVersion Option<int>→Seq<int>, CheckVersion→set-membership, valid-but-non-matching/weak/list → 412 not 400; 4 endpoints — Playout×2, Collection, MultiCollection, RerunCollection — lack a malformed-If-Match test today). Disjoint surface from #269 (shared parser + CheckVersion, not the handler saves).
  • The SaveChangesForcingVersion delta-rebase (this PR) touches the shared helper used by all 25 bumpers (incl. the pre-existing Add*ToPlaylist / schedule-item writers) — coordinate if another session edits it.
  • Rebased onto main mid-session (the #303 H9/H10 hooks landed in parallel); the H9 decisions.md append-only guard and H10 review-verdict gate are now live and were exercised end-to-end.
## Session 2026-07-12 (evening) — #269 cross-editor ETag rotation **Closed**: #269 (PR #307, `7a9b30de`, fast-forward to main). Arc: a #253 priority-pickup follow-up. **Filed**: #308 — two *concurrent* same-item `Add*ToCollection` can both pass the membership check and the loser 500s on the composite-PK violation (`SaveChangesForcingVersion` catches only `DbUpdateConcurrencyException`). **Triage**: **#308 → backlog** (`bug`, `priority: low`) — pre-existing, narrow (two identical requests racing), no corruption (PK guarantees ≤1 row). **Reviews**: no adversarial-reviewer audits picked up. PR #307 ran the full independent loop: cold zero-context fork (**MERGEABLE**) + Codex (**BLOCKED** — F1 real: `SaveChangesForcingVersion` dropped the pending `Version++` under a race, silently losing rotation) → **Fable** reconciled the fork-vs-Codex disagreement (Codex right on all 3; applied the delta-rebase fix + strengthened the race tests, negative-controlled by reverting the fix) → **Codex re-review of the fix commit** (F1 CLOSED, no new findings). The #303 H6 (`## Done-when`) + H10 (`Review-verdict @ head`) merge gates were both satisfied at merge. **Arc change**: **body edited** — #269 moved to Done (history); the #253 priority-pickup line updated (#269 DONE, **#265** now the remaining un-`in-progress` piece); #308 added to backlog. **Dependency/conflict warnings**: - **#265** (RFC-7232 If-Match parser reshape) is the remaining #253 piece — Fable-ratified design + full recon captured this session (10 controllers short-circuit Malformed→400 before the handler; reshape `ExpectedVersion` `Option<int>`→`Seq<int>`, `CheckVersion`→set-membership, valid-but-non-matching/weak/list → 412 not 400; 4 endpoints — Playout×2, Collection, MultiCollection, RerunCollection — lack a malformed-If-Match test today). Disjoint surface from #269 (shared parser + `CheckVersion`, not the handler saves). - The **`SaveChangesForcingVersion` delta-rebase** (this PR) touches the shared helper used by **all 25 bumpers** (incl. the pre-existing `Add*ToPlaylist` / schedule-item writers) — coordinate if another session edits it. - Rebased onto `main` mid-session (the #303 H9/H10 hooks landed in parallel); the H9 `decisions.md` append-only guard and H10 review-verdict gate are now live and were exercised end-to-end.
timothy added the priority: high label 2026-07-12 20:32:45 +02:00
Author
Owner

Session 2026-07-12 (cont.) — #303 hook program: H10, H11, H12, format CI

Closed: #311 + #312 (PR #313 → main 9b3b3963); #303 itself stays OPEN (2 non-hook follow-ups remain: security scan, migration smoke). Earlier this session: H10 (PR #306) and #303 follow-ups 1+2 (PR #309) also merged. All #303-program, not arc items.
Filed: #310 — de-BOM the 17 legacy .cs #269 touched; #311 — H11 rebase-hook + PR-scoped format CI; #312 — H12 session-end qualification audit.
Triage: #310 backlog (priority: low); #311 backlog→DONE (priority: medium); #312 backlog→DONE (priority: low). All qualified; H12 audit now reports the whole open queue clean (also labelled #237→high, #22→low).
Reviews: 5 in-session adversarial fork passes — 3 on the H10 PR (one per fix commit, dogfooding H10) + 1 on PR #313 (all MERGEABLE); #309 review-skipped per the new Codex rubric (pure-docs leaf, stated). None from the adversarial-reviewer repo.
Arc change: none (this body untouched; all #303-program, parallel to the #197 arc).
Dependency/conflict warnings — NEW enforcement live on main, affects every future session:

  • H11: git push from a branch behind origin/main is blocked — git fetch origin main && git rebase origin/main (do NOT merge main in). Escape ETV_SKIP_REBASE_CHECK=1. The lore is flipped to rebase-not-merge.
  • format CI job: a PR that touches a legacy-BOM .cs (2530 of them) must de-BOM/normalize that file (dotnet format --include <file>) or the blocking format check fails — the fix-as-you-touch rule (contributing.md §7). .cs-free PRs pass trivially.
  • H10 merge-gate is proven on non-docs PRs: a code PR now needs a Review-verdict: MERGEABLE @ <head-sha> comment + all ## Done-when boxes ticked to merge (PR #313 was the first to pass it live).
  • H12: run scripts/issue-qualification-audit.sh at session end; label anything it flags.
## Session 2026-07-12 (cont.) — #303 hook program: H10, H11, H12, format CI **Closed**: #311 + #312 (PR #313 → main `9b3b3963`); #303 itself stays OPEN (2 non-hook follow-ups remain: security scan, migration smoke). Earlier this session: H10 (PR #306) and #303 follow-ups 1+2 (PR #309) also merged. All #303-program, not arc items. **Filed**: #310 — de-BOM the 17 legacy `.cs` #269 touched; #311 — H11 rebase-hook + PR-scoped format CI; #312 — H12 session-end qualification audit. **Triage**: #310 backlog (`priority: low`); #311 backlog→DONE (`priority: medium`); #312 backlog→DONE (`priority: low`). All qualified; H12 audit now reports the whole open queue clean (also labelled #237→high, #22→low). **Reviews**: 5 in-session adversarial fork passes — 3 on the H10 PR (one per fix commit, dogfooding H10) + 1 on PR #313 (all MERGEABLE); #309 review-skipped per the new Codex rubric (pure-docs leaf, stated). None from the adversarial-reviewer repo. **Arc change**: none (this body untouched; all #303-program, parallel to the #197 arc). **Dependency/conflict warnings — NEW enforcement live on main, affects every future session**: - **H11**: `git push` from a branch **behind origin/main is blocked** — `git fetch origin main && git rebase origin/main` (do NOT merge main in). Escape `ETV_SKIP_REBASE_CHECK=1`. The lore is flipped to rebase-not-merge. - **format CI job**: a PR that touches a legacy-BOM `.cs` (2530 of them) **must de-BOM/normalize that file** (`dotnet format --include <file>`) or the blocking `format` check fails — the fix-as-you-touch rule (contributing.md §7). `.cs`-free PRs pass trivially. - **H10 merge-gate is proven on non-docs PRs**: a code PR now needs a `Review-verdict: MERGEABLE @ <head-sha>` comment + all `## Done-when` boxes ticked to merge (PR #313 was the first to pass it live). - **H12**: run `scripts/issue-qualification-audit.sh` at session end; label anything it flags.
Author
Owner

Session 2026-07-12 (night) — #303 close-out (spin off the 2 non-hook follow-ups)

Closed: #303 (process-hardening tracker) — the whole H1–H12 hook suite + api-docs/decisions-guard/format CI gates were already live on main; the only remainder was two non-hook follow-ups, now filed as their own issues per the body's stated close condition. No code this session (issue bookkeeping only).

Filed:

  • #314 — out-of-ecosystem black-box security scan (ZAP/nuclei/semgrep vs the running container).
  • #315 — release-path migration-on-prod-copy smoke (restore backup → migrate → boot → health-check).

Triage:

  • #314 → gate-blocker (input to #197), security + review + priority: high. It's a #197 exit criterion (HARD GATE before remote exposure); sequence it LATE in #197 — after auth (#295) and /api/v1 versioning (#286) settle the surface — though a first baseline pass can run against the current surface now.
  • #315 → backlog, ci-cd + enhancement + priority: medium. Release-path data-plane rigor, not a gate. Cross-repo seam: the Komodo pre-deploy backup/hook is server-management territory (companion issue there if the hook changes); the migrate/boot logic + docs are ours.

Reviews: none picked up/filed in timothy/adversarial-reviewer. No code diff → no adversarial code review needed (pure issue bookkeeping).

Arc change: none — arc body untouched. #303 was off-arc (backlog process work); #314 feeds the existing #197 frontier gate, #315 is backlog. Nothing moved in the arc or Done-history.

Dependency/conflict warnings: #314 is a #197 exit criterion — whoever works the #197 gate should treat it as a required late step, not a parallel nicety. #315 needs a server-management companion issue for the pre-deploy hook if implemented as a Komodo step.

## Session 2026-07-12 (night) — #303 close-out (spin off the 2 non-hook follow-ups) **Closed**: #303 (process-hardening tracker) — the whole H1–H12 hook suite + api-docs/decisions-guard/format CI gates were already live on main; the only remainder was two non-hook follow-ups, now filed as their own issues per the body's stated close condition. No code this session (issue bookkeeping only). **Filed**: - #314 — out-of-ecosystem black-box security scan (ZAP/nuclei/semgrep vs the running container). - #315 — release-path migration-on-prod-copy smoke (restore backup → migrate → boot → health-check). **Triage**: - **#314 → gate-blocker (input to #197)**, `security` + `review` + `priority: high`. It's a **#197 exit criterion** (HARD GATE before remote exposure); sequence it LATE in #197 — after auth (#295) and `/api/v1` versioning (#286) settle the surface — though a first baseline pass can run against the current surface now. - **#315 → backlog**, `ci-cd` + `enhancement` + `priority: medium`. Release-path data-plane rigor, not a gate. Cross-repo seam: the Komodo pre-deploy backup/hook is **server-management** territory (companion issue there if the hook changes); the migrate/boot logic + docs are ours. **Reviews**: none picked up/filed in timothy/adversarial-reviewer. No code diff → no adversarial code review needed (pure issue bookkeeping). **Arc change**: none — arc body untouched. #303 was off-arc (backlog process work); #314 feeds the existing #197 frontier gate, #315 is backlog. Nothing moved in the arc or Done-history. **Dependency/conflict warnings**: #314 is a #197 exit criterion — whoever works the #197 gate should treat it as a required late step, not a parallel nicety. #315 needs a server-management companion issue for the pre-deploy hook if implemented as a Komodo step.
Author
Owner

Session 2026-07-12 (night, cont.) — merge-gate double-prompt fix + #314 baseline security scan

Continuation of the #303 close-out comment above; picked up the two spun-off followups + a user-reported gate bug.

Closed: #317 (PR #318, FF main 90c8348e) — merge-consent gate double-prompt. Root cause: the H6/H10 hook's satisfied path did a bare exit 0, which doesn't auto-approve a PreToolUse call (only declines to block), so the raw MCP prompt still fired on top of the conversational ask. Fix: satisfied path (CI green + all Done-when ticked + positive Review-verdict @ head) now emits permissionDecision: allow — true auto-grant. The gate is now live on main and auto-grants — future sessions: do NOT ask conversationally to merge a satisfied PR; a human prompt remains only when the gate ASKS (non-derivable) or for gate-config PRs (.claude//.gitea//.husky/, which stay a passthrough prompt).

Filed:

  • #319 — set security response headers (CSP/nosniff/Permissions-Policy/COEP) on the host; ZAP baseline finding, #197 input.
  • (earlier this session: #314 security scan, #315 migration smoke — the #303 spin-offs.)

Progress (not closed):

  • #314 — ran the first unauthenticated baseline scan vs the test container: ZAP DAST (FAIL 0; only real class = missing security headers → #319), semgrep SAST (0 findings in shipped C#/SPA code; 19 in design-system/ mockups = non-runtime), nuclei deferred (template fetch blocked in the runner env). Kept OPEN, re-scoped to the authenticated deep scan = the actual #197 exit criterion (late, after #295/#286). in-progress removed.

Triage:

  • #319 → gate input to #197 (security+review+priority: medium). Cheap, standard hardening; medium not high (behind auth+LAN proxy today).
  • #317 → was Blocker/High process bug, now CLOSED.
  • #314 remains the #197 exit-criterion gate input (was priority: high); #315 backlog (done earlier).

Reviews: 2 independent cold fork passes on PR #318 (initial diff MERGEABLE-WITH-NITS — caught the docs-exemption silent-auto-merge nit → fixed → fix-commit re-review MERGEABLE @ 90c8348e). No adversarial-reviewer-repo audits picked up/filed.

Arc change: none — arc body untouched. All work was off-arc process/tooling (#303 lineage) plus #197 inputs (#319, #314). #197 frontier unchanged.

Dependency/conflict warnings: #319 + #314-remaining are both #197 exit inputs — whoever works the #197 gate should fold them in (headers fix + authenticated scan) rather than treat as parallel nicety. The merge gate now auto-grants (see Closed above) — the kickoff HARD CONSTRAINT was updated to match.

## Session 2026-07-12 (night, cont.) — merge-gate double-prompt fix + #314 baseline security scan Continuation of the #303 close-out comment above; picked up the two spun-off followups + a user-reported gate bug. **Closed**: #317 (PR #318, FF main `90c8348e`) — merge-consent gate double-prompt. Root cause: the H6/H10 hook's satisfied path did a bare `exit 0`, which doesn't auto-approve a PreToolUse call (only declines to block), so the raw MCP prompt still fired on top of the conversational ask. Fix: satisfied path (CI green + all Done-when ticked + positive Review-verdict @ head) now emits `permissionDecision: allow` — true auto-grant. **The gate is now live on main and auto-grants — future sessions: do NOT ask conversationally to merge a satisfied PR; a human prompt remains only when the gate ASKS (non-derivable) or for gate-config PRs (`.claude/`/`.gitea/`/`.husky/`, which stay a passthrough prompt).** **Filed**: - #319 — set security response headers (CSP/nosniff/Permissions-Policy/COEP) on the host; ZAP baseline finding, #197 input. - (earlier this session: #314 security scan, #315 migration smoke — the #303 spin-offs.) **Progress (not closed)**: - #314 — ran the first **unauthenticated** baseline scan vs the test container: ZAP DAST (FAIL 0; only real class = missing security headers → #319), semgrep SAST (**0 findings in shipped C#/SPA code**; 19 in `design-system/` mockups = non-runtime), nuclei deferred (template fetch blocked in the runner env). Kept OPEN, re-scoped to the **authenticated deep scan** = the actual #197 exit criterion (late, after #295/#286). `in-progress` removed. **Triage**: - **#319 → gate input to #197** (`security`+`review`+`priority: medium`). Cheap, standard hardening; medium not high (behind auth+LAN proxy today). - **#317 → was Blocker/High process bug, now CLOSED.** - #314 remains the #197 exit-criterion gate input (was `priority: high`); #315 backlog (done earlier). **Reviews**: 2 independent cold fork passes on PR #318 (initial diff MERGEABLE-WITH-NITS — caught the docs-exemption silent-auto-merge nit → fixed → fix-commit re-review MERGEABLE @ 90c8348e). No adversarial-reviewer-repo audits picked up/filed. **Arc change**: none — arc body untouched. All work was off-arc process/tooling (#303 lineage) plus #197 *inputs* (#319, #314). #197 frontier unchanged. **Dependency/conflict warnings**: #319 + #314-remaining are both #197 exit inputs — whoever works the #197 gate should fold them in (headers fix + authenticated scan) rather than treat as parallel nicety. The merge gate now auto-grants (see Closed above) — the kickoff HARD CONSTRAINT was updated to match.
Author
Owner

Session 2026-07-12 — #295 PR2 (SPA session-auth cutover) + #301

Closed: #295 (browser-SPA session auth — cookie-only, boot gate, machine-key screen, logout) + #301 (side-effecting-GET POST-ification) — PR #316 (be9cf5f3 → merge). Both are #197 Phase-1 gate sub-items.

Filed: #320 — pre-existing playback segment-wait loop doesn't check notifier.IsFailed (spins if ffmpeg dies after the playlist appears); more relevant now the manifest fetch is a server-held POST.

Triage: #320 → backlog (bug, priority: low — narrow window, self-heals on client cancel).

Reviews: no adversarial-reviewer-repo audits. PR #316 ran the full independent loop: cold fork (MERGEABLE-WITH-NITS) + Codex (BLOCKED, 5 findings) — Codex caught a side-effecting GET the fork missed (GET /api/graphics-elements?refresh=true writes the DB) + lock-race→422-not-409 + a fast-completion screen race + machine-key missing no-store + an AuthSession wire-mismatch. Fable reconciled (ratified Codex's BLOCKED, prescribed the fix plan, rejected 1 fork nit, deferred 1 to #320). Fixes landed → fork fix-commit re-review MERGEABLE. Live-E2E 5/5. (An orchestrator-caught MachineKey {apiKey}-vs-{key} wire bug and stale OpenAPI test cases were fixed during assembly — mocked unit tests missed both; live-E2E confirmed the fix.)

Arc change: body edited — #295 removed from the #197 frontier open-sub-items line; #295+#301 added to Done (history) under "#197 sub-issues shipped".

Dependency/conflict warnings:

  • CI-gate fix shipped in this PR: scripts/update-openapi.sh now runs a full dotnet build before -t:GenerateOpenApiDocuments. The new #303 api-docs gate was failing on any /api-surface PR with deps.json does not exist (it only ran dotnet restore); this PR was the first to exercise it. A parallel session mis-diagnosed the same failure as a runner cache flake (9dc48118 "re-trigger CI" chore) — the real fix is here. Future API PRs will now pass the gate.
  • #197 frontier now: #286 (/api/v1 versioning — sequence LAST) + #289 (MCP hardening, parked on PR#76). #295's ApiAuthorizationFilter/session surface is live on main — coordinate any auth-filter edits (#286's route sweep touches it).
  • Main moved 3× during this session (#303 H10/#311/#315/#317 process work); expect frequent rebases on longer-lived branches. The #317 merge-consent auto-grant + #311 rebase-not-merge conventions are now live.
## Session 2026-07-12 — #295 PR2 (SPA session-auth cutover) + #301 **Closed**: #295 (browser-SPA session auth — cookie-only, boot gate, machine-key screen, logout) + #301 (side-effecting-GET POST-ification) — PR #316 (`be9cf5f3` → merge). Both are #197 Phase-1 gate sub-items. **Filed**: #320 — pre-existing playback segment-wait loop doesn't check `notifier.IsFailed` (spins if ffmpeg dies after the playlist appears); more relevant now the manifest fetch is a server-held POST. **Triage**: #320 → **backlog** (`bug`, `priority: low` — narrow window, self-heals on client cancel). **Reviews**: no adversarial-reviewer-repo audits. PR #316 ran the full independent loop: cold fork (MERGEABLE-WITH-NITS) + **Codex (BLOCKED, 5 findings)** — Codex caught a side-effecting GET the fork missed (`GET /api/graphics-elements?refresh=true` writes the DB) + lock-race→422-not-409 + a fast-completion screen race + machine-key missing `no-store` + an `AuthSession` wire-mismatch. **Fable reconciled** (ratified Codex's BLOCKED, prescribed the fix plan, rejected 1 fork nit, deferred 1 to #320). Fixes landed → **fork fix-commit re-review MERGEABLE**. Live-E2E 5/5. (An orchestrator-caught `MachineKey {apiKey}`-vs-`{key}` wire bug and stale OpenAPI test cases were fixed during assembly — mocked unit tests missed both; live-E2E confirmed the fix.) **Arc change**: body edited — #295 removed from the #197 frontier open-sub-items line; #295+#301 added to Done (history) under "#197 sub-issues shipped". **Dependency/conflict warnings**: - **CI-gate fix shipped in this PR**: `scripts/update-openapi.sh` now runs a full `dotnet build` before `-t:GenerateOpenApiDocuments`. The new #303 api-docs gate was failing on any `/api`-surface PR with `deps.json does not exist` (it only ran `dotnet restore`); this PR was the first to exercise it. A parallel session mis-diagnosed the same failure as a runner cache flake (`9dc48118` "re-trigger CI" chore) — the real fix is here. Future API PRs will now pass the gate. - **#197 frontier now**: #286 (`/api/v1` versioning — sequence LAST) + #289 (MCP hardening, parked on PR#76). #295's `ApiAuthorizationFilter`/session surface is live on main — coordinate any auth-filter edits (#286's route sweep touches it). - Main moved 3× during this session (#303 H10/#311/#315/#317 process work); expect frequent rebases on longer-lived branches. The #317 merge-consent auto-grant + #311 rebase-not-merge conventions are now live.
Author
Owner

Session 2026-07-12 (night, cont.) — #315 migration-smoke + merge-gate creds gotcha

Closed: #315 (PR #321, FF main 9dc48118) — scripts/migration-smoke.sh rehearses a release's EF migrations against a throwaway copy of the latest prod backup (gates on the Done applying database migrations log line, since the migrator is a BackgroundService concurrent with Kestrel; validated live twice vs the 283 MB prod backup). Docs in docs/ci-cd.md + docs/decisions.md.

Filed: server-management#589 — wire the smoke into the Komodo pre-deploy step (cross-repo: we own the script, that repo owns the hook).

Triage: #315 was backlog priority: medium → done. server-management#589 cross-repo companion.

Reviews: 2 independent cold fork passes on PR #321 (initial MERGEABLE-WITH-NITS — dropped a fragile FAIL_RE log-scan + validated --timeout + log image id → fix-commit re-review MERGEABLE). No adversarial-reviewer-repo audits.

Arc change: none — off-arc backlog + #197 inputs.

Dependency/conflict warnings (important for future sessions):

  • The merge-gate auto-grant (#317) needs Gitea read-creds in the session env to fire. Without them the hook fails open to ask → you still get the mechanical prompt (this is why #321's merge still prompted). Now wired on the main dev machine via the gitignored .claude/settings.local.json env (ETV_GITEA_BASICAUTH) — takes effect after a Claude Code restart (env loads at session start). After that, a satisfied merge (CI green + Done-when ticked + Review-verdict @ head) auto-grants with no prompt.
  • api-docs CI gate has a latent build-before-regen bug (job restores but never builds the app, so update-openapi.sh → dotnet-getdocument fails on a missing deps.json) — bites any PR that changes the API surface; another implementer is fixing it on docker-build.yml. Distinct from the one-off actions/checkout cache flake that hit PR #321 (which path-gates api-docs to a skip anyway).
  • server-management#589 tracks the pre-deploy wiring for migration-smoke.
## Session 2026-07-12 (night, cont.) — #315 migration-smoke + merge-gate creds gotcha **Closed**: #315 (PR #321, FF main `9dc48118`) — `scripts/migration-smoke.sh` rehearses a release's EF migrations against a throwaway copy of the latest prod backup (gates on the `Done applying database migrations` log line, since the migrator is a `BackgroundService` concurrent with Kestrel; validated live twice vs the 283 MB prod backup). Docs in `docs/ci-cd.md` + `docs/decisions.md`. **Filed**: server-management#589 — wire the smoke into the Komodo pre-deploy step (cross-repo: we own the script, that repo owns the hook). **Triage**: #315 was backlog `priority: medium` → done. server-management#589 cross-repo companion. **Reviews**: 2 independent cold fork passes on PR #321 (initial MERGEABLE-WITH-NITS — dropped a fragile `FAIL_RE` log-scan + validated `--timeout` + log image id → fix-commit re-review MERGEABLE). No adversarial-reviewer-repo audits. **Arc change**: none — off-arc backlog + #197 inputs. **Dependency/conflict warnings (important for future sessions):** - **The merge-gate auto-grant (#317) needs Gitea read-creds in the session env to fire.** Without them the hook fails open to `ask` → you still get the mechanical prompt (this is why #321's merge still prompted). Now wired on the main dev machine via the gitignored `.claude/settings.local.json` `env` (`ETV_GITEA_BASICAUTH`) — **takes effect after a Claude Code restart** (env loads at session start). After that, a satisfied merge (CI green + Done-when ticked + `Review-verdict @ head`) auto-grants with no prompt. - **`api-docs` CI gate has a latent build-before-regen bug** (job restores but never *builds* the app, so `update-openapi.sh` → `dotnet-getdocument` fails on a missing `deps.json`) — bites any PR that changes the API surface; another implementer is fixing it on `docker-build.yml`. Distinct from the one-off `actions/checkout` cache flake that hit PR #321 (which path-gates api-docs to a skip anyway). - server-management#589 tracks the pre-deploy wiring for migration-smoke.
Author
Owner

Session 2026-07-12 (night) — #319 enforcing CSP + Permissions-Policy

Closed: #319 (PR #323, 6537697f, fast-forward to main). A #197 security-posture gate sub-item (ZAP-baseline hardening from #314).

Filed: none.

Triage: n/a — no new issues. Three low-value follow-on hardenings noted in the #319 close comment + decisions.md but deliberately not filed (self-host Geist font, Scalar-tuned CSP for /docs, report-uri) — surface them only if #197's authenticated deep scan wants them.

Reviews: no adversarial-reviewer-repo audits picked up (5 open there: #20–#24). PR #323 ran one cold zero-context fork review over the diff → MERGEABLE @ 873b3e54; its one LOW (guard hashed source not built artifact) was fixed in 6537697f (test+docs only → independent re-review skipped per the rubric, stated on the PR), NIT (report-uri) deferred. Review-verdict: MERGEABLE @ 6537697f posted.

Arc change: body edited — #319 appended to "#197 sub-issues shipped" in Done (history). #197 frontier line unchanged (open sub-items remain #286 versioning [LAST] + #289 [parked on PR#76]).

Dependency/conflict warnings:

  • #265 (If-Match 412, the remaining #253 piece) is in-progress in a sibling worktree (.claude/worktrees/265-ifmatch-412) owned by another session — do not touch that branch/checkout.
  • SecurityHeadersMiddleware is now the single home for host security headers — future CSP/font/Scalar hardening lands there.
  • Gotcha for future tests: the built SPA (ErsatzTV/wwwroot/app) is gitignored/absent in CI — hash/read the committed web/ source (Vite copies inline scripts verbatim), or prefer-built-else-source as #319's guard test does.
## Session 2026-07-12 (night) — #319 enforcing CSP + Permissions-Policy **Closed**: #319 (PR #323, `6537697f`, fast-forward to main). A #197 security-posture gate sub-item (ZAP-baseline hardening from #314). **Filed**: none. **Triage**: n/a — no new issues. Three low-value follow-on hardenings noted in the #319 close comment + decisions.md but deliberately **not filed** (self-host Geist font, Scalar-tuned CSP for /docs, report-uri) — surface them only if #197's authenticated deep scan wants them. **Reviews**: no adversarial-reviewer-repo audits picked up (5 open there: #20–#24). PR #323 ran one cold zero-context fork review over the diff → **MERGEABLE @ 873b3e54**; its one LOW (guard hashed source not built artifact) was fixed in `6537697f` (test+docs only → independent re-review skipped per the rubric, stated on the PR), NIT (report-uri) deferred. `Review-verdict: MERGEABLE @ 6537697f` posted. **Arc change**: body edited — #319 appended to "#197 sub-issues shipped" in Done (history). #197 frontier line unchanged (open sub-items remain #286 versioning [LAST] + #289 [parked on PR#76]). **Dependency/conflict warnings**: - **#265** (If-Match 412, the remaining #253 piece) is `in-progress` in a **sibling worktree** (`.claude/worktrees/265-ifmatch-412`) owned by another session — do not touch that branch/checkout. - **`SecurityHeadersMiddleware` is now the single home for host security headers** — future CSP/font/Scalar hardening lands there. - Gotcha for future tests: the built SPA (`ErsatzTV/wwwroot/app`) is **gitignored/absent in CI** — hash/read the committed `web/` source (Vite copies inline scripts verbatim), or prefer-built-else-source as #319's guard test does.
Author
Owner

Session 2026-07-12 (night) — #265 If-Match RFC 7232 (last #253 piece)

Closed: #265 (PR #322, c197de42 → main) + #253 (umbrella replace-all concurrency contract — #265 was its last piece; closed as COMPLETE, no separate PR).

Filed: none.

Triage: no new issues this session.

Reviews: no adversarial-reviewer-repo audits picked up. PR #322 ran the full independent loop — cold zero-context fork (MERGEABLE-WITH-NITS, 2 nits applied) + Codex (BLOCKED, 2 real Highs the fork missed): (1) "3,5" — a comma is a valid etagc, so Split(',') broke one valid tag into two → 400 instead of 412; (2) * — string.Trim() strips NBSP, letting a malformed value masquerade as the * force-write. Fixed with a quote-aware position scanner + RFC-OWS-only trim (Trim(' ', '\t')) → Codex fix-commit re-review MERGEABLE. H6/H10 gate auto-granted the merge (CI green + Done-when ticked + Review-verdict @ head).

Arc change: body edited — the #253 priority-pickup line cleared (now the only-standing pickup is COMPLETE); #253 + #265 moved to Done (history).

Dependency/conflict warnings:

  • #197 gate input satisfied: the "#253 replace-all concurrency contract" listed under the #197 frontier inputs is now fully shipped (all 10 aggregates + #269 rotation + #265 RFC eval). Whoever works #197 can treat that input as done.
  • Format-gate gotcha for future macOS sessions: dotnet format --include <files> (and the whole-.sln form) silently no-ops locally here (workspace-load warning + --include path-matching), so it does NOT catch the #311 fix-as-you-touch violations CI enforces. Reliable local repro: run dotnet format whitespace <project> (no --include), git add only your changed files, git checkout -- . to revert the legacy-file noise. Cost me two CI round-trips this session.
  • #286 (/api/v1 versioning) remains explicitly LAST in the #197 arc — sequence after all other in-flight /api work; this PR added none of that.
## Session 2026-07-12 (night) — #265 If-Match RFC 7232 (last #253 piece) **Closed**: #265 (PR #322, `c197de42` → main) + **#253** (umbrella replace-all concurrency contract — #265 was its last piece; closed as COMPLETE, no separate PR). **Filed**: none. **Triage**: no new issues this session. **Reviews**: no adversarial-reviewer-repo audits picked up. PR #322 ran the full independent loop — cold zero-context fork (**MERGEABLE-WITH-NITS**, 2 nits applied) + **Codex (BLOCKED, 2 real Highs the fork missed)**: (1) `"3,5"` — a comma is a valid `etagc`, so `Split(',')` broke one valid tag into two → 400 instead of 412; (2) ` * ` — `string.Trim()` strips NBSP, letting a malformed value masquerade as the `*` force-write. Fixed with a quote-aware position scanner + RFC-OWS-only trim (`Trim(' ', '\t')`) → **Codex fix-commit re-review MERGEABLE**. H6/H10 gate auto-granted the merge (CI green + Done-when ticked + `Review-verdict @ head`). **Arc change**: body edited — the #253 priority-pickup line cleared (now the only-standing pickup is COMPLETE); #253 + #265 moved to Done (history). **Dependency/conflict warnings**: - **#197 gate input satisfied**: the "#253 replace-all concurrency contract" listed under the #197 frontier inputs is now fully shipped (all 10 aggregates + #269 rotation + #265 RFC eval). Whoever works #197 can treat that input as done. - **Format-gate gotcha for future macOS sessions**: `dotnet format --include <files>` (and the whole-`.sln` form) silently **no-ops locally** here (workspace-load warning + `--include` path-matching), so it does NOT catch the #311 fix-as-you-touch violations CI enforces. Reliable local repro: run `dotnet format whitespace <project>` (no `--include`), `git add` only your changed files, `git checkout -- .` to revert the legacy-file noise. Cost me two CI round-trips this session. - **#286** (`/api/v1` versioning) remains explicitly LAST in the #197 arc — sequence after all other in-flight `/api` work; this PR added none of that.
Author
Owner

Session 2026-07-12 (night, cont.) — #238 TopBar primary-action audit

Closed: #238 (PR #324, FF main 51b67dea) — the TopBar primary-action button was inert on every screen but Schedules (and rendered a bare "+" on the ~10 primaryAction: '' routes, since the button rendered unconditionally). Shared usePrimaryAction hook + TopBar renders the button only when primaryAction is non-empty; WIRE the 8 single-create-flow list screens, DROP the rest. High-priority product bug (not an arc/gate item) — picked as a priority-pickup while the #197 frontier's own sub-items were all sequenced-away or claimed.

Filed: none.

Triage: n/a — no new issues.

Reviews: no adversarial-reviewer-repo audits picked up (5 still open there: #20–#24). Two independent cold fork passes on PR #324 — full-PR MERGEABLE-WITH-NITS (2 findings: trakt detail-sub-route no-op + spot-check-only test) → both fixed in 51b67dea → fix-commit re-review MERGEABLE. Review-verdict: MERGEABLE @ 51b67dea posted. Independent review on a pure-SPA leaf was optional per the skip rubric; ran it anyway (10 files + a cross-cutting convention).

Arc change: none — body untouched. #238 was a High-priority bug in the ChicoryTV modularization milestone, not a numbered-arc item; nothing to move in Done(history). #197 frontier unchanged (#286 versioning [LAST] + #289 [parked on PR#76]).

Dependency/conflict warnings:

  • #247 (shell extraction) is now unblocked — #238 was its stated dependency. #238 established the usePrimaryAction seam + the route-label⟺screen-subscription convention (spa-conventions §10); #247 should formalize that coupling into the extracted shell rather than re-derive it.
  • #238 landed on App.tsx (TopBar + full route table) and 7 screen files. The App.tsx modularization epic (#245 Playouts / #246 Dashboard-Guide-Libraries / #247 shell) touches the same file — rebase/coordinate. New file web/src/primaryAction.ts.
  • Rebased onto the #265 (If-Match, PR #322) merge that landed mid-session; decisions.md append-conflict resolved keeping both entries.
## Session 2026-07-12 (night, cont.) — #238 TopBar primary-action audit **Closed**: #238 (PR #324, FF main `51b67dea`) — the TopBar primary-action button was inert on every screen but Schedules (and rendered a bare "+" on the ~10 `primaryAction: ''` routes, since the button rendered unconditionally). Shared `usePrimaryAction` hook + TopBar renders the button only when `primaryAction` is non-empty; WIRE the 8 single-create-flow list screens, DROP the rest. High-priority product bug (not an arc/gate item) — picked as a priority-pickup while the #197 frontier's own sub-items were all sequenced-away or claimed. **Filed**: none. **Triage**: n/a — no new issues. **Reviews**: no adversarial-reviewer-repo audits picked up (5 still open there: #20–#24). Two independent cold fork passes on PR #324 — full-PR **MERGEABLE-WITH-NITS** (2 findings: trakt detail-sub-route no-op + spot-check-only test) → both fixed in `51b67dea` → fix-commit re-review **MERGEABLE**. `Review-verdict: MERGEABLE @ 51b67dea` posted. Independent review on a pure-SPA leaf was optional per the skip rubric; ran it anyway (10 files + a cross-cutting convention). **Arc change**: none — body untouched. #238 was a High-priority bug in the *ChicoryTV modularization* milestone, not a numbered-arc item; nothing to move in Done(history). #197 frontier unchanged (#286 versioning [LAST] + #289 [parked on PR#76]). **Dependency/conflict warnings**: - **#247 (shell extraction) is now unblocked** — #238 was its stated dependency. #238 established the `usePrimaryAction` seam + the route-label⟺screen-subscription convention (spa-conventions §10); #247 should formalize that coupling into the extracted shell rather than re-derive it. - **#238 landed on `App.tsx` (TopBar + full route table) and 7 screen files.** The App.tsx modularization epic (#245 Playouts / #246 Dashboard-Guide-Libraries / #247 shell) touches the same file — rebase/coordinate. New file `web/src/primaryAction.ts`. - Rebased onto the #265 (If-Match, PR #322) merge that landed mid-session; decisions.md append-conflict resolved keeping both entries.
Author
Owner

Session 2026-07-13 — #286 /api/v1 versioning (the #197 C1 BLOCKER)

Closed: #286 (PR #326, 682dceec → main FF). Whole /api surface mounted at /api/v1 (uniform incl. auth/scan/scripted) + ApiVersionRewriteMiddleware legacy rewrite; /api/v1 frozen additive-only; route convention standardized + enforced by ApiRouteVersioningTests. Also closed PR #76 / #289 (v0 read-only MCP) as superseded and re-scoped #58 → the MCP is rebuilt fresh (read+write) against the frozen /api/v1, after #286 (owner-directed).

Filed: none.

Triage: no new issues this session.

Reviews: no adversarial-reviewer-repo audits picked up. PR #326 ran the full independent loop — cold fork (MERGEABLE-WITH-NITS) + Codex (BLOCKED, 2 findings the fork under-weighted): (1) the versioning test silently skipped non-/api routes → a stray absolute non-/api route would escape both the test and ApiAuthorizationFilter; (2) the deprecation Link broke under ETV_BASE_URL. Both fixed in 682dceec; cold fix-commit re-review MERGEABLE. Live-E2E green (versioned routes, legacy rewrite header, /api/v2 passthrough, no auth bypass, SPA boot 0 console errors).

Arc change: body edited — #286 moved to Done (history); #197 frontier open-sub-items line refreshed (#286 done, #289/PR#76 superseded → the remaining #197 gate work is #314 authenticated deep security scan; MCP #58 rebuilds against /api/v1).

Dependency/conflict warnings:

  • The entire /api surface moved to /api/v1. Any in-flight branch with /api/* route attributes or SPA request literals will conflict on rebase — regenerate OpenAPI artifacts (update-openapi.sh + npm run generate:api) after rebasing rather than hand-resolving, and the new ApiRouteVersioningTests will flag any unversioned route. Legacy /api/* still works via the rewrite shim (Deprecation header) during the 2-release sunset.
  • #58 (MCP rebuild) is now unblocked — builds read+write tools against the frozen /api/v1; still gated on the #197 security MUST-FIX before go-live.
  • #314 is the remaining #197 exit criterion (authenticated deep scan against the settled surface).
## Session 2026-07-13 — #286 `/api/v1` versioning (the #197 C1 BLOCKER) **Closed**: #286 (PR #326, `682dceec` → main FF). Whole `/api` surface mounted at `/api/v1` (uniform incl. auth/scan/scripted) + `ApiVersionRewriteMiddleware` legacy rewrite; `/api/v1` frozen additive-only; route convention standardized + enforced by `ApiRouteVersioningTests`. Also closed **PR #76 / #289 (v0 read-only MCP)** as **superseded** and **re-scoped #58** → the MCP is rebuilt fresh (read+write) against the frozen `/api/v1`, after #286 (owner-directed). **Filed**: none. **Triage**: no new issues this session. **Reviews**: no adversarial-reviewer-repo audits picked up. PR #326 ran the full independent loop — cold fork (MERGEABLE-WITH-NITS) + **Codex (BLOCKED, 2 findings the fork under-weighted)**: (1) the versioning test silently skipped non-`/api` routes → a stray absolute non-`/api` route would escape both the test and `ApiAuthorizationFilter`; (2) the deprecation `Link` broke under `ETV_BASE_URL`. Both fixed in `682dceec`; cold fix-commit re-review MERGEABLE. Live-E2E green (versioned routes, legacy rewrite header, `/api/v2` passthrough, no auth bypass, SPA boot 0 console errors). **Arc change**: body edited — #286 moved to Done (history); #197 frontier open-sub-items line refreshed (#286 done, #289/PR#76 superseded → the remaining #197 gate work is **#314** authenticated deep security scan; MCP #58 rebuilds against `/api/v1`). **Dependency/conflict warnings**: - **The entire `/api` surface moved to `/api/v1`.** Any in-flight branch with `/api/*` route attributes or SPA request literals will conflict on rebase — regenerate OpenAPI artifacts (`update-openapi.sh` + `npm run generate:api`) after rebasing rather than hand-resolving, and the new `ApiRouteVersioningTests` will flag any unversioned route. Legacy `/api/*` still works via the rewrite shim (Deprecation header) during the 2-release sunset. - **#58 (MCP rebuild) is now unblocked** — builds read+write tools against the frozen `/api/v1`; still gated on the #197 security MUST-FIX before go-live. - **#314** is the remaining #197 exit criterion (authenticated deep scan against the settled surface).
Author
Owner

Session 2026-07-13 — #172 scheduling API hardening (#197 gate-cluster)

Closed: #172 (PR #325, FF main → 2e4e07a2). A review+api+bug gate-cluster item. Three genuinely-live findings fixed: null/empty-name → 500 across 10 create/replace handlers (Block/Template/DecoTemplate/Deco + UpdateFFmpegProfile/CreatePlaylist), exact-duplicate template items bypassing overlap validation, and unreachable-404 ProducesResponseType on all 4 create-group actions. The other 6 findings + tail were verified deliberate-documented / moot / already-fixed against live main (not blindly re-fixed).

Filed: #327 — playlist rename (ReplacePlaylistItems) does no name validation (empty / >50-char accepted); pre-existing create-vs-update asymmetry, not the NRE class.

Triage: #327 → backlog (api+bug+priority: low — display nuisance, no crash/security). Qualified (H12 audit clean).

Reviews: no adversarial-reviewer-repo audits picked up (5 still open there: #20–#24). One cold zero-context fork review over PR #325 — caught a real CI-red blocker (the 404 trim left stale ApiErrorResponseMetadataTests assertions) my own stale-dll local run had masked; fix-commit re-reviewed → MERGEABLE. Independent pass was mandatory here (API write-path handler validation).

Arc change: none — the #197 frontier body is already current (down to #314 as the exit criterion; thanks parallel session). #172 was a gate-cluster item, not a numbered-arc item.

Dependency/conflict warnings:

  • #286 (/api/v1 sweep) merged mid-session — I rebased #172 onto it (clean auto-merge on the 4 shared controllers + keep-both decisions.md; regenerated v1.json). Any other in-flight branch that predates 682dceec will hit the same /api/*→/api/v1/* route churn on rebase — expect the whole controller/spec/SPA-literal surface to have moved.
  • The #197 frontier is now just #314 (authenticated out-of-ecosystem deep scan) + go-live; #286/#289 both landed/superseded.
  • Stale-dll false-pass reminder (bit me this session): dotnet test reported exit 0 on a stale build after git clean nuked bin/obj — the metadata-test regression only surfaced on a clean rebuild + the cold review's fresh run. Grep the build for Build succeeded before trusting a green.
## Session 2026-07-13 — #172 scheduling API hardening (#197 gate-cluster) **Closed**: #172 (PR #325, FF `main` → `2e4e07a2`). A `review`+`api`+`bug` gate-cluster item. Three genuinely-live findings fixed: null/empty-name → 500 across **10 create/replace handlers** (Block/Template/DecoTemplate/Deco + `UpdateFFmpegProfile`/`CreatePlaylist`), exact-duplicate template items bypassing overlap validation, and unreachable-404 `ProducesResponseType` on all 4 create-group actions. The other 6 findings + tail were verified deliberate-documented / moot / already-fixed against live `main` (not blindly re-fixed). **Filed**: #327 — playlist rename (`ReplacePlaylistItems`) does no name validation (empty / >50-char accepted); pre-existing create-vs-update asymmetry, not the NRE class. **Triage**: #327 → **backlog** (`api`+`bug`+`priority: low` — display nuisance, no crash/security). Qualified (H12 audit clean). **Reviews**: no adversarial-reviewer-repo audits picked up (5 still open there: #20–#24). One cold zero-context fork review over PR #325 — caught a real **CI-red blocker** (the 404 trim left stale `ApiErrorResponseMetadataTests` assertions) my own stale-dll local run had masked; fix-commit re-reviewed → MERGEABLE. Independent pass was mandatory here (API write-path handler validation). **Arc change**: none — the #197 frontier body is already current (down to #314 as the exit criterion; thanks parallel session). #172 was a gate-cluster item, not a numbered-arc item. **Dependency/conflict warnings**: - **#286 (`/api/v1` sweep) merged mid-session** — I rebased #172 onto it (clean auto-merge on the 4 shared controllers + keep-both `decisions.md`; regenerated `v1.json`). Any other in-flight branch that predates `682dceec` will hit the same `/api/*`→`/api/v1/*` route churn on rebase — expect the whole controller/spec/SPA-literal surface to have moved. - The **#197 frontier is now just #314** (authenticated out-of-ecosystem deep scan) + go-live; #286/#289 both landed/superseded. - **Stale-dll false-pass reminder** (bit me this session): `dotnet test` reported exit 0 on a stale build after `git clean` nuked `bin/obj` — the metadata-test regression only surfaced on a clean rebuild + the cold review's fresh run. Grep the build for `Build succeeded` before trusting a green.
Author
Owner

Session 2026-07-13 (cont.) — #314 authenticated security scan (#197 exit criterion)

Closed: #314 (PR #331, a042623a → main FF). Authenticated OWASP ZAP api-scan (imports /openapi/v1.json, injects X-Api-Key) vs a throwaway :latest container → 0 High/Med/Crit across all 160 /api/v1 ops; semgrep 0. Shipped scripts/security-scan.sh + docs/ci-cd.md. Fixed a go-live-blocking dep-vuln en route (Microsoft.OpenApi 2.0.0→2.7.5, GHSA-v5pm-xwqc-g5wc High).

Filed:

  • #330 — Cross-Origin-Resource-Policy header missing (ZAP Low, extends #319).

Triage:

  • #330 → backlog (security+review+priority: low) — defense-in-depth header hygiene; not go-live-blocking (the API surface passed the authenticated active scan clean; behind auth + LAN proxy today).
  • Microsoft.OpenApi vuln → fixed in-PR (go-live-blocker, not a separate open issue).

Reviews: no adversarial-reviewer-repo audits picked up. PR #331 ran cold review (MERGEABLE-WITH-NITS — independently verified the dep-pin resolves + no key-injection + can't mis-target prod) → 2 nits fixed → fix-commit re-review MERGEABLE @ a042623a.

Arc change: body edited — #314 moved to Done (history); #197 frontier line refreshed: both gate sub-items (#286 + #314) are now closed, the authenticated scan is clean, and no open code sub-item remains in the #197 cluster — the remaining step is the owner go-live go/no-go on #197, then arc item 6 (Go-live). #330 slotted to backlog.

Dependency/conflict warnings:

  • The Microsoft.OpenApi 2.7.5 pin is on main (Directory.Packages.props + ErsatzTV.csproj); the weekly dependency-scan job (which was red on this High vuln) will go green.
  • scripts/security-scan.sh runs on the docker host (bumblebee), not the Mac — re-run each release; it's a manual host gate, not a per-PR CI job.
  • #197 is now an owner decision, not a code pickup — next session should NOT re-pick #286/#314 (closed) or treat #197 as code work; surface the go-live go/no-go to the user.
## Session 2026-07-13 (cont.) — #314 authenticated security scan (#197 exit criterion) **Closed**: #314 (PR #331, `a042623a` → main FF). Authenticated OWASP ZAP `api-scan` (imports `/openapi/v1.json`, injects `X-Api-Key`) vs a throwaway `:latest` container → **0 High/Med/Crit** across all 160 `/api/v1` ops; semgrep 0. Shipped `scripts/security-scan.sh` + `docs/ci-cd.md`. Fixed a go-live-blocking dep-vuln en route (**Microsoft.OpenApi 2.0.0→2.7.5**, GHSA-v5pm-xwqc-g5wc High). **Filed**: - #330 — Cross-Origin-Resource-Policy header missing (ZAP Low, extends #319). **Triage**: - **#330 → backlog** (`security`+`review`+`priority: low`) — defense-in-depth header hygiene; not go-live-blocking (the API surface passed the authenticated active scan clean; behind auth + LAN proxy today). - Microsoft.OpenApi vuln → **fixed in-PR** (go-live-blocker, not a separate open issue). **Reviews**: no adversarial-reviewer-repo audits picked up. PR #331 ran cold review (MERGEABLE-WITH-NITS — independently verified the dep-pin resolves + no key-injection + can't mis-target prod) → 2 nits fixed → fix-commit re-review MERGEABLE @ `a042623a`. **Arc change**: body edited — #314 moved to Done (history); **#197 frontier line refreshed**: both gate sub-items (#286 + #314) are now closed, the authenticated scan is clean, and **no open code sub-item remains in the #197 cluster** — the remaining step is the **owner go-live go/no-go on #197**, then arc item 6 (Go-live). #330 slotted to backlog. **Dependency/conflict warnings**: - The **Microsoft.OpenApi 2.7.5 pin** is on main (`Directory.Packages.props` + `ErsatzTV.csproj`); the weekly `dependency-scan` job (which was red on this High vuln) will go green. - `scripts/security-scan.sh` runs on the **docker host** (bumblebee), not the Mac — re-run each release; it's a manual host gate, not a per-PR CI job. - **#197 is now an owner decision**, not a code pickup — next session should NOT re-pick #286/#314 (closed) or treat #197 as code work; surface the go-live go/no-go to the user.
Author
Owner

Session 2026-07-13 — #197 final code/security gate

Closed: #197 (inspection-only; owner accepted GO-WITH-NONBLOCKING-FOLLOWUPS for the application/code gate; no PR)
Filed: #334 — keep local-library filesystem paths authenticated when anonymous catalog reads are enabled
Triage: #334 → backlog, Medium/security (default-secure posture means non-gating). H12 also qualified PRs #328/#329 as Low dependency updates and PR #191 as Medium CI/CD.
Reviews: Two independent cold read-only passes (source-control audit + final go-live judge) agreed on the code-gate verdict and the #334 classification; no adversarial-reviewer audit was claimed.
Arc change: body edited — moved #197 to Done history; advanced the frontier to Arc 6; removed closed #66 from backlog; added #293/#334; corrected the now-satisfied #238 modularization dependency; clarified the current floating-:prod vs PR #191 pinned-version deployment-policy decision.
Dependency/conflict warnings: Arc 6 still requires an exact-release dependency/security scan, verified backup path, trusted-proxy configuration, and local-admin bootstrap before exposure. Resolve PR #191's pinned-tag/backup design against the current floating :prod path. #333 remains separate real-server integration qualification and is recommended before release.

## Session 2026-07-13 — #197 final code/security gate **Closed**: #197 (inspection-only; owner accepted GO-WITH-NONBLOCKING-FOLLOWUPS for the application/code gate; no PR) **Filed**: #334 — keep local-library filesystem paths authenticated when anonymous catalog reads are enabled **Triage**: #334 → backlog, Medium/security (default-secure posture means non-gating). H12 also qualified PRs #328/#329 as Low dependency updates and PR #191 as Medium CI/CD. **Reviews**: Two independent cold read-only passes (source-control audit + final go-live judge) agreed on the code-gate verdict and the #334 classification; no adversarial-reviewer audit was claimed. **Arc change**: body edited — moved #197 to Done history; advanced the frontier to Arc 6; removed closed #66 from backlog; added #293/#334; corrected the now-satisfied #238 modularization dependency; clarified the current floating-`:prod` vs PR #191 pinned-version deployment-policy decision. **Dependency/conflict warnings**: Arc 6 still requires an exact-release dependency/security scan, verified backup path, trusted-proxy configuration, and local-admin bootstrap before exposure. Resolve PR #191's pinned-tag/backup design against the current floating `:prod` path. #333 remains separate real-server integration qualification and is recommended before release.
Author
Owner

Session 2026-07-13 — ChicoryTV v26.8.0 go-live

Closed: #335 (release-boundary PR #337; production configuration server-management#590 / PR #591); #336 (PR #337); milestone “ChicoryTV go-live”.
Filed: #336 — release-boundary docs slice (closed); #338 — distinguish ZAP warning exit 2 from security FAIL.
Triage: #336 → gate sub-issue, High, completed; server-management#590 → Arc 6 gate-blocker, High, completed; #338 → backlog, Low; #333 → backlog/non-gating integration validation, Medium, explicitly deferred to an isolated credentialed window.
Reviews: no adversarial-reviewer audit picked up; PR #337 used its docs-only exemption, and server-management PR #591 recorded a bounded declarative-config/runbook self-review exemption followed by exact-image and live production verification.
Arc change: body edited after re-reading live state — removed the closed Arc 6 frontier, added #335 to Done history, declared all six numbered arcs complete, moved modularization to post-go-live wording, and listed #333/#338 in backlog.
Dependency/conflict warnings: #333 needs a refreshed isolated current test instance plus explicit Jellyfin/Emby and Plex credentials; #338 is independent release-tooling cleanup.

## Session 2026-07-13 — ChicoryTV v26.8.0 go-live **Closed**: #335 (release-boundary PR #337; production configuration server-management#590 / PR #591); #336 (PR #337); milestone “ChicoryTV go-live”. **Filed**: #336 — release-boundary docs slice (closed); #338 — distinguish ZAP warning exit 2 from security FAIL. **Triage**: #336 → gate sub-issue, High, completed; server-management#590 → Arc 6 gate-blocker, High, completed; #338 → backlog, Low; #333 → backlog/non-gating integration validation, Medium, explicitly deferred to an isolated credentialed window. **Reviews**: no adversarial-reviewer audit picked up; PR #337 used its docs-only exemption, and server-management PR #591 recorded a bounded declarative-config/runbook self-review exemption followed by exact-image and live production verification. **Arc change**: body edited after re-reading live state — removed the closed Arc 6 frontier, added #335 to Done history, declared all six numbered arcs complete, moved modularization to post-go-live wording, and listed #333/#338 in backlog. **Dependency/conflict warnings**: #333 needs a refreshed isolated current test instance plus explicit Jellyfin/Emby and Plex credentials; #338 is independent release-tooling cleanup.
Author
Owner

Goal complete

All six numbered rewrite arcs are closed. ChicoryTV v26.8.0 is running in production with the exact-image security gate, fail-closed backup/migration rehearsal, trusted-network restriction, and seed-free local-admin login verified.

This tracker can close; the explicitly listed non-gating items remain independent backlog issues and should be selected from live Gitea state using the lower-effort selector rule recorded in the kickoff handoff.

## Goal complete All six numbered rewrite arcs are closed. ChicoryTV v26.8.0 is running in production with the exact-image security gate, fail-closed backup/migration rehearsal, trusted-network restriction, and seed-free local-admin login verified. This tracker can close; the explicitly listed non-gating items remain independent backlog issues and should be selected from live Gitea state using the lower-effort selector rule recorded in the kickoff handoff.
Author
Owner

Session 2026-07-13 — #330 CORP + Kodi artwork diagnosis

Closed: #330 (PR #341, fast-forward 2609b4ce)
Filed: #340 — optional advertised IPTV base URL for stable M3U/XMLTV absolute URLs
Triage: #340 → backlog / Medium (Kodi programme artwork is broken, but streams/guide work and a resolvable-source operational workaround exists). Existing timothy/server-management#501 was updated with the Dispatcharr EPG-origin cache bug, Docker-only programme URLs, and stale logo-ID evidence; no external GitHub issue was created. H12 script lacked raw creds and failed open; equivalent authenticated MCP audit found 0 open ErsatzTV issues missing a priority: label.
Reviews: PR #341 received a cold independent security review with no findings; Review-verdict: MERGEABLE @ 2609b4ce593eeaf7203fb50c78feec030638984f. Full CI green.
Arc change: none — the numbered rewrite arc remains complete and #237 remains closed; body untouched.
Dependency/conflict warnings: #340 is independent backlog code hardening; its immediate workaround/config evidence and the separate Dispatcharr channel-logo failures live on server-management#501. No numbered-arc dependency.

## Session 2026-07-13 — #330 CORP + Kodi artwork diagnosis **Closed**: #330 (PR #341, fast-forward `2609b4ce`) **Filed**: #340 — optional advertised IPTV base URL for stable M3U/XMLTV absolute URLs **Triage**: #340 → **backlog / Medium** (Kodi programme artwork is broken, but streams/guide work and a resolvable-source operational workaround exists). Existing `timothy/server-management#501` was updated with the Dispatcharr EPG-origin cache bug, Docker-only programme URLs, and stale logo-ID evidence; no external GitHub issue was created. H12 script lacked raw creds and failed open; equivalent authenticated MCP audit found 0 open ErsatzTV issues missing a `priority:` label. **Reviews**: PR #341 received a cold independent security review with no findings; `Review-verdict: MERGEABLE @ 2609b4ce593eeaf7203fb50c78feec030638984f`. Full CI green. **Arc change**: none — the numbered rewrite arc remains complete and #237 remains closed; body untouched. **Dependency/conflict warnings**: #340 is independent backlog code hardening; its immediate workaround/config evidence and the separate Dispatcharr channel-logo failures live on `server-management#501`. No numbered-arc dependency.
Author
Owner

Session 2026-07-13 — #333 real-server media-source validation

Closed: none; documentation PR #346 merged, while #333 remains open for its Plex server/library criterion
Filed: #344 — successful remote-connection save falsely triggers the dirty guard; #345 — authorized Plex accounts with zero servers get contradictory status and no sign-out
Triage: #344 → backlog, High; #345 → backlog, High
Reviews: none; #333 retained its documented test-results-only adversarial-review exemption
Arc change: none
Dependency/conflict warnings: #333 needs an owned, reachable Plex Media Server attached to the test account before its remaining discovery/library checks can run; #344 and #345 both touch media-source SPA flows, so coordinate concurrent edits

## Session 2026-07-13 — #333 real-server media-source validation **Closed**: none; documentation PR #346 merged, while #333 remains open for its Plex server/library criterion **Filed**: #344 — successful remote-connection save falsely triggers the dirty guard; #345 — authorized Plex accounts with zero servers get contradictory status and no sign-out **Triage**: #344 → backlog, High; #345 → backlog, High **Reviews**: none; #333 retained its documented test-results-only adversarial-review exemption **Arc change**: none **Dependency/conflict warnings**: #333 needs an owned, reachable Plex Media Server attached to the test account before its remaining discovery/library checks can run; #344 and #345 both touch media-source SPA flows, so coordinate concurrent edits
Author
Owner

Session 2026-07-14 — picker correction + remote dirty guard

Closed: #344 (PR #349); #347 (PR #348)
Filed: #347 — queue selector skipped open milestone work and repicked completed reviewer audits.
Triage: #347 → backlog/process bug, High; fixed and closed in-session. H12 credentialed script path was unavailable, so the equivalent authenticated MCP audit enumerated all 43 open issues: none is missing a priority: label.
Reviews: #344 cold exact-head adversarial review passed after a follow-up commit fixed in-flight draft ownership and added App-level navigation coverage; no reviewer-repo audit was picked up.
Arc change: #237 body edited: reviewer comment-claim lifetime/deliverable exclusion made explicit; candidate discovery now enumerates every open milestone/open issue plus open review issues; milestone/review tier outranks unmilestoned backlog; epic/tie ordering made deterministic. Closed numbered arcs no longer imply a terminal no-pick state.
Dependency/conflict warnings: #245/#246/#247 share App.tsx; coordinate claims and extractions to avoid overlapping changes.

## Session 2026-07-14 — picker correction + remote dirty guard **Closed**: #344 (PR #349); #347 (PR #348) **Filed**: #347 — queue selector skipped open milestone work and repicked completed reviewer audits. **Triage**: #347 → backlog/process bug, High; fixed and closed in-session. H12 credentialed script path was unavailable, so the equivalent authenticated MCP audit enumerated all 43 open issues: none is missing a `priority:` label. **Reviews**: #344 cold exact-head adversarial review passed after a follow-up commit fixed in-flight draft ownership and added App-level navigation coverage; no reviewer-repo audit was picked up. **Arc change**: #237 body edited: reviewer comment-claim lifetime/deliverable exclusion made explicit; candidate discovery now enumerates every open milestone/open issue plus open review issues; milestone/review tier outranks unmilestoned backlog; epic/tie ordering made deterministic. Closed numbered arcs no longer imply a terminal no-pick state. **Dependency/conflict warnings**: #245/#246/#247 share `App.tsx`; coordinate claims and extractions to avoid overlapping changes.
Author
Owner

Session 2026-07-14 — #245 Playouts screen extraction

Closed: #245 via PR #351 (fast-forward at 7cbd871ab7601e8eb8896203d04a86fe67690404).

Filed: no additional product issue. The selector-routing correction is being carried separately in ErsatzTV PR #352 and server-management PR #597.

Triage: the H12 script had no local Gitea credentials, so the authenticated MCP-equivalent audit was run instead. It found #350 missing priority qualification; #350 is now labeled bug / priority: medium. No remaining real issue in the audited open set lacks a priority label.

Reviews: two blocking review passes closed the direct-screen coverage gaps. The final cold review reported no findings and Review-verdict: MERGEABLE @ 7cbd871ab7601e8eb8896203d04a86fe67690404. Required CI passed before merge.

Arc change: none. The tracker body remains unchanged; this is still post-go-live ChicoryTV modularization work.

Dependency/conflict warnings: #246 and #247 both touch the residual App.tsx. #246 must account for Libraries already having been extracted under #202 and should scope the remaining work accordingly. PR #352 is docs-only and does not overlap the application extraction.

## Session 2026-07-14 — #245 Playouts screen extraction **Closed**: #245 via PR #351 (fast-forward at `7cbd871ab7601e8eb8896203d04a86fe67690404`). **Filed**: no additional product issue. The selector-routing correction is being carried separately in ErsatzTV PR #352 and server-management PR #597. **Triage**: the H12 script had no local Gitea credentials, so the authenticated MCP-equivalent audit was run instead. It found #350 missing priority qualification; #350 is now labeled `bug` / `priority: medium`. No remaining real issue in the audited open set lacks a priority label. **Reviews**: two blocking review passes closed the direct-screen coverage gaps. The final cold review reported no findings and `Review-verdict: MERGEABLE @ 7cbd871ab7601e8eb8896203d04a86fe67690404`. Required CI passed before merge. **Arc change**: none. The tracker body remains unchanged; this is still post-go-live ChicoryTV modularization work. **Dependency/conflict warnings**: #246 and #247 both touch the residual `App.tsx`. #246 must account for Libraries already having been extracted under #202 and should scope the remaining work accordingly. PR #352 is docs-only and does not overlap the application extraction.
Author
Owner

Session 2026-07-14 — backlog capture + cheap-worker routing

Closed: #355 (PR #356)
Filed:

  • #354 — ChicoryTV subtitles progressively run ahead of dialogue during longer playback
  • #355 — record the tested Codex cheap-worker launch configuration
  • #357 — Jellyfin/Kodi player-owned channel-playback architecture PoC
    Triage:
  • #354 → backlog, medium
  • #355 → backlog, low; closed via PR #356
  • #357 → backlog, high; feasibility PoC before any product-direction pivot
    Reviews: none in adversarial-reviewer; PR #356 received a fresh cold MERGEABLE review at its final head
    Arc change: none
    Dependency/conflict warnings: #357 must preserve/reuse the API, MCP, channel-wizard, programming, M3U/XMLTV, EPG, and Kodi work; coordinate its PoC/reuse matrix with #178, #176, #58, and #62 before opening implementation slices.
## Session 2026-07-14 — backlog capture + cheap-worker routing **Closed**: #355 (PR #356) **Filed**: - #354 — ChicoryTV subtitles progressively run ahead of dialogue during longer playback - #355 — record the tested Codex cheap-worker launch configuration - #357 — Jellyfin/Kodi player-owned channel-playback architecture PoC **Triage**: - #354 → backlog, medium - #355 → backlog, low; closed via PR #356 - #357 → backlog, high; feasibility PoC before any product-direction pivot **Reviews**: none in adversarial-reviewer; PR #356 received a fresh cold MERGEABLE review at its final head **Arc change**: none **Dependency/conflict warnings**: #357 must preserve/reuse the API, MCP, channel-wizard, programming, M3U/XMLTV, EPG, and Kodi work; coordinate its PoC/reuse matrix with #178, #176, #58, and #62 before opening implementation slices.
Author
Owner

Session 2026-07-15 — #357 Jellyfin player-owned playback feasibility

Closed: none — #357 remains open under its original gate; PR #358 is open, reviewed, and green
Filed: none
Triage: no new issues. The H12 script had no raw credentials and failed open; the authenticated MCP-equivalent audit checked all 46 open ErsatzTV issues and found none missing a priority: label
Reviews: exact-head cold review reported no findings at 97397906e3771baa68355248b3a6bf8f4d4733d0; Review-verdict: MERGEABLE is posted on PR #358, explicitly limited to the feasibility artifact. Actions run #641 passed
Arc change: none — tracker body remains unchanged and closed
Dependency/conflict warnings: #357's server-plugin-only path is a documented no-go on stock Jellyfin 12.0-rc2; retain M3U/XMLTV and do not start Kodi-specific implementation until an offset/session/transition seam is proven. Coordinate any resumed work with #178, #176, #58, and #62

## Session 2026-07-15 — #357 Jellyfin player-owned playback feasibility **Closed**: none — #357 remains open under its original gate; PR #358 is open, reviewed, and green **Filed**: none **Triage**: no new issues. The H12 script had no raw credentials and failed open; the authenticated MCP-equivalent audit checked all 46 open ErsatzTV issues and found none missing a `priority:` label **Reviews**: exact-head cold review reported no findings at `97397906e3771baa68355248b3a6bf8f4d4733d0`; `Review-verdict: MERGEABLE` is posted on PR #358, explicitly limited to the feasibility artifact. Actions run #641 passed **Arc change**: none — tracker body remains unchanged and closed **Dependency/conflict warnings**: #357's server-plugin-only path is a documented no-go on stock Jellyfin 12.0-rc2; retain M3U/XMLTV and do not start Kodi-specific implementation until an offset/session/transition seam is proven. Coordinate any resumed work with #178, #176, #58, and #62
Author
Owner

Session 2026-07-15 — #246 Dashboard + Guide extraction

Closed

  • #246 — Dashboard and Guide extractions completed through PR #360 and PR #359.

Filed

  • none.

Triage

  • none required; no issues were filed this session.
  • H12 script took its documented no-credentials fail-open path. An exact-repository MCP audit found no open issues missing a priority: label.

Reviews

  • PR #360: cold adversarial review passed on bceffef856bb43efb7d638511ea1e066afa3075c; CI green.
  • PR #359: initial review passed, then a fresh post-rebase conflict/lifecycle review passed on final head 3876809a83088b5c6f3393cb56d00cd296cf5efe; CI green.

Arc change

  • #246 Batch A is complete. Dashboard and Guide are screen-owned; Libraries remained completed through #202. Parent #243 remains the live shell/routing modularization arc.

Dependency/conflict warnings

  • none. Dashboard landed first; Guide was rebased onto current main, reverified, and rereviewed before landing.
## Session 2026-07-15 — #246 Dashboard + Guide extraction **Closed** - #246 — Dashboard and Guide extractions completed through PR #360 and PR #359. **Filed** - none. **Triage** - none required; no issues were filed this session. - H12 script took its documented no-credentials fail-open path. An exact-repository MCP audit found no open issues missing a `priority:` label. **Reviews** - PR #360: cold adversarial review passed on `bceffef856bb43efb7d638511ea1e066afa3075c`; CI green. - PR #359: initial review passed, then a fresh post-rebase conflict/lifecycle review passed on final head `3876809a83088b5c6f3393cb56d00cd296cf5efe`; CI green. **Arc change** - #246 Batch A is complete. Dashboard and Guide are screen-owned; Libraries remained completed through #202. Parent #243 remains the live shell/routing modularization arc. **Dependency/conflict warnings** - none. Dashboard landed first; Guide was rebased onto current `main`, reverified, and rereviewed before landing.
Author
Owner

Session 2026-07-16 — ChicoryTV modularization final phase

Closed: #247 (PR #361, merge 82467f188bcf09834e21b8fd5160c94e5d859ede); parent epic #243; milestone 8 (zero open issues).

Filed: none.

Triage: no new issues. The H12 shell audit failed open because Gitea credentials are not exported to the shell; the equivalent bounded MCP audit checked both result pages and found no open non-PR issue missing a priority: label (the apparent #358/#362 hits were pull requests, not issues).

Reviews: PR #361 received a cold-context adversarial review. Its App-level Builder/Settings composition-smoke finding was fixed in follow-up 947c8aac; exact-head re-review verdict: MERGEABLE. No adversarial-reviewer audit was picked up or filed.

Arc change: no tracker-body edit. Live state now records #247 and parent #243 closed and the ChicoryTV modularization milestone complete, superseding earlier session prose that still described this epic as open.

Dependency/conflict warnings: none from this work. #345 (priority: high) and #340 (priority: medium) remain open, unclaimed, unmilestoned backlog items; live labels, not earlier recommendation prose, determine their ordering.

## Session 2026-07-16 — ChicoryTV modularization final phase **Closed**: #247 (PR #361, merge `82467f188bcf09834e21b8fd5160c94e5d859ede`); parent epic #243; milestone 8 (zero open issues). **Filed**: none. **Triage**: no new issues. The H12 shell audit failed open because Gitea credentials are not exported to the shell; the equivalent bounded MCP audit checked both result pages and found no open non-PR issue missing a `priority:` label (the apparent #358/#362 hits were pull requests, not issues). **Reviews**: PR #361 received a cold-context adversarial review. Its App-level Builder/Settings composition-smoke finding was fixed in follow-up `947c8aac`; exact-head re-review verdict: MERGEABLE. No adversarial-reviewer audit was picked up or filed. **Arc change**: no tracker-body edit. Live state now records #247 and parent #243 closed and the ChicoryTV modularization milestone complete, superseding earlier session prose that still described this epic as open. **Dependency/conflict warnings**: none from this work. #345 (priority: high) and #340 (priority: medium) remain open, unclaimed, unmilestoned backlog items; live labels, not earlier recommendation prose, determine their ordering.
Author
Owner

Session 2026-07-16 — functional-E2E CI harness (#299)

Closed: #299 (PR #362, merged to main 0d780307 via fast-forward). Note: the ChicoryTV SPA/API arc was already complete when this session started — no open arc frontier; picked #299 as the highest-ranked unmilestoned backlog (priority: high) after the selector confirmed #243's only open child #247 was in-progress (it has since merged as #247).
Filed: #363 — follow-up: add the deferred functional-E2E flows (racy 409 re-trigger, external-collections/playout-build lock 409s, Playwright UI-interactive) that need scanner+seeded media or a browser to be deterministic.
Triage: #363 → backlog, priority: medium (ci-cd, enhancement) — H12-qualified. Also surfaced a non-#299 observation for a future look: POST /api/v1/channels with {} returns 500 not a 400 (possible validation gap) — not yet filed.
Reviews: one cold-context, same-model (Claude) review-only pass on the PR diff → MERGEABLE-WITH-NITS (no Critical/High/Medium; verified no vacuous-pass paths in the harness). Same-model cold review judged sufficient for a CI+shell+docs leaf with no concurrency/auth-handler/migration C# and <150 changed C# lines; cross-model/frontier reserved for those risk classes. No reviewer-repo audits picked up.
Arc change: none (arc already fully in "Done history"; this was backlog work).
Dependency/conflict warnings: none for the harness itself. #363 (its follow-up) will need the docs/e2e-local.md seeding recipe wired into the CI job + ffmpeg/scanner present; coordinate with anyone extending the functional-e2e job. Rebased cleanly over the #247 merge (only docs/decisions.md append-conflict).

## Session 2026-07-16 — functional-E2E CI harness (#299) **Closed**: #299 (PR #362, merged to main `0d780307` via fast-forward). Note: the ChicoryTV SPA/API arc was already complete when this session started — no open arc frontier; picked #299 as the highest-ranked unmilestoned backlog (`priority: high`) after the selector confirmed #243's only open child #247 was `in-progress` (it has since merged as #247). **Filed**: #363 — follow-up: add the deferred functional-E2E flows (racy 409 re-trigger, external-collections/playout-build lock 409s, Playwright UI-interactive) that need scanner+seeded media or a browser to be deterministic. **Triage**: #363 → backlog, `priority: medium` (`ci-cd`, `enhancement`) — H12-qualified. Also surfaced a non-#299 observation for a future look: `POST /api/v1/channels` with `{}` returns 500 not a 400 (possible validation gap) — not yet filed. **Reviews**: one cold-context, same-model (Claude) review-only pass on the PR diff → MERGEABLE-WITH-NITS (no Critical/High/Medium; verified no vacuous-pass paths in the harness). Same-model cold review judged sufficient for a CI+shell+docs leaf with no concurrency/auth-handler/migration C# and <150 changed C# lines; cross-model/frontier reserved for those risk classes. No reviewer-repo audits picked up. **Arc change**: none (arc already fully in "Done history"; this was backlog work). **Dependency/conflict warnings**: none for the harness itself. #363 (its follow-up) will need the `docs/e2e-local.md` seeding recipe wired into the CI job + ffmpeg/scanner present; coordinate with anyone extending the `functional-e2e` job. Rebased cleanly over the #247 merge (only `docs/decisions.md` append-conflict).
Author
Owner

Correction — 2026-07-16 queue selection

The earlier version of this comment incorrectly treated the empty arc/milestone/review pools as the end of selection. That omitted #237's explicit backlog fall-through.

Live corrected ranking:

  1. #99 — priority: medium, open and unclaimed
  2. #333 — priority: medium, open and unclaimed
  3. #334 — priority: medium, open and unclaimed

#99 is now claimed (in-progress; claim comment posted). This session is continuing on its remaining channel-state wiring. A proper session-end comment will be added when the work finishes or is parked.

The reviewer-pool statement remains accurate: adversarial-reviewer #20–#24 already contain posted deliverables and are not pickups. The H12 script was a no-op because this shell lacks ETV_GITEA_TOKEN / ETV_GITEA_BASICAUTH.

## Correction — 2026-07-16 queue selection The earlier version of this comment incorrectly treated the empty arc/milestone/review pools as the end of selection. That omitted #237's explicit backlog fall-through. Live corrected ranking: 1. #99 — `priority: medium`, open and unclaimed 2. #333 — `priority: medium`, open and unclaimed 3. #334 — `priority: medium`, open and unclaimed #99 is now claimed (`in-progress`; claim comment posted). This session is continuing on its remaining channel-state wiring. A proper session-end comment will be added when the work finishes or is parked. The reviewer-pool statement remains accurate: adversarial-reviewer #20–#24 already contain posted deliverables and are not pickups. The H12 script was a no-op because this shell lacks `ETV_GITEA_TOKEN` / `ETV_GITEA_BASICAUTH`.
Author
Owner

Session log — 2026-07-16 (Claude Code / Opus 4.8 orchestrator)

Note: this tracker (#237) is closed — the arc is fully done. This was a post-arc backlog session (no open frontier, no open milestone issues, no open review items, no reviewer-repo audits). Logging here for continuity.

Picked up & landed: #345 (priority: high) — Plex pin flow reported connected but stranded authorized accounts with no servers. Chosen as the top available unclaimed high-priority item; the other priority: high (#357) is in flight under PR #358 + active worktrees, so excluded.

  • Fix: separated Plex authorization (stored token) from connection (discovered servers) in the SPA; new authorized-no-servers pin-flow terminal; hasPlexAccount gates Sign in vs Sign out; empty-servers messaging gated on !isLocked. Pure SPA, no API change.
  • Review: two cold-context adversarial passes. Initial diff BLOCKED (real High — empty-servers messaging ignored isLocked, would fire during discovery and urge sign-out mid-sync); fix re-review MERGEABLE @ 6532929c. Regression guards negative-control-proven.
  • Merged PR #366 → main @ ee044ab2 (H6/H10 gate auto-granted; Done-when ticked + verdict @ head + CI green). Main post-merge CI monitored.

Triage verdicts for new issues filed this session:

  • #367 (bug, frontend, priority: low) — backlog. Pre-existing, orthogonal gap surfaced in review: the budget-exhausted Plex message says "Use Refresh to check again," but there is no global Refresh control when zero servers exist. Not a gate item.

H12 audit: clean — every open issue carries a priority: label.

Candidate pool for next session (verify live state at pickup — do not trust this line): remaining priority: high backlog and open milestone-less bugs, e.g. #354 (subtitles drift, medium), #350 (cold-start ~7s), #340 (advertised IPTV base URL). #357 remains in flight under PR #358.

### Session log — 2026-07-16 (Claude Code / Opus 4.8 orchestrator) Note: this tracker (#237) is **closed** — the arc is fully done. This was a **post-arc backlog** session (no open frontier, no open milestone issues, no open `review` items, no reviewer-repo audits). Logging here for continuity. **Picked up & landed: #345** (`priority: high`) — Plex pin flow reported connected but stranded authorized accounts with no servers. Chosen as the top available unclaimed high-priority item; the other `priority: high` (#357) is in flight under PR #358 + active worktrees, so excluded. - Fix: separated Plex authorization (stored token) from connection (discovered servers) in the SPA; new `authorized-no-servers` pin-flow terminal; `hasPlexAccount` gates Sign in vs Sign out; empty-servers messaging gated on `!isLocked`. Pure SPA, no API change. - Review: two cold-context adversarial passes. Initial diff **BLOCKED** (real High — empty-servers messaging ignored `isLocked`, would fire during discovery and urge sign-out mid-sync); fix re-review **MERGEABLE @ 6532929c**. Regression guards negative-control-proven. - Merged **PR #366** → `main` @ `ee044ab2` (H6/H10 gate auto-granted; Done-when ticked + verdict @ head + CI green). Main post-merge CI monitored. **Triage verdicts for new issues filed this session:** - **#367** (`bug`, `frontend`, `priority: low`) — backlog. Pre-existing, orthogonal gap surfaced in review: the `budget-exhausted` Plex message says "Use Refresh to check again," but there is no global Refresh control when zero servers exist. Not a gate item. **H12 audit:** clean — every open issue carries a `priority:` label. **Candidate pool for next session (verify live state at pickup — do not trust this line):** remaining `priority: high` backlog and open milestone-less bugs, e.g. #354 (subtitles drift, medium), #350 (cold-start ~7s), #340 (advertised IPTV base URL). #357 remains in flight under PR #358.
Author
Owner

Session 2026-07-16 — #99 direct-stream on-air state

Closed: #99 (PR #368, merge ca5ac9446605157bf1fd7087c6888991750c3305)
Filed: none
Triage: no new issues. The H12 shell audit took its documented no-credentials no-op path; the equivalent MCP audit checked all 40 open non-PR issues and found none missing a priority: label.
Reviews: PR #368 received a cold review-only pass. One Low stale-route documentation finding was fixed in d1a52eaa; fresh exact-head re-review returned MERGEABLE, and CI run 661 passed on that head. No adversarial-reviewer audit was picked up or filed.
Arc change: none — this was post-arc backlog work; live #99 state is closed and the tracker body remains unchanged.
Dependency/conflict warnings: none.

## Session 2026-07-16 — #99 direct-stream on-air state **Closed**: #99 (PR #368, merge `ca5ac9446605157bf1fd7087c6888991750c3305`) **Filed**: none **Triage**: no new issues. The H12 shell audit took its documented no-credentials no-op path; the equivalent MCP audit checked all 40 open non-PR issues and found none missing a `priority:` label. **Reviews**: PR #368 received a cold review-only pass. One Low stale-route documentation finding was fixed in `d1a52eaa`; fresh exact-head re-review returned `MERGEABLE`, and CI run 661 passed on that head. No adversarial-reviewer audit was picked up or filed. **Arc change**: none — this was post-arc backlog work; live #99 state is closed and the tracker body remains unchanged. **Dependency/conflict warnings**: none.
Author
Owner

Done — Session 2026-07-16: cascading selector protocol

What was done / Closed: PR #369 merged at 5c1d2f63fa9268a9debf7938b89bf6b1605d67b0; no issue was closed because tracker #237 was already closed. Fresh sessions now build a bounded top-five shortlist by cascading arc → open milestone → review/audits → priority high → medium → low, with mandatory fall-through and a bounded raw-evidence fallback for cheap workers lacking repository-scoped listing.

Root cause: The prior selector contract capped output correctly but let empty arc/milestone/review pools terminate selection before repository-wide priority tiers were searched. A mini-worker tool limitation could also be mistaken for an empty tier.

Files changed / Arc change: docs/handoffs/chicorytv-issue-queue.md; #237 body edited to mirror the canonical tier order, top-five stop condition, exclusions, deterministic ordering, and tool-limitation fallback.

Verification / Reviews: git diff --check; stale three-candidate/two-fallback clauses absent; pre-push API drift, ESLint, typecheck, and production web build green; fresh low-cost restart simulation exercised the tool-limitation fallback and returned exactly #333, #334, #15, #69, #70 before stopping. Cold review and fallback re-review: MERGEABLE. CI run 663 green on reviewed head e8580f4a84e97941a7d2943fb9ebd7a9e716f5f8.

Filed / Follow-up issues: None.

Triage: No new issues. H12 shell audit took its no-credentials no-op path; the bounded MCP audit already confirmed all 40 open non-PR issues carry a priority: label.

Deferred / Dependency-conflict warnings: None.

Docs updated: Persistent handoff and canonical tracker body now agree.

## Done — Session 2026-07-16: cascading selector protocol **What was done / Closed**: PR #369 merged at `5c1d2f63fa9268a9debf7938b89bf6b1605d67b0`; no issue was closed because tracker #237 was already closed. Fresh sessions now build a bounded top-five shortlist by cascading arc → open milestone → review/audits → priority high → medium → low, with mandatory fall-through and a bounded raw-evidence fallback for cheap workers lacking repository-scoped listing. **Root cause**: The prior selector contract capped output correctly but let empty arc/milestone/review pools terminate selection before repository-wide priority tiers were searched. A mini-worker tool limitation could also be mistaken for an empty tier. **Files changed / Arc change**: `docs/handoffs/chicorytv-issue-queue.md`; #237 body edited to mirror the canonical tier order, top-five stop condition, exclusions, deterministic ordering, and tool-limitation fallback. **Verification / Reviews**: `git diff --check`; stale three-candidate/two-fallback clauses absent; pre-push API drift, ESLint, typecheck, and production web build green; fresh low-cost restart simulation exercised the tool-limitation fallback and returned exactly #333, #334, #15, #69, #70 before stopping. Cold review and fallback re-review: MERGEABLE. CI run 663 green on reviewed head `e8580f4a84e97941a7d2943fb9ebd7a9e716f5f8`. **Filed / Follow-up issues**: None. **Triage**: No new issues. H12 shell audit took its no-credentials no-op path; the bounded MCP audit already confirmed all 40 open non-PR issues carry a `priority:` label. **Deferred / Dependency-conflict warnings**: None. **Docs updated**: Persistent handoff and canonical tracker body now agree.
Author
Owner

Session 2026-07-16 — ChicoryTV subtitle timing diagnosis

Closed: #354 (no PR; exact production playback traced to mismatched source sidecars, with remediation documented)
Filed: timothy/server-management#599 — replace/resync the two verified mismatched sidecars and rescan Jellyfin/ErsatzTV
Triage: timothy/server-management#599 → backlog, priority: medium (content remediation; not an ErsatzTV arc blocker)
Reviews: none
Arc change: none
Dependency/conflict warnings: #599 owns the media-file correction; no ErsatzTV code surface changed. #357 remains explicitly parked and unclaimed pending owner authorization; this session did not reopen its architecture work.

## Session 2026-07-16 — ChicoryTV subtitle timing diagnosis **Closed**: #354 (no PR; exact production playback traced to mismatched source sidecars, with remediation documented) **Filed**: timothy/server-management#599 — replace/resync the two verified mismatched sidecars and rescan Jellyfin/ErsatzTV **Triage**: timothy/server-management#599 → backlog, priority: medium (content remediation; not an ErsatzTV arc blocker) **Reviews**: none **Arc change**: none **Dependency/conflict warnings**: #599 owns the media-file correction; no ErsatzTV code surface changed. #357 remains explicitly parked and unclaimed pending owner authorization; this session did not reopen its architecture work.
Author
Owner

Session 2026-07-16 — local-path auth + selector eligibility

Closed: #334 (PR #371)
Filed: none
Triage: No new issues. The H12 script no-op'd because its credential environment was absent; the equivalent Gitea MCP audit found all 38 open issues carry a priority: label.
Reviews: PR #371 cold security review was clean; its test-hardening follow-up was re-reviewed cleanly. PR #372 cold protocol review was clean.
Arc change: Body edited: automatic pickup now excludes parked issues, milestone records cannot be candidates, and missing repository-scoped issue listing requires literal TOOL_LIMITATION. Matching durable guidance merged in PR #372. #357 was parked as experimental; #333 was parked because the required Plex server does not exist. Both are explicit-pick-only.
Dependency/conflict warnings: none

## Session 2026-07-16 — local-path auth + selector eligibility **Closed**: #334 (PR #371) **Filed**: none **Triage**: No new issues. The H12 script no-op'd because its credential environment was absent; the equivalent Gitea MCP audit found all 38 open issues carry a `priority:` label. **Reviews**: PR #371 cold security review was clean; its test-hardening follow-up was re-reviewed cleanly. PR #372 cold protocol review was clean. **Arc change**: Body edited: automatic pickup now excludes `parked` issues, milestone records cannot be candidates, and missing repository-scoped issue listing requires literal `TOOL_LIMITATION`. Matching durable guidance merged in PR #372. #357 was parked as experimental; #333 was parked because the required Plex server does not exist. Both are explicit-pick-only. **Dependency/conflict warnings**: none
Author
Owner

Session 2026-07-16 — #15 coverage increment + login-border backlog

Closed: none — PR #370 merged (3654fad1, main CI green incl. :latest image) but it's a bounded increment of #15, which stays OPEN as the increments tracker.
Filed: #373 — white border around the login / shell-less boot pages (missing body { margin:0 } reset).
Triage: #373 → backlog (cosmetic, priority: low, no arc impact). Root cause is inferred from source, NOT rendered/reproduced — issue body flags this and its Done-when requires a render check before trusting the "~1 line" candidate fix.
Reviews: none picked up or delivered this session (two internal cold-context review passes on PR #370 only).
Arc change: none.
Dependency/conflict warnings:

  • Queue is drained to backlog priority: medium/low — arc/milestones/review tiers all empty. #334 (Medium) was already claimed (in-progress) by another session this session; do not re-pick.
  • Future coverage/#15 work MUST keep the coverlet.runsettings **/Migrations/*.cs exclusion — removing it re-introduces the exit-137 OOM on the shared Build & test job (whole-solution Architecture.Tests instruments 2.59M generated migration lines; exclusion cuts it 2.5M→20k coverable).

H12 audit: clean (all open issues carry a priority: label).

## Session 2026-07-16 — #15 coverage increment + login-border backlog **Closed**: none — PR #370 merged (`3654fad1`, main CI green incl. `:latest` image) but it's a bounded increment of **#15**, which stays OPEN as the increments tracker. **Filed**: #373 — white border around the login / shell-less boot pages (missing `body { margin:0 }` reset). **Triage**: #373 → **backlog** (cosmetic, `priority: low`, no arc impact). Root cause is inferred from source, NOT rendered/reproduced — issue body flags this and its Done-when requires a render check before trusting the "~1 line" candidate fix. **Reviews**: none picked up or delivered this session (two internal cold-context review passes on PR #370 only). **Arc change**: none. **Dependency/conflict warnings**: - Queue is drained to backlog `priority: medium/low` — arc/milestones/`review` tiers all empty. #334 (Medium) was already claimed (`in-progress`) by another session this session; do not re-pick. - **Future coverage/#15 work MUST keep the `coverlet.runsettings` `**/Migrations/*.cs` exclusion** — removing it re-introduces the exit-137 OOM on the shared `Build & test` job (whole-solution `Architecture.Tests` instruments 2.59M generated migration lines; exclusion cuts it 2.5M→20k coverable). H12 audit: clean (all open issues carry a `priority:` label).
Author
Owner

Session 2026-07-16 — #333 Plex real-server integration validation

Closed: #333 (docs commit dd273fbb on main, fixes #333) — the sole remaining #202 real-server item, the Plex interactive pin-flow, run against ersatztv-test + the newly-provisioned disposable test Plex server. Auth, server/library discovery (the 2026-07-13 servers:[] gap), sync-preference persistence, and sign-out cleanup all passed with captured API-state evidence.
Filed: none — no new defects in this happy-path run.
Triage: n/a (nothing filed).
Reviews: none — independent-review exemption recorded on #333 per its last Done-when box (only repo change is test-result documentation, no runtime behavior).
Arc change: none. (#333 was Backlog; it's now closed. Left the body Backlog line as-is — closed issues are already excluded from every pickup tier, so no re-pick risk.)
Dependency/conflict warnings:

  • The disposable plex service can now be removed from the media-servers stack (server-management-owned compose) — validation is done.
  • ersatztv-test local-admin credential was reset to admin / homelab default this session (prior session's password was unrecorded); noted in the #333 close comment.
  • #344 and #345 remain open as filed; #345's authorized/zero-server edge was NOT re-exercised (this run always had a discovered server).
## Session 2026-07-16 — #333 Plex real-server integration validation **Closed**: #333 (docs commit `dd273fbb` on `main`, `fixes #333`) — the sole remaining #202 real-server item, the Plex interactive pin-flow, run against `ersatztv-test` + the newly-provisioned disposable test Plex server. Auth, **server/library discovery** (the 2026-07-13 `servers:[]` gap), sync-preference persistence, and sign-out cleanup all passed with captured API-state evidence. **Filed**: none — no new defects in this happy-path run. **Triage**: n/a (nothing filed). **Reviews**: none — independent-review exemption recorded on #333 per its last Done-when box (only repo change is test-result documentation, no runtime behavior). **Arc change**: none. (#333 was Backlog; it's now closed. Left the body Backlog line as-is — closed issues are already excluded from every pickup tier, so no re-pick risk.) **Dependency/conflict warnings**: - The disposable `plex` service can now be removed from the `media-servers` stack (server-management-owned compose) — validation is done. - `ersatztv-test` local-admin credential was reset to `admin` / homelab default this session (prior session's password was unrecorded); noted in the #333 close comment. - #344 and #345 remain open as filed; #345's authorized/zero-server edge was NOT re-exercised (this run always had a discovered server).
Author
Owner

Session log — 2026-07-17 (Claude/Opus, backlog mode)

Picked #77 (top eligible: milestone-10 Scheduling: refactor + distribution, priority:medium; #380 in-progress + #70 blocked-by-#380 excluded; user excluded #388/#380).

Shipped: #77 CLOSED via PR #393 (merged 33d6c607, CI all-green, cold review MERGEABLE). Finding: clock-boundary padding is already implemented (upstream FillerMode.Pad+PadToNearestMinute for Classic, pad_to_next/pad_until for Sequential, Block time-anchored) — so scoped to verify+document, not build. Added a builder golden (Classic_clock_padded) + guide-projector test + docs (decisions/domain-model/testing). No production code. Live-E2E substituted with the two deterministic tests (no prod-code change → write-path justification N/A).

Filed (both blocked-by #388, per the standing "all UI work gated on #388" rule this session):

  • #392 — clock-align convenience toggle + 60-min increment (the #77 UI residual). priority:low.
  • #396 — collapsible left sidebar + collapsible nav groups (design-prototype change, user-requested capture). priority:medium.

Also: pushed a queue-lore fix to main (88287bcd) — Gitea's ?milestones=<name> filter silently no-ops on names with :/+ (returns ALL open issues), which mis-tiered #72 during selection; recipe = fetch-open-once + filter locally on .milestone.title.

Triage: #392 low / #396 medium — both correctly qualified (H12 audit clean). Note: #388's design-system mirror commit (e1d77ebe) landed on main this session (another workstream) — if #388 closes, #392/#396 auto-unblock.

**Session log — 2026-07-17 (Claude/Opus, backlog mode)** Picked #77 (top eligible: milestone-10 `Scheduling: refactor + distribution`, priority:medium; #380 in-progress + #70 blocked-by-#380 excluded; user excluded #388/#380). **Shipped:** #77 **CLOSED** via PR #393 (merged `33d6c607`, CI all-green, cold review MERGEABLE). Finding: clock-boundary padding is **already implemented** (upstream `FillerMode.Pad`+`PadToNearestMinute` for Classic, `pad_to_next`/`pad_until` for Sequential, Block time-anchored) — so scoped to verify+document, not build. Added a builder golden (`Classic_clock_padded`) + guide-projector test + docs (decisions/domain-model/testing). No production code. Live-E2E substituted with the two deterministic tests (no prod-code change → write-path justification N/A). **Filed (both blocked-by #388, per the standing "all UI work gated on #388" rule this session):** - #392 — clock-align convenience toggle + 60-min increment (the #77 UI residual). priority:low. - #396 — collapsible left sidebar + collapsible nav groups (design-prototype change, user-requested capture). priority:medium. **Also:** pushed a queue-lore fix to main (`88287bcd`) — Gitea's `?milestones=<name>` filter silently no-ops on names with `:`/`+` (returns ALL open issues), which mis-tiered #72 during selection; recipe = fetch-open-once + filter locally on `.milestone.title`. **Triage:** #392 low / #396 medium — both correctly qualified (H12 audit clean). Note: #388's design-system mirror commit (`e1d77ebe`) landed on main this session (another workstream) — if #388 closes, #392/#396 auto-unblock.
Author
Owner

Session log — 2026-07-17 (Claude/Opus 4.8, backlog mode)

Picked #73 (Seasonal / date-conditional channels & schedule rules). Closed as already implemented; the real gap was docs discoverability. No production code.

Area What happened Refs
Shipped #73 CLOSED as already-implemented. Date-conditional scheduling is fully present (IAlternateScheduleItem → ProgramScheduleAlternate/PlayoutTemplate, evaluated by AlternateScheduleSelector), SPA-reachable, and already unit-tested. Landed a task-shaped docs recipe instead of a build. commit 62c15da4 on main (fixes #73), docs-only
Docs channels.md → "Recipe: seasonal / holiday programming" (both engines + 5 gotchas); domain-model.md glossary row (mirrors the #77 row); decisions.md entry + index link 3 files, +110
Design Frontier verdict (evidence-contradicts-the-issue escalation): close, don't #77-re-scope — #77 had two gaps (undocumented AND untested), #73 had neither, so even #77's re-scope target was already met. Close-don't-park then forbids holding it open for a hypothetical build. decisions.md 2026-07-17
Routed "Prioritize collection X during a date range" → #70, not built here. Decomposes as date-scoping (shipped) × soft-prioritization (#70's weighting primitive). Building a second weighting primitive while #70 is mid-flight on the same PlaybackOrder switch sites, same milestone, would have been a direct collision. Composition note posted on #70 with the one caveat (weighting must live on the item/block, not the playout). #70 comment
Backlog filed None, deliberately. Rejected as-designed: per-item date predicate (zero new expressible behaviour, engine-risk in the #163/#380 golden-net area), default-DecoId date-gating (a date-conditional fallback is a contradiction), UI rename (upstream vocabulary in routes/API/parity docs). "Auto-surface a channel in the lineup" is a genuine absence but speculative → named as a known non-goal, no placeholder issue (close-don't-park). —
Captured The load-bearing undocumented fact: blank StartYear/EndYear = repeats every year (AlternateScheduleSelector.cs:32-40) — years default to the queried date's year. Override needs both years (only one → silently still yearly), and explicit years force reverse = false, disabling wrap-around. Verified by reading the selector, not assumed. Also: the issue body's "Deco-like modifier" premise is wrong — decos carry no dates (DecoTemplateItem is time-of-day). channels.md, domain-model.md
Outstanding Nothing from this issue. —

Selection note (tier-1 was not empty — the cascade never reached the flat backlog): open milestones are Auto-Tune DetailPanel (11), Scheduling: refactor + distribution (10), Design system (6). The whole Auto-Tune milestone is unavailable — #383 is an umbrella container, #384 in-progress (claimed 08:49), #385 blocked-by open #70, #386 blocked-by #385+#384; #59 is likewise an epic container. Eligible tier-1 remainder: #73, #176 (both priority: low; #395 blocked-by open #381) → tie on priority → lowest number → #73.

⚠️ Selector-quality warning for future sessions. The cheap selector packet was wrong in two ways and would have caused a bad pickup if trusted:

  1. It reported #385 and #386 as deps:clear — both are actually blocked (#385 by open #70; #386 by open #385+#384). It evidently did not call GET /issues/{n}/dependencies.
  2. It missed #383 entirely (the only priority: medium in tier 1) and ranked purely by issue number, ignoring the priority-within-tier rule.

Both were caught only because the orchestrator re-derived tier-1 membership locally and ran the dependency checks itself. Recommendation: the orchestrator's "one focused recheck of the winner" is not sufficient when the selector's tier is contested — re-verify the whole tier's deps + priority ordering, not just the proposed winner. The existing lore already warns that the ?milestones=<name> filter silently no-ops on names with :/+ (both open milestones here have them); the fix used was the documented one — fetch all open issues once and filter locally on .milestone.title.

H12 audit: clean — every open issue carries a priority: label. No new issues filed, so nothing to label.
Claims: in-progress removed from #73. Did not touch #70 or #384 (both live claims by parallel sessions).
Review exemption (stated, not silent): docs-only, no server-state effect → independent review skipped per the rubric. No live-E2E: no production code changed, same reasoning as #77; mechanism stays pinned by existing AlternateScheduleSelectorTests + playoutTemplateCalendar.test.ts.

**Session log — 2026-07-17 (Claude/Opus 4.8, backlog mode)** Picked **#73** (*Seasonal / date-conditional channels & schedule rules*). Closed as **already implemented**; the real gap was docs discoverability. No production code. | Area | What happened | Refs | |---|---|---| | **Shipped** | **#73 CLOSED** as already-implemented. Date-conditional scheduling is fully present (`IAlternateScheduleItem` → `ProgramScheduleAlternate`/`PlayoutTemplate`, evaluated by `AlternateScheduleSelector`), SPA-reachable, and already unit-tested. Landed a task-shaped docs recipe instead of a build. | commit `62c15da4` on `main` (`fixes #73`), docs-only | | **Docs** | `channels.md` → "Recipe: seasonal / holiday programming" (both engines + 5 gotchas); `domain-model.md` glossary row (mirrors the #77 row); `decisions.md` entry + index link | 3 files, +110 | | **Design** | Frontier verdict (evidence-contradicts-the-issue escalation): **close**, don't #77-re-scope — #77 had two gaps (undocumented AND untested), #73 had neither, so even #77's re-scope target was already met. Close-don't-park then forbids holding it open for a hypothetical build. | `decisions.md` 2026-07-17 | | **Routed** | *"Prioritize collection X during a date range"* → **#70**, not built here. Decomposes as date-scoping (shipped) × soft-prioritization (#70's weighting primitive). Building a second weighting primitive while #70 is mid-flight on the same `PlaybackOrder` switch sites, same milestone, would have been a direct collision. Composition note posted on #70 with the one caveat (weighting must live on the item/block, not the playout). | #70 comment | | **Backlog filed** | **None, deliberately.** Rejected as-designed: per-item date predicate (zero new expressible behaviour, engine-risk in the #163/#380 golden-net area), default-`DecoId` date-gating (a date-conditional *fallback* is a contradiction), UI rename (upstream vocabulary in routes/API/parity docs). "Auto-surface a channel in the lineup" is a genuine absence but speculative → named as a **known non-goal**, no placeholder issue (close-don't-park). | — | | **Captured** | The load-bearing undocumented fact: **blank `StartYear`/`EndYear` = repeats every year** (`AlternateScheduleSelector.cs:32-40`) — years default to the queried date's year. Override needs **both** years (only one → silently still yearly), and explicit years force `reverse = false`, disabling wrap-around. Verified by reading the selector, not assumed. Also: the issue body's "Deco-like modifier" premise is wrong — decos carry no dates (`DecoTemplateItem` is time-of-day). | `channels.md`, `domain-model.md` | | **Outstanding** | Nothing from this issue. | — | **Selection note (tier-1 was not empty — the cascade never reached the flat backlog):** open milestones are `Auto-Tune DetailPanel` (11), `Scheduling: refactor + distribution` (10), `Design system` (6). The whole Auto-Tune milestone is unavailable — **#383 is an umbrella container**, **#384 `in-progress`** (claimed 08:49), **#385 blocked-by open #70**, **#386 blocked-by #385+#384**; #59 is likewise an epic container. Eligible tier-1 remainder: #73, #176 (both `priority: low`; #395 blocked-by open #381) → tie on priority → lowest number → **#73**. ⚠️ **Selector-quality warning for future sessions.** The cheap selector packet was **wrong in two ways** and would have caused a bad pickup if trusted: 1. It reported **#385 and #386 as `deps:clear`** — both are actually **blocked** (`#385` by open #70; `#386` by open #385+#384). It evidently did not call `GET /issues/{n}/dependencies`. 2. It **missed #383 entirely** (the only `priority: medium` in tier 1) and ranked purely by issue number, ignoring the priority-within-tier rule. Both were caught only because the orchestrator re-derived tier-1 membership locally and ran the dependency checks itself. **Recommendation: the orchestrator's "one focused recheck of the winner" is not sufficient when the selector's tier is contested — re-verify the whole tier's deps + priority ordering, not just the proposed winner.** The existing lore already warns that the `?milestones=<name>` filter silently no-ops on names with `:`/`+` (both open milestones here have them); the fix used was the documented one — fetch all open issues once and filter locally on `.milestone.title`. **H12 audit:** clean — every open issue carries a `priority:` label. No new issues filed, so nothing to label. **Claims:** `in-progress` removed from #73. Did not touch #70 or #384 (both live claims by parallel sessions). **Review exemption (stated, not silent):** docs-only, no server-state effect → independent review skipped per the rubric. No live-E2E: no production code changed, same reasoning as #77; mechanism stays pinned by existing `AlternateScheduleSelectorTests` + `playoutTemplateCalendar.test.ts`.
Author
Owner

Session — #384 shipped (Auto-Tune DetailPanel backend, read endpoint)

Client/model: Claude Code, Opus 4.8 (orchestrator), Haiku selector + Explore recon/review subagents.

Selected: cheap selector shortlisted the #383 Auto-Tune DetailPanel milestone bundle as the top eligible tier. Of its three children, #384 (read-only, no deps) was the only workable one — #385 and #386 were/are blocked (see below). Claimed + worked #384.

Shipped:

  • #384 CLOSED → PR #397 merged to main (1be45fbe → merge ed4b0d45). GET /api/v1/channels/auto-tune/members — lists a proposed channel's distinct content-source members by running the server-owned GenerateQuery through ISearchIndex and rolling leaf items up to distinct shows/movies (faithful to the built SmartCollection playout). Reuses existing paged DTO; 10 new tests; cold review MERGEABLE (one Low fixed: out-of-range axis → 500 → empty). Docs: api-conventions §5, decisions.md, domain-model. Rebased onto d38181cb (code byte-identical) to clear a docs-only conflict from the parallel #380/#73 merges.

Bundle finding (resolves the #383 sweep):

  • #385 is blocked-by #70 — a parallel session's recon established its rotation-weights half is #70's weighted-distribution scope, which must land on #380's just-merged ShuffleSourceBuilder seam (not the old scattered switches). Its per-channel-templateId/Advanced-overrides slice is ready-to-plumb once unblocked.
  • #386 (SPA) remains blocked-by both #384 (now done) and #385.
  • Umbrella #383 stays open until #385 + #386 land.

Triage: filed no new issues. H12 qualification audit clean (every open issue carries a priority: label).

Live-state next-session note (verify at pickup, don't trust this line): #70 (priority: medium) is now unblocked (#380 closed) but is currently in-progress (parallel session). With #70/#385/#386 all excluded (claimed or blocked), the next eligible pickup falls to the priority:-label backlog cascade — re-derive from live state.

### Session — #384 shipped (Auto-Tune DetailPanel backend, read endpoint) **Client/model:** Claude Code, Opus 4.8 (orchestrator), Haiku selector + Explore recon/review subagents. **Selected:** cheap selector shortlisted the #383 **Auto-Tune DetailPanel** milestone bundle as the top eligible tier. Of its three children, #384 (read-only, no deps) was the only workable one — #385 and #386 were/are blocked (see below). Claimed + worked #384. **Shipped:** - **#384 CLOSED** → PR **#397** merged to main (`1be45fbe` → merge `ed4b0d45`). `GET /api/v1/channels/auto-tune/members` — lists a proposed channel's distinct content-source members by running the server-owned `GenerateQuery` through `ISearchIndex` and rolling leaf items up to distinct shows/movies (faithful to the built SmartCollection playout). Reuses existing paged DTO; 10 new tests; cold review MERGEABLE (one Low fixed: out-of-range axis → 500 → empty). Docs: api-conventions §5, decisions.md, domain-model. Rebased onto `d38181cb` (code byte-identical) to clear a docs-only conflict from the parallel #380/#73 merges. **Bundle finding (resolves the #383 sweep):** - **#385** is `blocked-by #70` — a parallel session's recon established its rotation-weights half *is* #70's weighted-distribution scope, which must land on #380's just-merged `ShuffleSourceBuilder` seam (not the old scattered switches). Its per-channel-`templateId`/Advanced-overrides slice is ready-to-plumb once unblocked. - **#386** (SPA) remains `blocked-by` both #384 (now done) and #385. - Umbrella **#383** stays open until #385 + #386 land. **Triage:** filed no new issues. H12 qualification audit clean (every open issue carries a `priority:` label). **Live-state next-session note (verify at pickup, don't trust this line):** #70 (`priority: medium`) is now **unblocked** (#380 closed) but is currently `in-progress` (parallel session). With #70/#385/#386 all excluded (claimed or blocked), the next eligible pickup falls to the `priority:`-label backlog cascade — re-derive from live state.
Author
Owner

Session 2026-07-17 — #72 (channel health) shipped; #164 released with design

Claimed: #72 + #164 as the "channel health / status UX" bundle. Only #72 shipped; #164 released (see below).

Selection note (lore reinforced): the cheap selector ranked #383 first — but #383 is the auto-tune umbrella and #384 was already in-progress by a parallel session. Independent re-derivation caught it. Widened the standing lore: the selector's unreliable outputs include in-progress claim state and umbrella-vs-child, not just deps/priority. Picked #72/#164 (the only unclaimed, unblocked priority: medium bundle) instead.

Shipped

  • #72 CLOSED → PR #405 (merged, 1e35248e). Per-channel PlayoutCount on the channels API + a "No playout" badge/filter in the lineup. Root cause found under it: GET /api/v1/channels/{id} reported playoutCount: 0 for every channel (missing .Include(Playouts) on an AsNoTracking read), silently disabling the channel editor's playout-source guard. Two cold reviews (MERGEABLE). Landing took 2 rebases + 2 lost merge races against fast-moving main — merge_when_checks_succeed is the tool for that (H6 blocks a partial-green synchronous merge).
  • PR #417 (queued, docs-only) — standing-lore batch: batch-pushes / trust-the-queue / no-"main-checkout" / diagnosing-CI-reds.

Released, not shipped

  • #164 — health-check audit + guided remediation. Not started; claim released, in-progress removed, and the full design verdict posted on #164 (14-check Pass-1 classification, DTO evolution Brief+Remediation{Kind,Target}, the 3 real defects incl. the SPA ignoring check.link, PR-B/PR-C slicing). Next session starts from a plan.

Filed (all priority:-labeled; H12 ✓)

  • #401 (bug, low) — UpdateChannel silently coerces Mirror→Generated when a playout exists; should 422.
  • #414 (enhancement, low) — auto-tuned-vs-user origin marker; blocked-by #383 (dependency set + verified).
  • #415 (enhancement, low) — per-channel empty-schedule + broken-source fault detection (empty-schedule only correct for Classic today — the trap that deferred it).

Outstanding / handoff

  • Design-sync pull NOT done for the Channels screen — the change was mirrored into design-system/Channels.jsx without first pulling the current prototype from Claude Design. If that design moved, the mirror is on a stale base. Worth a /design-sync reconcile.
  • #164 is the obvious next pickup (design ready). Its PR C (health-check TTL cache) should be filed as its own issue when picked up — orthogonal to the UX.

Recommended next: #164 (design recorded, ready to implement). Verify live state at pickup per protocol.

## Session 2026-07-17 — #72 (channel health) shipped; #164 released with design **Claimed:** #72 + #164 as the "channel health / status UX" bundle. Only #72 shipped; #164 released (see below). **Selection note (lore reinforced):** the cheap selector ranked **#383 first — but #383 is the auto-tune *umbrella* and #384 was already `in-progress`** by a parallel session. Independent re-derivation caught it. Widened the standing lore: the selector's unreliable outputs include `in-progress` claim state and umbrella-vs-child, not just deps/priority. Picked #72/#164 (the only unclaimed, unblocked `priority: medium` bundle) instead. ### Shipped - **#72 CLOSED → PR #405 (merged, `1e35248e`).** Per-channel `PlayoutCount` on the channels API + a "No playout" badge/filter in the lineup. **Root cause found under it:** `GET /api/v1/channels/{id}` reported `playoutCount: 0` for *every* channel (missing `.Include(Playouts)` on an AsNoTracking read), silently disabling the channel editor's playout-source guard. Two cold reviews (MERGEABLE). Landing took 2 rebases + 2 lost merge races against fast-moving main — `merge_when_checks_succeed` is the tool for that (H6 blocks a partial-green synchronous merge). - **PR #417 (queued, docs-only)** — standing-lore batch: batch-pushes / trust-the-queue / no-"main-checkout" / diagnosing-CI-reds. ### Released, not shipped - **#164** — health-check audit + guided remediation. **Not started**; claim released, `in-progress` removed, and the **full design verdict posted on #164** (14-check Pass-1 classification, DTO evolution `Brief`+`Remediation{Kind,Target}`, the 3 real defects incl. the SPA ignoring `check.link`, PR-B/PR-C slicing). Next session starts from a plan. ### Filed (all `priority:`-labeled; H12 ✓) - **#401** (`bug`, low) — `UpdateChannel` silently coerces Mirror→Generated when a playout exists; should 422. - **#414** (`enhancement`, low) — auto-tuned-vs-user origin marker; **blocked-by #383** (dependency set + verified). - **#415** (`enhancement`, low) — per-channel empty-schedule + broken-source fault detection (empty-schedule only correct for Classic today — the trap that deferred it). ### Outstanding / handoff - **Design-sync pull NOT done** for the Channels screen — the change was mirrored *into* `design-system/Channels.jsx` without first pulling the current prototype from Claude Design. If that design moved, the mirror is on a stale base. Worth a `/design-sync` reconcile. - **#164 is the obvious next pickup** (design ready). Its PR C (health-check TTL cache) should be filed as its own issue when picked up — orthogonal to the UX. **Recommended next:** #164 (design recorded, ready to implement). Verify live state at pickup per protocol.
Author
Owner

Session — 2026-07-17 (Claude Code / Opus 4.8)

Worked: #376 — XMLTV access_token raw-interpolation (unmilestoned priority: medium backlog bug). Merged (PR #419), issue closed.

Queue selection notes (for the next selector):

  • Selector packet ranked #383 as the tier-2 winner — wrong: #383 is an epic/container (children #384 done, #385, #386 are the pickups; Done-when = "child endpoint issues shipped"). The whole Auto-Tune DetailPanel milestone is currently blocked: #385 blocked-by open #70 (in-progress), #386 blocked-by #385. Exclude #383 as a container.
  • Scheduling: refactor + distribution milestone: #70 in-progress; #395 blocked-by open #381; #176 is the only deps-clear member but it's a priority: low plan-mode "rethink creation flow" vision epic overlapping in-progress #70 — deferred (needs a brainstorming/design arc, coordinate with #70).
  • Both open milestone names contain :/+, so the server-side milestones= filter no-ops — filtered locally as required. Re-verified deps via GET /issues/{n}/dependencies on every tier-2 candidate (selector's deps data was again unreliable — it reported #383 clear but missed the container status).
  • With tier-2 having no clean pickup, fell through to unmilestoned priority: medium and (per user choice between the vision epic and a bounded bug) took the security/correctness bug #376.

Fix: SecurityElement.Escape on the token value in GetChannelGuideHandler (mirrors #340's {RequestBase} escaping); M3U path left unchanged (not XML). Regression test Guide_xml_escapes_access_token (non-vacuous). Cold adversarial review → MERGEABLE.

Filed: #421 (bug, priority: low) — M3U url-tvg="..." token-quoting concern (distinct, out-of-scope observation from the #376 review). H12 audit clean.

No in-progress labels left by this session.

### Session — 2026-07-17 (Claude Code / Opus 4.8) **Worked:** #376 — XMLTV `access_token` raw-interpolation (unmilestoned `priority: medium` backlog bug). **Merged (PR #419), issue closed.** **Queue selection notes (for the next selector):** - Selector packet ranked **#383 as the tier-2 winner — wrong**: #383 is an **epic/container** (children #384 done, #385, #386 are the pickups; Done-when = "child endpoint issues shipped"). The whole **Auto-Tune DetailPanel** milestone is currently blocked: #385 blocked-by open **#70** (`in-progress`), #386 blocked-by #385. Exclude #383 as a container. - **Scheduling: refactor + distribution** milestone: #70 `in-progress`; #395 blocked-by open #381; #176 is the only deps-clear member but it's a `priority: low` **plan-mode "rethink creation flow" vision epic** overlapping in-progress #70 — deferred (needs a brainstorming/design arc, coordinate with #70). - Both open milestone names contain `:`/`+`, so the server-side `milestones=` filter no-ops — filtered locally as required. Re-verified deps via `GET /issues/{n}/dependencies` on every tier-2 candidate (selector's deps data was again unreliable — it reported #383 clear but missed the container status). - With tier-2 having no clean pickup, fell through to unmilestoned `priority: medium` and (per user choice between the vision epic and a bounded bug) took the security/correctness bug #376. **Fix:** `SecurityElement.Escape` on the token value in `GetChannelGuideHandler` (mirrors #340's `{RequestBase}` escaping); M3U path left unchanged (not XML). Regression test `Guide_xml_escapes_access_token` (non-vacuous). Cold adversarial review → MERGEABLE. **Filed:** #421 (`bug`, `priority: low`) — M3U `url-tvg="..."` token-quoting concern (distinct, out-of-scope observation from the #376 review). H12 audit clean. **No `in-progress` labels left by this session.**
Author
Owner

Session — 2026-07-17 (Claude Code / Opus 4.8)

Shipped: #320 — TroubleshootController segment-wait loop now breaks on notifier.IsFailed (was spinning until client-cancel when ffmpeg died after the playlist but before segments, holding the request thread + troubleshooting lock). PR #423 merged (de63603a), issue closed. Regression test added (negative-control-verified); two cold review passes; local + CI green.

Selection notes (verify-the-selector, again): the cheap selector packet jumped straight to Tier-5 (priority: medium bugs) claiming "eligible milestones all have blocked/in-progress members." That was wrong: it read #385 as still blocked by #70, but #70 merged today, so Gitea auto-cleared the block — #385 is an eligible milestone-tier pickup. Re-deriving deps locally (GET /issues/{n}/dependencies) is what caught it. Corrected live milestone-tier state as of this session:

  • #383 = umbrella/container (Done-when = "child endpoints shipped") → exclude; its read-endpoint child #384 merged today (PR #397).
  • #385 = eligible now (blocker #70 closed). Note it was released at 09:02 today over a #70 scope-collision; #70 has since merged, so its documented 4-piece re-scope is now unblocked. Concrete write-path backend; will need independent review + live-E2E.
  • #386 blocked-by #385 (open); #395 blocked-by #381 (open) → both excluded.
  • #176 = large PseudoTV-style "rethink creation" design epic (no Done-when) — design-first, not a mechanical pickup.

Operator was offered the milestone-tier fork (revive #385 / start #176 design epic / clean Tier-5 bug) and chose a clean Tier-5 bug, hence #320.

New issues filed: none. H12 audit: clean (every open issue carries a priority: label).

No forward "recommended next" recorded (drift rule) — read live state. Whoever's next: #385 is now genuinely unblocked and is the top eligible milestone pickup if the Auto-Tune DetailPanel arc is to continue.

## Session — 2026-07-17 (Claude Code / Opus 4.8) **Shipped:** #320 — TroubleshootController segment-wait loop now breaks on `notifier.IsFailed` (was spinning until client-cancel when ffmpeg died after the playlist but before segments, holding the request thread + troubleshooting lock). PR #423 merged (`de63603a`), issue closed. Regression test added (negative-control-verified); two cold review passes; local + CI green. **Selection notes (verify-the-selector, again):** the cheap selector packet jumped straight to Tier-5 (`priority: medium` bugs) claiming "eligible milestones all have blocked/in-progress members." That was **wrong**: it read **#385 as still blocked by #70**, but **#70 merged today**, so Gitea auto-cleared the block — #385 is an *eligible* milestone-tier pickup. Re-deriving deps locally (`GET /issues/{n}/dependencies`) is what caught it. Corrected live milestone-tier state as of this session: - **#383** = umbrella/container (Done-when = "child endpoints shipped") → exclude; its read-endpoint child **#384 merged today** (PR #397). - **#385** = eligible now (blocker #70 closed). Note it was *released* at 09:02 today over a #70 scope-collision; #70 has since merged, so its documented 4-piece re-scope is now unblocked. Concrete write-path backend; will need independent review + live-E2E. - **#386** blocked-by #385 (open); **#395** blocked-by #381 (open) → both excluded. - **#176** = large PseudoTV-style "rethink creation" **design epic** (no Done-when) — design-first, not a mechanical pickup. Operator was offered the milestone-tier fork (revive #385 / start #176 design epic / clean Tier-5 bug) and chose a **clean Tier-5 bug**, hence #320. **New issues filed:** none. **H12 audit:** clean (every open issue carries a `priority:` label). **No forward "recommended next" recorded** (drift rule) — read live state. Whoever's next: #385 is now genuinely unblocked and is the top eligible milestone pickup if the Auto-Tune DetailPanel arc is to continue.
Author
Owner

Session — #385 Auto-Tune DetailPanel backend (per-channel overrides)

Picked: #385 (milestone Auto-Tune DetailPanel, tier 2). Selector's #1 was #383 — correctly rejected as the milestone umbrella (children #384/#385/#386 are the pickups; the lore-flagged trap). #385 had been claimed+released earlier today over a #70/#380 scope collision; verified all three cited blockers (#70 WeightedShuffle, #380 ShuffleSourceBuilder, #384 member-read) had since closed, so re-claimed.

Shipped (merged PR #428 → 1abc5df1): additive per-channel overrides on POST /api/v1/channels/auto-tune — optional templateId / advanced / logo, reusing the manual Channel Builder's CreateChannelFromLineupAdvancedOptionsRequest verbatim (no parallel DTO). Backward-compatible. Tests + OpenAPI regen + docs (decisions/api-conventions/domain-model).

Deviation — scope split (operator-approved): #385's body bundled per-source rotation weights, but honoring them needs a MultiCollection-of-per-source-SmartCollections redesign (#70's weights live only on MultiCollection join rows; an auto-tuned channel is one SmartCollection). Split the weights + query-corrections into #425 ([PLAN-MODE], milestone Auto-Tune DetailPanel, priority: low). Bug-initials/colour generated logo also deferred (needs persisted Channel state + FFmpeg-pipeline wiring). #385's Done-when was rescoped accordingly before merge.

Verification: cold independent review → MERGEABLE; live-E2E → PASS (overrides round-trip through channel-detail read-back, playout + M3U/XMLTV clean); CI green. Note: hit a decisions.md conflict treadmill — main advanced 13 then 2 commits during CI cycles, each re-conflicting only on the append-only decisions log; rebased through (no code change) and used Gitea auto-merge-on-green to land before the next conflict.

New issue triage: #425 (weights redesign) — qualified priority: low, milestoned; a real design task, not a throwaway.

Board delta: #385 CLOSED. #386 (SPA DetailPanel) now unblocked (deps #384+#385 both closed). #425 open (weights). Milestone Auto-Tune DetailPanel remaining: #386 (SPA) + #425 (weights).

Recommended next (guess — live state wins): #386 consumes this endpoint but is SPA/frontend; #425 is its backend sibling (weights). Either is eligible; #425 unblocks the full weight story #386 wants.

## Session — #385 Auto-Tune DetailPanel backend (per-channel overrides) **Picked**: #385 (milestone *Auto-Tune DetailPanel*, tier 2). Selector's #1 was #383 — correctly rejected as the milestone **umbrella** (children #384/#385/#386 are the pickups; the lore-flagged trap). #385 had been claimed+released earlier today over a #70/#380 scope collision; verified all three cited blockers (#70 WeightedShuffle, #380 ShuffleSourceBuilder, #384 member-read) had since **closed**, so re-claimed. **Shipped** (merged PR #428 → `1abc5df1`): additive per-channel overrides on `POST /api/v1/channels/auto-tune` — optional `templateId` / `advanced` / `logo`, reusing the manual Channel Builder's `CreateChannelFromLineupAdvancedOptionsRequest` verbatim (no parallel DTO). Backward-compatible. Tests + OpenAPI regen + docs (decisions/api-conventions/domain-model). **Deviation — scope split (operator-approved)**: #385's body bundled per-source **rotation weights**, but honoring them needs a MultiCollection-of-per-source-SmartCollections redesign (#70's weights live only on MultiCollection join rows; an auto-tuned channel is one SmartCollection). Split the weights + query-corrections into **#425** (`[PLAN-MODE]`, milestone *Auto-Tune DetailPanel*, priority: low). Bug-initials/colour generated logo also deferred (needs persisted Channel state + FFmpeg-pipeline wiring). #385's Done-when was rescoped accordingly before merge. **Verification**: cold independent review → MERGEABLE; live-E2E → PASS (overrides round-trip through channel-detail read-back, playout + M3U/XMLTV clean); CI green. Note: hit a `decisions.md` conflict **treadmill** — main advanced 13 then 2 commits during CI cycles, each re-conflicting only on the append-only decisions log; rebased through (no code change) and used Gitea **auto-merge-on-green** to land before the next conflict. **New issue triage**: #425 (weights redesign) — qualified priority: low, milestoned; a real design task, not a throwaway. **Board delta**: #385 CLOSED. #386 (SPA DetailPanel) now **unblocked** (deps #384+#385 both closed). #425 open (weights). Milestone *Auto-Tune DetailPanel* remaining: #386 (SPA) + #425 (weights). **Recommended next (guess — live state wins)**: #386 consumes this endpoint but is SPA/frontend; #425 is its backend sibling (weights). Either is eligible; #425 unblocks the full weight story #386 wants.
Author
Owner

Session — 2026-07-18 (Claude Code / Opus 4.8 orchestrator)

Worked: #164 (Health-checks UX / guided remediation) → merged, PR #430 (merge 0a34d1a5). Selector ranked milestone-tier #425 first, but recon showed it was [PLAN-MODE] design-first + a split-sibling of the then-in-progress #385 (weighted-distribution surface) — declined to collide; user chose #164 from the priority:medium tier. Implemented from the prior session's recorded design verdict.

Shipped: server-declared remediation {Kind, Target} on an additive health DTO (Brief + Remediation, deprecated Link kept); mapper made total (NotApplicable no longer throws); 2 stale Blazor links fixed; dead SettingsScreen /system/health row removed; SPA renders remediation. Full docs (decisions/api-conventions/blazor-route-parity) + OpenAPI regen. Independent review MERGEABLE (1 CSS finding fixed). Live-verified on a booted instance.

Filed: #431 — TTL-cache PerformHealthChecks (deferred perf half of #164; priority: medium, Done-when included).

Merge friction (for the next session): landing crossed the Gitea maintenance window — 10 commits (incl. #385) landed on main, forcing two rebases (append-only decisions.md + regenerated-OpenAPI conflicts; regenerated, never hand-merged). Then the decisions.md append-only CI job showed failure but was a killed job (missing server-side log, 10-min duration for a seconds-long git-diff) from a runner restart — re-triggered green via a no-op --amend push (rerun_run/cancel are 404 on this Gitea 1.25.4). H6 gate correctly blocked the merge while overall CI was red; merged once all-green.

H12 audit: clean (all open issues qualified). Deferred: design-sync mirror of the Dashboard health-panel remediation UI back to the Claude Design prototype (low novelty — reused existing Button/ctv-button).

### Session — 2026-07-18 (Claude Code / Opus 4.8 orchestrator) **Worked:** #164 (Health-checks UX / guided remediation) → **merged, PR #430** (merge `0a34d1a5`). Selector ranked milestone-tier #425 first, but recon showed it was `[PLAN-MODE]` design-first + a split-sibling of the then-in-progress #385 (weighted-distribution surface) — declined to collide; user chose #164 from the priority:medium tier. Implemented from the prior session's recorded design verdict. **Shipped:** server-declared remediation `{Kind, Target}` on an additive health DTO (`Brief` + `Remediation`, deprecated `Link` kept); mapper made total (NotApplicable no longer throws); 2 stale Blazor links fixed; dead SettingsScreen `/system/health` row removed; SPA renders remediation. Full docs (decisions/api-conventions/blazor-route-parity) + OpenAPI regen. Independent review MERGEABLE (1 CSS finding fixed). Live-verified on a booted instance. **Filed:** #431 — TTL-cache `PerformHealthChecks` (deferred perf half of #164; priority: medium, Done-when included). **Merge friction (for the next session):** landing crossed the **Gitea maintenance window** — 10 commits (incl. #385) landed on `main`, forcing **two rebases** (append-only `decisions.md` + regenerated-OpenAPI conflicts; regenerated, never hand-merged). Then the `decisions.md append-only` CI job showed `failure` but was a **killed job** (missing server-side log, 10-min duration for a seconds-long git-diff) from a runner restart — re-triggered green via a no-op `--amend` push (`rerun_run`/`cancel` are 404 on this Gitea 1.25.4). H6 gate correctly blocked the merge while overall CI was red; merged once all-green. **H12 audit:** clean (all open issues qualified). **Deferred:** design-sync mirror of the Dashboard health-panel remediation UI back to the Claude Design prototype (low novelty — reused existing Button/ctv-button).
Author
Owner

Session — #386 Auto-Tune DetailPanel slide-over (SPA) ✅ merged

Picked: #386 (tier-2, milestone Auto-Tune DetailPanel, priority: low) — the lowest-numbered eligible milestone issue; backend children #384/#385 both CLOSED so it was unblocked. Selector shortlist verified against live state (deps + in-progress re-derived myself per the lore).

Shipped (PR #432, merged): the SPA Configure slide-over. New reusable SlideOver primitive; extracted the Channel Builder advanced-options model to a shared builder/advancedOptions.tsx (ChannelBuilder imports it unchanged, tests byte-identical); getAutoTuneChannelMembers API; the DetailPanel + wiring. Docs: spa-conventions §11 + decisions.md #386.

Scope call (issue body was stale): the create DTO has no weight/exclude field — #385 split those to #425. Deferred weight steppers + corrections to #425 (in-pane hint); dropped MiniEpg / bug-initials / "generated query" as backend-less decoration. Frontier-reviewed (Fable) + recorded in decisions.md.

Verify: 808 web tests + lint + build green; cold adversarial review (no blockers, 3 Mediums fixed) + fix-commit re-review clean; live-E2E on a seeded local instance (browser: preview→Configure→toggle Shuffle→Create; created channel schedule = PlaybackOrder=Shuffle overridden vs SeasonEpisode axis-default). H10 verdict posted @ head; merge gate auto-granted (all Done-when ticked, CI green).

New issues filed: none. Follow-up: #425 (weights/corrections) remains the open sibling in this milestone.

H12 audit: clean — every open issue carries a priority: label.

### Session — #386 Auto-Tune DetailPanel slide-over (SPA) ✅ merged **Picked:** #386 (tier-2, milestone *Auto-Tune DetailPanel*, `priority: low`) — the lowest-numbered eligible milestone issue; backend children #384/#385 both CLOSED so it was unblocked. Selector shortlist verified against live state (deps + in-progress re-derived myself per the lore). **Shipped (PR #432, merged):** the SPA Configure slide-over. New reusable `SlideOver` primitive; extracted the Channel Builder advanced-options model to a shared `builder/advancedOptions.tsx` (ChannelBuilder imports it unchanged, tests byte-identical); `getAutoTuneChannelMembers` API; the DetailPanel + wiring. Docs: spa-conventions §11 + decisions.md #386. **Scope call (issue body was stale):** the create DTO has no weight/exclude field — #385 split those to **#425**. Deferred weight steppers + corrections to #425 (in-pane hint); dropped MiniEpg / bug-initials / "generated query" as backend-less decoration. Frontier-reviewed (Fable) + recorded in decisions.md. **Verify:** 808 web tests + lint + build green; cold adversarial review (no blockers, 3 Mediums fixed) + fix-commit re-review clean; live-E2E on a seeded local instance (browser: preview→Configure→toggle Shuffle→Create; created channel schedule = `PlaybackOrder=Shuffle` overridden vs `SeasonEpisode` axis-default). H10 verdict posted @ head; merge gate auto-granted (all Done-when ticked, CI green). **New issues filed:** none. **Follow-up:** #425 (weights/corrections) remains the open sibling in this milestone. **H12 audit:** clean — every open issue carries a `priority:` label.
Author
Owner

Session (Claude Code / Opus 4.8) — #425 shipped

Worked: #425 — Auto-Tune per-source rotation weights + query corrections. Selected as top-of-queue (arc frontier empty → tier-2 milestone Auto-Tune DetailPanel; re-derived the tier myself: #383 umbrella, #386/#176 in-progress, #395 blocked by open #381, leaving #425).

Outcome: MERGED (PR #439). [PLAN-MODE] design decision recorded in decisions.md (Option A — reuse #70's MultiCollection/WeightedShuffle machinery). Backend-only, create-flow (extends the existing bulk-create request, not a post-hoc PUT — corrected the plan draft after finding #386 is a create-wizard). Dual-provider migration AddCollectionOwnedByChannelId. Live-E2E proved 3:1 weighting end-to-end (playout + guide). Two-pass cold-Opus review found + fixed a blocker (search pickers leaked owned collections → user-schedule-delete via FK cascade). 2268+ tests green. Docs (decisions/domain-model/api-conventions) + OpenAPI trio updated.

Note for the queue: #425 was scoped backend-only. The SPA weight steppers (DetailPanel UI wiring the new sources field) remain a follow-up — #386 shipped the panel but deferred the steppers to #425, which delivered only the backend. Whoever picks up the Auto-Tune DetailPanel SPA next should wire the steppers/exclude/add-untagged to POST /api/v1/channels/auto-tune's sources. Not yet filed as its own issue.

Filed: none. H12 audit: clean (every open issue has a priority label).

### Session (Claude Code / Opus 4.8) — #425 shipped **Worked:** #425 — Auto-Tune per-source rotation weights + query corrections. Selected as top-of-queue (arc frontier empty → tier-2 milestone `Auto-Tune DetailPanel`; re-derived the tier myself: #383 umbrella, #386/#176 in-progress, #395 blocked by open #381, leaving #425). **Outcome:** MERGED (PR #439). `[PLAN-MODE]` design decision recorded in decisions.md (Option A — reuse #70's MultiCollection/WeightedShuffle machinery). Backend-only, create-flow (extends the existing bulk-create request, not a post-hoc PUT — corrected the plan draft after finding #386 is a create-wizard). Dual-provider migration `AddCollectionOwnedByChannelId`. Live-E2E proved 3:1 weighting end-to-end (playout + guide). Two-pass cold-Opus review found + fixed a blocker (search pickers leaked owned collections → user-schedule-delete via FK cascade). 2268+ tests green. Docs (decisions/domain-model/api-conventions) + OpenAPI trio updated. **Note for the queue:** #425 was scoped backend-only. The **SPA weight steppers** (DetailPanel UI wiring the new `sources` field) remain a follow-up — #386 shipped the panel but deferred the steppers to #425, which delivered only the backend. Whoever picks up the Auto-Tune DetailPanel SPA next should wire the steppers/exclude/add-untagged to `POST /api/v1/channels/auto-tune`'s `sources`. Not yet filed as its own issue. **Filed:** none. **H12 audit:** clean (every open issue has a priority label).
Author
Owner

Session — 2026-07-18 (Claude Code / Opus 4.8)

Worked: #308 — Concurrent same-item Add*ToCollection can 500 on composite-PK violation. ✅ CLOSED, PR #441 merged to main (2281f2e7).

Selection: arc / gate / open-milestone / review tiers all empty (expected steady state). Fell through to the priority: backlog; #308 was the lowest-numbered eligible priority: medium (deps clear, unclaimed). Verified the selector's mechanical claims myself (deps via GET /dependencies, in-progress, umbrella) — all clean.

What shipped: TrySaveChangesForcingVersion (catches only a classified unique/PK violation → false); 10 single-item handlers no-op on it, the bulk handler retries against fresh membership so it never drops non-colliding items. Provider detection via a TvContext.IsUniqueConstraintViolation static delegate (matches the existing IsSqlite/LastInsertedRowId seam) wired to Sqlite/MySqlErrorClassifier. Add*ToPlaylist audited → not affected (identity PK, duplicates legal).

Verification: full suite green (~3,800); every fix-dependent test verified to fail with the fix disabled (non-vacuous). Independent cold review MERGEABLE. Live-E2E on real SQLite WAL genuinely fired 6 UNIQUE constraint failed collisions — all caught, all 204, exactly-one-row invariant held.

Root cause: the #269 membership pre-check isn't atomic with the insert. No migration / no OpenAPI regen needed.

Docs: api-conventions.md §7a (new "Idempotent insert under concurrency"), decisions/optimistic-concurrency.md (dated entry).

New issues filed: none. Deferred: none. H12 audit: ✓ clean.

Notable technique (saved to memory): deterministic reproduction of a DB load→save race needs a shared-cache multi-connection SQLite harness + a SavingChanges interceptor that interposes the racer after the handler's read; inserting the racer before Handle is a vacuity trap (the handler's own reload picks it up → pre-check no-ops → the test passes even with the fix broken; the negative control caught this).

Recommended next: back to the priority: backlog — next eligible medium after #308 (verify live state; the arc/milestone/review tiers remain empty).

### Session — 2026-07-18 (Claude Code / Opus 4.8) **Worked:** #308 — *Concurrent same-item `Add*ToCollection` can 500 on composite-PK violation.* ✅ **CLOSED**, PR #441 merged to `main` (`2281f2e7`). **Selection:** arc / gate / open-milestone / `review` tiers all empty (expected steady state). Fell through to the `priority:` backlog; #308 was the lowest-numbered eligible `priority: medium` (deps clear, unclaimed). Verified the selector's mechanical claims myself (deps via `GET /dependencies`, in-progress, umbrella) — all clean. **What shipped:** `TrySaveChangesForcingVersion` (catches only a classified unique/PK violation → `false`); 10 single-item handlers no-op on it, the bulk handler retries against fresh membership so it never drops non-colliding items. Provider detection via a `TvContext.IsUniqueConstraintViolation` static delegate (matches the existing `IsSqlite`/`LastInsertedRowId` seam) wired to `Sqlite`/`MySqlErrorClassifier`. `Add*ToPlaylist` audited → **not affected** (identity PK, duplicates legal). **Verification:** full suite green (~3,800); every fix-dependent test verified to fail with the fix disabled (non-vacuous). Independent cold review MERGEABLE. Live-E2E on real SQLite WAL genuinely fired 6 `UNIQUE constraint failed` collisions — all caught, all 204, exactly-one-row invariant held. **Root cause:** the #269 membership pre-check isn't atomic with the insert. No migration / no OpenAPI regen needed. **Docs:** `api-conventions.md` §7a (new "Idempotent insert under concurrency"), `decisions/optimistic-concurrency.md` (dated entry). **New issues filed:** none. **Deferred:** none. **H12 audit:** ✓ clean. **Notable technique (saved to memory):** deterministic reproduction of a DB load→save race needs a shared-cache multi-connection SQLite harness + a `SavingChanges` interceptor that interposes the racer *after* the handler's read; inserting the racer before `Handle` is a vacuity trap (the handler's own reload picks it up → pre-check no-ops → the test passes even with the fix broken; the negative control caught this). **Recommended next:** back to the `priority:` backlog — next eligible medium after #308 (verify live state; the arc/milestone/review tiers remain empty).
Author
Owner

Session (Claude Code / Opus 4.8 orchestrator) — 2026-07-18

Worked / closed: #293 — Page/cap GET /api/v1/search/all-items (DoS hardening). Merged via PR #442.

Selection: arc frontier empty; tier-2 (open milestones) had no eligible pickup — #383 is the Auto-Tune DetailPanel epic container (children #384/#385/#386 all closed), #425 + #176 in-progress, #395 gated on open #381. Tier-3 empty: no review-labeled issue, and all five ersatztv reviewer audits (adversarial-reviewer #20–#24) already have posted deliverables. No priority: high issues exist → fell through to priority: medium, lowest eligible = #293. (Selector packet's "deps clear" was wrong on #395 as usual — re-derived the tier myself.)

Outcome: paginated the endpoint (pageNum/pageSize clamped; additive per-kind Totals), handler threads native Lucene skip/limit, SPA add-all pages to completeness (option (a), operator-confirmed). Two cold adversarial passes → MERGEABLE; both LOW findings fixed in-diff. Rebased onto a 26-commits-ahead origin/main (regenerated OpenAPI so main's new /search/fields survived). CI green, merged, #293 closed.

New issues filed: none. H12 audit: clean (every open issue carries a priority: label).

Queue note (live state wins): Auto-Tune DetailPanel milestone still open (owned by live #425); Scheduling: refactor + distribution still has #176 (in-progress) + #395 (gated on #381). Next eligible backlog pickup by the cascade is the next-lowest priority: medium after #293 (e.g. #308/#350/#363/#396/#398/#403/#404/#412/#420/#431) — re-derive live at pickup, don't trust this list.

**Session (Claude Code / Opus 4.8 orchestrator) — 2026-07-18** **Worked / closed: #293** — Page/cap `GET /api/v1/search/all-items` (DoS hardening). Merged via PR #442. **Selection**: arc frontier empty; tier-2 (open milestones) had no eligible pickup — #383 is the Auto-Tune DetailPanel **epic container** (children #384/#385/#386 all closed), #425 + #176 `in-progress`, #395 gated on open #381. Tier-3 empty: no `review`-labeled issue, and all five `ersatztv` reviewer audits (adversarial-reviewer #20–#24) already have posted deliverables. No `priority: high` issues exist → fell through to `priority: medium`, lowest eligible = **#293**. (Selector packet's "deps clear" was wrong on #395 as usual — re-derived the tier myself.) **Outcome**: paginated the endpoint (`pageNum`/`pageSize` clamped; additive per-kind `Totals`), handler threads native Lucene `skip`/`limit`, SPA add-all pages to completeness (option (a), operator-confirmed). Two cold adversarial passes → MERGEABLE; both LOW findings fixed in-diff. Rebased onto a 26-commits-ahead `origin/main` (regenerated OpenAPI so main's new `/search/fields` survived). CI green, merged, #293 closed. **New issues filed**: none. **H12 audit**: clean (every open issue carries a `priority:` label). **Queue note (live state wins)**: Auto-Tune DetailPanel milestone still open (owned by live #425); `Scheduling: refactor + distribution` still has #176 (in-progress) + #395 (gated on #381). Next eligible backlog pickup by the cascade is the next-lowest `priority: medium` after #293 (e.g. #308/#350/#363/#396/#398/#403/#404/#412/#420/#431) — re-derive live at pickup, don't trust this list.
Author
Owner

Session — 2026-07-18 (Claude Code orchestrator)

Selection. Arc/gate/review/priority: high tiers empty (steady state). Re-derived the milestone tier myself (didn't trust the selector): both open-milestone issues are ineligible — #383 is a done umbrella epic (children #384/#385/#386 all closed), #395 is blocked-by open #381. Fell through to priority: medium, lowest number. #350 was claimed by a parallel session ~2 min before me (its comment reasoned identically) → backed off per later-claimant-backs-off → claimed the next-lowest eligible medium, #363.

Worked #363 → PR #443 (MERGED, main c4695b1095; fixes #363). functional-E2E harness: added two deterministic lock-contention 409 flows (library-scan already-scanning, external-collections already-scanning) to scripts/e2e-functional.sh. Probed a real instance + measured the ~4s scan window before sizing the seed; verified 38/38 across 4+ fresh runs + green on CI. 2-round independent cold review → MERGEABLE @ head. Docs updated in-PR (e2e-local.md, ci-cd.md, workflow comment).

Deferred (spun out, not shipped flaky): #444 playout-build lock 409 + isLocked (#215, priority: medium) and #445 Playwright headless flows (priority: low) — both need heavier seeding / a wide observation window, and the harness rule is only add deterministic assertions.

Also filed: #447 (priority: low) — flaky LibrariesScreen polling SPA test that intermittently reds the required Build & test gate (failed then passed on the same web/ code during this PR's CI). H12 audit clean.

Recommended next: normal backlog cascade — remaining priority: medium (e.g. #396, #398, #403, #404, #412, #420, #431). Note the new #444 (medium) is blocked in practice on real seeding work but not marked blocked-by. (Live Gitea state is truth; this line is a guess.)

### Session — 2026-07-18 (Claude Code orchestrator) **Selection.** Arc/gate/review/`priority: high` tiers empty (steady state). Re-derived the milestone tier myself (didn't trust the selector): both open-milestone issues are ineligible — #383 is a **done umbrella epic** (children #384/#385/#386 all closed), #395 is **blocked-by open #381**. Fell through to `priority: medium`, lowest number. #350 was claimed by a parallel session ~2 min before me (its comment reasoned identically) → backed off per later-claimant-backs-off → claimed the next-lowest eligible medium, **#363**. **Worked #363 → PR #443 (MERGED, main c4695b1095; `fixes #363`).** functional-E2E harness: added two deterministic lock-contention 409 flows (library-scan already-scanning, external-collections already-scanning) to `scripts/e2e-functional.sh`. Probed a real instance + measured the ~4s scan window before sizing the seed; verified 38/38 across 4+ fresh runs + green on CI. 2-round independent cold review → MERGEABLE @ head. Docs updated in-PR (e2e-local.md, ci-cd.md, workflow comment). **Deferred (spun out, not shipped flaky):** #444 playout-build lock 409 + `isLocked` (#215, priority: medium) and #445 Playwright headless flows (priority: low) — both need heavier seeding / a wide observation window, and the harness rule is *only add deterministic assertions*. **Also filed:** #447 (priority: low) — flaky `LibrariesScreen` polling SPA test that intermittently reds the required `Build & test` gate (failed then passed on the same `web/` code during this PR's CI). H12 audit clean. **Recommended next:** normal backlog cascade — remaining `priority: medium` (e.g. #396, #398, #403, #404, #412, #420, #431). Note the new #444 (medium) is blocked in practice on real seeding work but not marked `blocked-by`. (Live Gitea state is truth; this line is a guess.)
Author
Owner

Session — 2026-07-18 (Claude Code orchestrator)

Picked: #350 (Channel cold-start ~7s to first segment) — lowest-numbered eligible priority: medium. Both open-milestone candidates were ineligible: #383 is an umbrella epic (children #384/#385/#386 closed), #395 is blocked-by open #381. No arc/review/priority: high tier open (maintenance mode, as expected). Selection cross-checked by a cheap selector + independent dep re-derivation (GET /issues/{n}/dependencies on every plausible candidate). Ran with 2 other startup sessions in flight — claimed fast; no collision.

Delivered: PR #446 (merged, CI-green, independent review MERGEABLE) — HLS cold-start latency instrumentation for #350. Scoped instrumentation-only per measure-before-optimize (the issue's driver ranking was a single-observation estimate); one structured Information log per tune-in with a setup/startup/fill phase split + cleanly-detectable feature flags (subtitle burn-in, hwaccel). Watermark/HDR deliberately not flagged (overlay-ambiguity in args). Log-only, no transcode behavior change.

#350 left OPEN as the optimization tracker — the actual fix (warm-start pre-roll / conditional subtitle skip / readrate burst) awaits real prod numbers from the new log. in-progress removed.

New issues filed: none. H12 audit: clean (every open issue carries a priority: label).

Queue note for next session: the top eligible priority: medium backlog after #350 is #363 (may already be claimed — a #363 E2E PR #443 merged this session), then #396, #398, #403, #404, #412, #420, #431.

### Session — 2026-07-18 (Claude Code orchestrator) **Picked:** #350 (Channel cold-start ~7s to first segment) — lowest-numbered eligible `priority: medium`. Both open-milestone candidates were ineligible: #383 is an umbrella epic (children #384/#385/#386 closed), #395 is blocked-by open #381. No arc/`review`/`priority: high` tier open (maintenance mode, as expected). Selection cross-checked by a cheap selector + independent dep re-derivation (`GET /issues/{n}/dependencies` on every plausible candidate). Ran with 2 other startup sessions in flight — claimed fast; no collision. **Delivered:** PR #446 (merged, CI-green, independent review MERGEABLE) — HLS cold-start latency **instrumentation** for #350. Scoped instrumentation-only per measure-before-optimize (the issue's driver ranking was a single-observation estimate); one structured Information log per tune-in with a `setup/startup/fill` phase split + cleanly-detectable feature flags (subtitle burn-in, hwaccel). Watermark/HDR deliberately not flagged (overlay-ambiguity in args). Log-only, no transcode behavior change. **#350 left OPEN** as the optimization tracker — the actual fix (warm-start pre-roll / conditional subtitle skip / readrate burst) awaits real prod numbers from the new log. `in-progress` removed. **New issues filed:** none. **H12 audit:** clean (every open issue carries a `priority:` label). **Queue note for next session:** the top eligible `priority: medium` backlog after #350 is #363 (may already be claimed — a #363 E2E PR #443 merged this session), then #396, #398, #403, #404, #412, #420, #431.
Author
Owner

Session 2026-07-18 (Claude Code orchestrator)

Picked: #396 — Collapsible left sidebar + nav-group accordions (SPA). Selector shortlist was unreliable — its #1 pick #383 is an umbrella whose children #384/#385/#386 are all closed (surfaced for closing); #395 is blocked by open #381; it also missed eligible #396/#431/#444. Re-derived the tier locally → tier-5 unmilestoned priority: medium, lowest number, unblocked = #396 (blocker #388 closed).

Shipped: PR #448 → merged to main (6e976644). 60px icon rail + accordion groups (Media/System collapsible, Primary always open, default-collapsed), inventory-driven, two ctv-sidebar-* localStorage keys, a11y label-in-tree + tooltip, active-route reveal in both states. Build/lint/855 vitest green; cold review found 2 issues (active-route-in-collapsed-group; mobile brand bar) — both fixed + re-verified; live-E2E (Playwright headless) green. Docs: spa-conventions §13 + decisions.md.

CI note: a decisions.md append-only job was killed under runner saturation (missing log). I then created avoidable churn (empty re-trigger commit + force-push a green sha, which re-ran the pipeline) — operator selectively re-ran the job (green) and cancelled the dup runs. Landed via operator-authorized force-merge. New lesson saved: don't push to clear a killed job.

Triage: filed no new issues. H12 audit clean (all open issues carry a priority: label).

Surfaced for the operator: umbrella #383 (Auto-Tune DetailPanel) — all children closed, likely just needs closing. #395 blocked by open #381 (near-duplicate golden-tests issue) — worth reconciling.

**Session 2026-07-18 (Claude Code orchestrator)** **Picked:** #396 — Collapsible left sidebar + nav-group accordions (SPA). Selector shortlist was unreliable — its #1 pick #383 is an umbrella whose children #384/#385/#386 are all closed (surfaced for closing); #395 is blocked by open #381; it also missed eligible #396/#431/#444. Re-derived the tier locally → tier-5 unmilestoned `priority: medium`, lowest number, unblocked = **#396** (blocker #388 closed). **Shipped:** PR #448 → merged to main (6e976644). 60px icon rail + accordion groups (Media/System collapsible, Primary always open, default-collapsed), inventory-driven, two `ctv-sidebar-*` localStorage keys, a11y label-in-tree + tooltip, active-route reveal in both states. Build/lint/855 vitest green; cold review found 2 issues (active-route-in-collapsed-group; mobile brand bar) — both fixed + re-verified; live-E2E (Playwright headless) green. Docs: spa-conventions §13 + decisions.md. **CI note:** a `decisions.md append-only` job was killed under runner saturation (missing log). I then created avoidable churn (empty re-trigger commit + force-push a green sha, which re-ran the pipeline) — operator selectively re-ran the job (green) and cancelled the dup runs. Landed via operator-authorized force-merge. New lesson saved: don't push to clear a killed job. **Triage:** filed no new issues. H12 audit clean (all open issues carry a `priority:` label). **Surfaced for the operator:** umbrella **#383** (Auto-Tune DetailPanel) — all children closed, likely just needs closing. **#395** blocked by open **#381** (near-duplicate golden-tests issue) — worth reconciling.
Author
Owner

Session — batched small-issue sweep (Claude Code, Opus 4.8)

Directive: knock out a load of small backlog items in one go; parallel worktrees, one PR per item, fitting model per slice.

Landed (8 issues / 7 PRs, all squash-merged to main):

PR Issue(s) Change Review
#449 #310 strip residual UTF-8 BOM (2 of 17; rest already clean) mechanical
#450 #367 Plex budget-exhausted msg + always-available "Check again" button SPA leaf
#451 #409, #447 LibrariesScreen "Never scanned" label + flaky-test fix SPA leaf
#452 #338 ZAP exit-2 warn vs exit-1 fail + --selftest + docs/ci-cd.md inline
#453 #410 scan-cancel → LogInformation (4 handlers) inline
#454 #327 playlist rename name validation → 422 cold review MERGEABLE + live-E2E
#456 #401 reject Mirror+playout → 422 + dead-else data-integrity fix cold review found the 2nd bug + live-E2E

Review earned its keep: the mandatory cold review of #401 caught that removing the old silent coercion exposed a pre-existing dead else-branch, letting a stray MirrorSourceChannelId/PlayoutOffset persist onto a Generated channel. Fixed + negative-controlled; live-E2E confirmed the null-through-read-path.

Live-E2E (scripts/e2e-local.sh, real /api, machine-key) run for both write-path PRs (#327/#401): 7/7 mutation checks pass with round-trip GETs.

New issue filed: #458 (priority: low) — playlist/group rename should reject duplicate names like create does (inherited gap, deferred from #327).

Deviation flagged: #409 fixed client-side (year<1900 sentinel) rather than the issue author's mild preference for API-boundary null normalization — chosen to avoid a DB migration for a priority:low bug; noted in the PR/close for a maintainer call.

Process notes: dotnet compiles are the local RAM pressure point on this Mac — serialized to one-at-a-time mid-session after operator flag; idle Roslyn build-server reclaim recovered 32%→71% free. All 7 branches BOM-clean before push; batched review verdicts; H12 audit clean.

Queue state: arc/gate empty (steady state); this was a priority:low backlog sweep. Next session: normal priority-label cascade (the selector this session surfaced #350/#403/#404/#412/#431 as the priority:medium single-pickups — verify live before claiming).

## Session — batched small-issue sweep (Claude Code, Opus 4.8) **Directive:** knock out a load of small backlog items in one go; parallel worktrees, one PR per item, fitting model per slice. **Landed (8 issues / 7 PRs, all squash-merged to main):** | PR | Issue(s) | Change | Review | |----|----------|--------|--------| | #449 | #310 | strip residual UTF-8 BOM (2 of 17; rest already clean) | mechanical | | #450 | #367 | Plex budget-exhausted msg + always-available "Check again" button | SPA leaf | | #451 | #409, #447 | LibrariesScreen "Never scanned" label + flaky-test fix | SPA leaf | | #452 | #338 | ZAP exit-2 warn vs exit-1 fail + --selftest + docs/ci-cd.md | inline | | #453 | #410 | scan-cancel → LogInformation (4 handlers) | inline | | #454 | #327 | playlist rename name validation → 422 | cold review MERGEABLE + **live-E2E** | | #456 | #401 | reject Mirror+playout → 422 **+ dead-else data-integrity fix** | cold review found the 2nd bug + **live-E2E** | **Review earned its keep:** the mandatory cold review of #401 caught that removing the old silent coercion exposed a pre-existing dead `else`-branch, letting a stray `MirrorSourceChannelId`/`PlayoutOffset` persist onto a Generated channel. Fixed + negative-controlled; live-E2E confirmed the null-through-read-path. **Live-E2E** (scripts/e2e-local.sh, real /api, machine-key) run for both write-path PRs (#327/#401): 7/7 mutation checks pass with round-trip GETs. **New issue filed:** #458 (priority: low) — playlist/group *rename* should reject duplicate names like *create* does (inherited gap, deferred from #327). **Deviation flagged:** #409 fixed client-side (year<1900 sentinel) rather than the issue author's mild preference for API-boundary null normalization — chosen to avoid a DB migration for a priority:low bug; noted in the PR/close for a maintainer call. **Process notes:** dotnet compiles are the local RAM pressure point on this Mac — serialized to one-at-a-time mid-session after operator flag; idle Roslyn build-server reclaim recovered 32%→71% free. All 7 branches BOM-clean before push; batched review verdicts; H12 audit clean. **Queue state:** arc/gate empty (steady state); this was a priority:low backlog sweep. Next session: normal priority-label cascade (the selector this session surfaced #350/#403/#404/#412/#431 as the priority:medium single-pickups — verify live before claiming).
Author
Owner

Session 2026-07-18 (Claude Code orchestrator, Opus 4.8)

Worked: #403 — Scheduling: unsupported PlaybackOrder fails silently at 5/6 dispatch sites → CLOSED via PR #457 (merged, CI green).

Selection notes (for the next session):

  • Arc / open-milestone / review / priority: high tiers were all empty (expected steady state). Fell through to priority: medium backlog.
  • Selector's #1 pick was #350, but it's already claimed by a parallel session (claimed 14:53; instrumentation shipped via merged PR #446; remaining optimization is deferred + data-gated) — excluded. Its claim comment noted "two other sessions in startup," so expect collisions on the top backlog picks. Took the next unclaimed tier-5 by number: #403.
  • Other tier-5 mediums seen and left OPEN/unclaimed: #404 (SPA weights UI — blocker #388 now CLOSED, so it's newly eligible; good next pick), #412 (CI peak-anon — note #406's lesson about that measurement being in-flight), #431 (health-check TTL cache).

What shipped in #403: three silent build-time sites (Classic/Playlist/Block) now log a Warning + keep a non-fatal fallback; explicit Block Random arm; new PlaybackOrderSupport matrix + tripwire test (a new order OR engine kind fails CI until classified); Block's allow-list now reads the matrix. Write-path rejection and the reverse _ => None mappings deliberately left alone (documented in the close comment + decisions.md). No golden drift.

Review: independent cold adversarial pass + focused re-review of the fix commit → MERGEABLE @ 1d95cbfe. Two Low findings (misleading Marathon log; engine-axis coverage gap) fixed before merge.

Docs: docs/decisions.md appended (main had NOT touched decisions.md this cycle — clean, no conflict).

H12 audit: clean — every open issue carries a priority: label. No new issues filed.

## Session 2026-07-18 (Claude Code orchestrator, Opus 4.8) **Worked:** #403 — Scheduling: unsupported `PlaybackOrder` fails silently at 5/6 dispatch sites → **CLOSED** via PR #457 (merged, CI green). **Selection notes (for the next session):** - Arc / open-milestone / `review` / `priority: high` tiers were all empty (expected steady state). Fell through to `priority: medium` backlog. - Selector's #1 pick was **#350**, but it's **already claimed by a parallel session** (claimed 14:53; instrumentation shipped via merged PR #446; remaining optimization is deferred + data-gated) — excluded. Its claim comment noted "two other sessions in startup," so expect collisions on the top backlog picks. Took the next unclaimed tier-5 by number: **#403**. - Other tier-5 mediums seen and left OPEN/unclaimed: **#404** (SPA weights UI — blocker #388 now CLOSED, so it's newly eligible; good next pick), **#412** (CI peak-anon — note #406's lesson about that measurement being in-flight), **#431** (health-check TTL cache). **What shipped in #403:** three silent build-time sites (Classic/Playlist/Block) now log a Warning + keep a non-fatal fallback; explicit Block `Random` arm; new `PlaybackOrderSupport` matrix + tripwire test (a new order OR engine kind fails CI until classified); Block's allow-list now reads the matrix. Write-path rejection and the reverse `_ => None` mappings deliberately left alone (documented in the close comment + decisions.md). No golden drift. **Review:** independent cold adversarial pass + focused re-review of the fix commit → MERGEABLE @ 1d95cbfe. Two Low findings (misleading Marathon log; engine-axis coverage gap) fixed before merge. **Docs:** `docs/decisions.md` appended (main had NOT touched decisions.md this cycle — clean, no conflict). **H12 audit:** clean — every open issue carries a `priority:` label. No new issues filed.
Author
Owner

Session — 2026-07-18 (Claude Code, Opus 4.8 orchestrator)

Worked the CI-redundancy cluster #398 + #420 (operator asked for a large plan/design issue).

Issue Outcome
#398 CI build-once CLOSED — measured & rejected. Fully designed + implemented (a compile job → single 972 MB gz artifact → --no-build consumers) and went green on CI (run 830), then rejected on measurement: ~12% slot-occupancy saving for a ~40–85% per-run wall-clock regression. Root cause: the 2.5 GB/972 MB artifact (tar ~82s CPU + ~180s transport eats most of the ~465s saved) and compile serializing before migrations' long ef-replay (DB work build-once can't shorten).
#420 cross-run tree-skip DONE — merged (PR #455, 2fff5fc0). Shipped standalone (per-job revalidate detect mirroring the #416 docs-only pattern; no compile job/artifact; no branch-protection change). Review MERGEABLE, live-verified vs the Gitea API. Fail-safe; build always runs (no image ships unvalidated). Honest hit-rate: near-inert — rebase-before-merge (this repo's decisions.md treadmill) + parallel merges mint a new tree that was never CI-validated, so it correctly declines almost always.

Reusable findings (recorded in docs/decisions.md):

  • actions/upload-artifact@v4 does NOT work on this Gitea (GHESNotSupportedError — @actions/artifact v2 rejects any non-github.com host). Use @v3 for any future artifact use.
  • The A→C decision rule (roundtrip vs compile-CPU) was insufficient — it missed the tar CPU and the serialization wall-clock regression. Measure real per-job timings, not the paper estimate. (Reinforces the existing "measure CI-behavior, not green" / "measure before optimizing from the issue premise" lessons.)

New issues filed: none. H12 audit: clean (all open issues qualified). No in-progress labels left.

Recommended next: back to the priority: backlog cascade. The remaining CI-cluster sibling #412 (peak-anon memory instrument) is unblocked and independent; the other priority: medium backlog items (#396 now merged as #448, #403 merged as #457, #431 health-check TTL cache, #404 SPA weight inputs) — re-derive live state before claiming (this comment is a guess written before the next session acts).

### Session — 2026-07-18 (Claude Code, Opus 4.8 orchestrator) Worked the CI-redundancy cluster #398 + #420 (operator asked for a large plan/design issue). | Issue | Outcome | |---|---| | **#398** CI build-once | **CLOSED — measured & rejected.** Fully designed + implemented (a `compile` job → single 972 MB gz artifact → `--no-build` consumers) and went **green on CI** (run 830), then rejected on measurement: **~12% slot-occupancy saving for a ~40–85% per-run wall-clock regression**. Root cause: the 2.5 GB/972 MB artifact (tar ~82s CPU + ~180s transport eats most of the ~465s saved) and `compile` serializing *before* migrations' long ef-replay (DB work build-once can't shorten). | | **#420** cross-run tree-skip | **DONE — merged (PR #455, `2fff5fc0`).** Shipped standalone (per-job `revalidate` detect mirroring the #416 docs-only pattern; no compile job/artifact; no branch-protection change). Review MERGEABLE, live-verified vs the Gitea API. Fail-safe; `build` always runs (no image ships unvalidated). **Honest hit-rate: near-inert** — rebase-before-merge (this repo's decisions.md treadmill) + parallel merges mint a new tree that was never CI-validated, so it correctly declines almost always. | **Reusable findings (recorded in `docs/decisions.md`):** - `actions/upload-artifact@v4` does NOT work on this Gitea (`GHESNotSupportedError` — `@actions/artifact` v2 rejects any non-github.com host). **Use `@v3`** for any future artifact use. - The A→C decision *rule* (roundtrip vs compile-CPU) was insufficient — it missed the tar CPU and the serialization wall-clock regression. Measure real per-job timings, not the paper estimate. (Reinforces the existing "measure CI-behavior, not green" / "measure before optimizing from the issue premise" lessons.) **New issues filed:** none. **H12 audit:** clean (all open issues qualified). No `in-progress` labels left. **Recommended next:** back to the `priority:` backlog cascade. The remaining CI-cluster sibling **#412** (peak-anon memory instrument) is unblocked and independent; the other `priority: medium` backlog items (#396 now merged as #448, #403 merged as #457, #431 health-check TTL cache, #404 SPA weight inputs) — re-derive live state before claiming (this comment is a guess written before the next session acts).
Author
Owner

Addendum (same session): #409 was reworked to the maintainer-preferred API/MCP-parity fix (PR #459, merged) — the client-side SPA heuristic was replaced with a read-boundary coercion in GetAllMediaSourcesForApiHandler + a dual-provider data migration nulling the historical 0001-01-01 sentinel rows, and the SPA reverted to a plain null check. Cold review caught that MediaSourceRepository still WRITES the sentinel today on disable-sync flows, so the read-coercion is the load-bearing permanent guard (a migration-only fix would have regressed). Live-E2E confirmed lastScan=null via the real API. Filed follow-ups: #458 (playlist/group rename should reject duplicate names) and #460 (normalize MediaSourceRepository writes to null). Notes: hit the decisions.md conflict treadmill twice — resolved via rebase + auto-merge (merge_when_checks_succeed); one decisions-guard flake (guard passed locally + on synthetic merge, 0 deletions) cleared on re-trigger.

**Addendum (same session):** #409 was reworked to the maintainer-preferred API/MCP-parity fix (PR #459, merged) — the client-side SPA heuristic was replaced with a read-boundary coercion in GetAllMediaSourcesForApiHandler + a dual-provider data migration nulling the historical 0001-01-01 sentinel rows, and the SPA reverted to a plain null check. Cold review caught that MediaSourceRepository still WRITES the sentinel today on disable-sync flows, so the read-coercion is the load-bearing permanent guard (a migration-only fix would have regressed). Live-E2E confirmed lastScan=null via the real API. Filed follow-ups: #458 (playlist/group rename should reject duplicate names) and #460 (normalize MediaSourceRepository writes to null). Notes: hit the decisions.md conflict treadmill twice — resolved via rebase + auto-merge (merge_when_checks_succeed); one decisions-guard flake (guard passed locally + on synthetic merge, 0 deletions) cleared on re-trigger.
Author
Owner

Session comment — 2026-07-19 (Claude Code orchestrator)

No backlog issue claimed — session pivoted to an operator-requested release + a tooling improvement.

1. Cut release v26.10.0 (unblocks #350 measurement)

Queue preflight surfaced #350 (cold-start instrumentation) as the top priority: medium pick, but its remaining optimization is data-gated: PR #446's instrumentation had never reached prod (:prod was the 2026-07-16 image, predating #446), so zero HLS cold-start samples existed anywhere. Operator chose to ship it to prod via a release.

  • Pre-tag gates: main green @ df8c5202; dotnet list package --vulnerable clean (15 projects); decisions.md had no superseded entries to consolidate (1944 lines, but the only "Superseded" hits are convention text).
  • Tag v26.10.0 pushed on df8c5202 → release build (run 850) green → :26.10.0 + :prod in registry. Release notes published (170 commits: Auto-Tune, WeightedShuffle distribution, health-check remediation, cold-start instrumentation, security hardening).
  • Security scan clean: ZAP DAST 0-FAIL / 118-PASS / 1 benign WARN (SPA catch-all content-type); semgrep SAST 0 findings across 4017 .cs.
  • ci-image run 849 "failure" = killed-during-cleanup (log ends mid buildx rm, no failure marker; the toolchain image built+pushed+smoke-tested green). Not real, not release-relevant.
  • Promotion deferred to Komodo Global Auto Update @ 03:00 (operator's choice) — same DeployStack + fail-closed backup/migration-smoke. #350 measurement is unblocked once prod picks up :26.10.0.

2. Improved the queue selector (PR #461)

Operator feedback: "rather than have the lore make us redo the selector's work, improve the selector." Added scripts/select-queue.sh — deterministic dependency exclusion / milestone-state tiering / priority-within-tier ordering / in-progress+parked+PR exclusion in CODE, leaving only CLAIM?/UMBRELLA? flags for judgment. Handoff kickoff + the three "cheap selector unreliable" lore bullets rewired to "run the script" (bullets kept as the evidence). Reproduces this session's hand-verified state (#395 correctly skipped as blocked-by open #381). shellcheck clean, fail-open. PR #461 open, CI running.

Queue state for next session

Arc empty (steady state). scripts/select-queue.sh top-5 = #350 (data-gated until prod update), #404 (SPA weight inputs — fully actionable, backend #402 merged), #412, #431, #444 — all priority: medium, deps clear. H12 audit clean (every open issue has a priority label).

## Session comment — 2026-07-19 (Claude Code orchestrator) **No backlog issue claimed** — session pivoted to an operator-requested release + a tooling improvement. ### 1. Cut release v26.10.0 (unblocks #350 measurement) Queue preflight surfaced #350 (cold-start instrumentation) as the top `priority: medium` pick, but its remaining optimization is **data-gated**: PR #446's instrumentation had never reached prod (`:prod` was the 2026-07-16 image, predating #446), so zero `HLS cold-start` samples existed anywhere. Operator chose to ship it to prod via a release. - Pre-tag gates: `main` green @ `df8c5202`; `dotnet list package --vulnerable` clean (15 projects); decisions.md had **no** superseded entries to consolidate (1944 lines, but the only "Superseded" hits are convention text). - Tag `v26.10.0` pushed on `df8c5202` → release build (run 850) green → `:26.10.0` + `:prod` in registry. Release notes published (170 commits: Auto-Tune, WeightedShuffle distribution, health-check remediation, cold-start instrumentation, security hardening). - Security scan clean: **ZAP DAST** 0-FAIL / 118-PASS / 1 benign WARN (SPA catch-all content-type); **semgrep SAST** 0 findings across 4017 `.cs`. - ci-image run 849 "failure" = **killed-during-cleanup** (log ends mid `buildx rm`, no failure marker; the toolchain image built+pushed+smoke-tested green). Not real, not release-relevant. - **Promotion deferred to Komodo Global Auto Update @ 03:00** (operator's choice) — same DeployStack + fail-closed backup/migration-smoke. #350 measurement is unblocked once prod picks up `:26.10.0`. ### 2. Improved the queue selector (PR #461) Operator feedback: "rather than have the lore make us redo the selector's work, improve the selector." Added `scripts/select-queue.sh` — deterministic dependency exclusion / milestone-state tiering / priority-within-tier ordering / in-progress+parked+PR exclusion in CODE, leaving only `CLAIM?`/`UMBRELLA?` flags for judgment. Handoff kickoff + the three "cheap selector unreliable" lore bullets rewired to "run the script" (bullets kept as the evidence). Reproduces this session's hand-verified state (#395 correctly skipped as blocked-by open #381). shellcheck clean, fail-open. **PR #461 open, CI running.** ### Queue state for next session Arc empty (steady state). `scripts/select-queue.sh` top-5 = **#350** (data-gated until prod update), **#404** (SPA weight inputs — fully actionable, backend #402 merged), #412, #431, #444 — all `priority: medium`, deps clear. H12 audit clean (every open issue has a priority label).
Author
Owner

Session comment — 2026-07-19 (Claude Code orchestrator)

Claimed + shipped: #404 — weighted-distribution SPA (PR #462, merged)

Selector surfaced #350 as the top priority: medium, but its remaining work is still data-gated (no channel tuned since the v26.10.0 auto-update, so zero HLS cold-start samples yet) — deferred, not claimable this session. Fell through to the next priority: medium: #404, the UI half of #70's weighted distribution (backend/API shipped in #402).

  • Multi-collection editor (/app/multi-collections): per-source weight input (1..1000, native stepper, clamp-on-blur), computed % share, "Reset to fair share"; weight round-trips through the draft so the replace-all PUT never silently resets it.
  • Classic schedule editor: WeightedShuffle offered only for MultiCollection sources; excluded from fillWithGroup like ShuffleInOrder. Input gained min/max/inputMode/onBlur/ariaLabel passthroughs (reusable by #425).
  • Verification: 861 web tests + lint + typecheck + prod build green; cold review MERGEABLE; live-E2E on a real instance (weighted create+read round-trips; rename-only PUT preserves weights; WeightedShuffle persists on a classic MultiCollection schedule item; block/playlist editors omit it). 3:1 ratio pinned by the existing PlayoutBuildGoldenTests.Classic_weighted.
  • Accuracy fix the E2E surfaced: the classic write path ACCEPTS WeightedShuffle on a plain Collection (degrades to fair-share) — so MultiCollection-only is a meaningfulness choice, not a write-path mirror; the rejection is on playlist/block paths. Corrected in itemRules.ts + decisions.md.
  • Docs: domain-model.md, spa-conventions.md (replace-all round-trip trap + bounded-numeric input pattern), decisions.md. Deferred: DesignSync push-back of the weight UI to MultiCollections.jsx (main-session-only, needs claude.ai approval).

Operator-reported backlog bugs → filed + one fixed

  • #463 — schedules "Active schedule" selector stretched full-width (missing fullWidth={false} → .ctv-field-full{width:100%} in a flex header, overlapping the title + buttons). Fixed this session (PR #465, merged): fullWidth={false} + bounded 150–260px + regression test; Playwright-headless screenshots at 1280/560px confirm no overlap. Triage: bug, priority: medium.
  • #464 — guide (and channels list) show generated fallback icons instead of real channel logos. Root cause: the guide DTO ChannelGuideChannelResponseModel has no logo field, the handler never populates one, and the SPA passes no src to ChannelLogo. Filed + queued (backend DTO + handler + SPA change; per operator, left for the normal queue). Triage: bug, priority: medium.

Housekeeping

  • H12 audit clean (every open issue carries a priority: label). No lingering in-progress labels.
  • Merged both PRs via the derived-consent gate (Done-when ticked + CI green + Review-verdict @ head).

Queue state (live-derived, not a forward guess)

Arc/gate/milestone/review tiers empty (steady state). Top priority: medium backlog now includes the two new bugs (#464, plus the pre-existing #350 data-gated, #412, #431, #444). Next session: run scripts/select-queue.sh and claim the top eligible.

## Session comment — 2026-07-19 (Claude Code orchestrator) ### Claimed + shipped: #404 — weighted-distribution SPA (PR #462, merged) Selector surfaced #350 as the top `priority: medium`, but its remaining work is still **data-gated** (no channel tuned since the v26.10.0 auto-update, so zero `HLS cold-start` samples yet) — deferred, not claimable this session. Fell through to the next `priority: medium`: **#404**, the UI half of #70's weighted distribution (backend/API shipped in #402). - Multi-collection editor (`/app/multi-collections`): per-source weight input (1..1000, native stepper, clamp-on-blur), computed % share, "Reset to fair share"; `weight` round-trips through the draft so the replace-all PUT never silently resets it. - Classic schedule editor: `WeightedShuffle` offered only for MultiCollection sources; excluded from `fillWithGroup` like `ShuffleInOrder`. `Input` gained `min`/`max`/`inputMode`/`onBlur`/`ariaLabel` passthroughs (reusable by #425). - Verification: 861 web tests + lint + typecheck + prod build green; cold review MERGEABLE; **live-E2E** on a real instance (weighted create+read round-trips; rename-only PUT preserves weights; WeightedShuffle persists on a classic MultiCollection schedule item; block/playlist editors omit it). 3:1 ratio pinned by the existing `PlayoutBuildGoldenTests.Classic_weighted`. - **Accuracy fix the E2E surfaced**: the classic write path ACCEPTS WeightedShuffle on a plain Collection (degrades to fair-share) — so MultiCollection-only is a *meaningfulness* choice, not a write-path mirror; the rejection is on playlist/block paths. Corrected in `itemRules.ts` + `decisions.md`. - Docs: `domain-model.md`, `spa-conventions.md` (replace-all round-trip trap + bounded-numeric input pattern), `decisions.md`. Deferred: DesignSync push-back of the weight UI to `MultiCollections.jsx` (main-session-only, needs claude.ai approval). ### Operator-reported backlog bugs → filed + one fixed - **#463** — schedules "Active schedule" selector stretched full-width (missing `fullWidth={false}` → `.ctv-field-full{width:100%}` in a flex header, overlapping the title + buttons). **Fixed this session (PR #465, merged)**: `fullWidth={false}` + bounded 150–260px + regression test; Playwright-headless screenshots at 1280/560px confirm no overlap. Triage: `bug`, `priority: medium`. - **#464** — guide (and channels list) show generated fallback icons instead of real channel logos. Root cause: the guide DTO `ChannelGuideChannelResponseModel` has no logo field, the handler never populates one, and the SPA passes no `src` to `ChannelLogo`. **Filed + queued** (backend DTO + handler + SPA change; per operator, left for the normal queue). Triage: `bug`, `priority: medium`. ### Housekeeping - H12 audit clean (every open issue carries a `priority:` label). No lingering `in-progress` labels. - Merged both PRs via the derived-consent gate (Done-when ticked + CI green + Review-verdict @ head). ### Queue state (live-derived, not a forward guess) Arc/gate/milestone/review tiers empty (steady state). Top `priority: medium` backlog now includes the two new bugs (#464, plus the pre-existing #350 data-gated, #412, #431, #444). Next session: run `scripts/select-queue.sh` and claim the top eligible.
Author
Owner

Session — Claude Code (Opus 4.8), 2026-07-19

Worked / landed: #431 (Backend: TTL-cache PerformHealthChecks) → PR #468 merged to main. 30s TTL cache of the full health-result list inside HealthCheckService; API GET poll reads the cache, startup + troubleshooting force fresh; GET /api/v1/health?refresh=true (additive) wired to the SPA "Refresh health" button. Read-path only, cold adversarial review MERGEABLE (no findings). Docs: decisions.md 2026-07-19 (#431), api-conventions §2, regenerated v1.json.

Selection (arc/gate/review/priority:high all empty — steady state): three priority: medium candidates surfaced (#412, #431, #444). Skipped #412 (CI peak-memory instrument) — it's the follow-up to #406 and needs live bumblebee measurement, colliding with the two active #406 worktrees in flight right now. Picked #431 (clean, self-contained, no shared-host contention). #444 (E2E lock flow) and #412 remain eligible for a later session.

New issue filed: #469 — CI: speed up the Formatting job (ci-cd, priority: low). Root cause: dotnet format --include still loads the whole-solution Roslyn workspace, so the job is slow and queued behind the long jobs; options = project-scope the format, dotnet format whitespace, cache obj/. Triage: low (dev-experience, not correctness).

CI notes for the next session:

  • decisions.md append-only flaked on the first run (be25df67) and passed on the fresh run (ca54853b) with an identical insertion-only diff — confirms it's the transient flake the operator flagged, not a real violation.
  • The Formatting gate caught legacy whitespace (not BOM) in GetTroubleshootingInfoHandler.cs, dragged in-scope by a one-line-comment touch (fix-as-you-touch #311). Fixed with a whitespace-only style commit (git diff -w empty). Lesson recorded: --verify the full touched-.cs set every push, not a subset; scope to the touched files' projects when the whole-.sln format is too slow to wait on.

Merge-gate: derived-consent auto-grant path (Done-when ticked + Review-verdict @ ca54853b + CI green) — no manual merge prompt needed.

### Session — Claude Code (Opus 4.8), 2026-07-19 **Worked / landed:** #431 (Backend: TTL-cache `PerformHealthChecks`) → **PR #468 merged** to `main`. 30s TTL cache of the full health-result list inside `HealthCheckService`; API GET poll reads the cache, startup + troubleshooting force fresh; `GET /api/v1/health?refresh=true` (additive) wired to the SPA "Refresh health" button. Read-path only, cold adversarial review MERGEABLE (no findings). Docs: decisions.md 2026-07-19 (#431), api-conventions §2, regenerated v1.json. **Selection (arc/gate/review/priority:high all empty — steady state):** three `priority: medium` candidates surfaced (#412, #431, #444). **Skipped #412** (CI peak-memory instrument) — it's the follow-up to #406 and needs live bumblebee measurement, colliding with the **two active `#406` worktrees** in flight right now. Picked #431 (clean, self-contained, no shared-host contention). #444 (E2E lock flow) and #412 remain eligible for a later session. **New issue filed:** **#469** — CI: speed up the Formatting job (`ci-cd`, `priority: low`). Root cause: `dotnet format --include` still loads the whole-solution Roslyn workspace, so the job is slow *and* queued behind the long jobs; options = project-scope the format, `dotnet format whitespace`, cache `obj/`. Triage: low (dev-experience, not correctness). **CI notes for the next session:** - `decisions.md append-only` **flaked** on the first run (`be25df67`) and **passed on the fresh run** (`ca54853b`) with an identical insertion-only diff — confirms it's the transient flake the operator flagged, not a real violation. - The `Formatting` gate caught **legacy whitespace** (not BOM) in `GetTroubleshootingInfoHandler.cs`, dragged in-scope by a one-line-comment touch (fix-as-you-touch #311). Fixed with a whitespace-only `style` commit (`git diff -w` empty). Lesson recorded: `--verify` the **full** touched-.cs set every push, not a subset; scope to the touched files' *projects* when the whole-`.sln` format is too slow to wait on. **Merge-gate:** derived-consent auto-grant path (Done-when ticked + `Review-verdict @ ca54853b` + CI green) — no manual merge prompt needed.
Author
Owner

Session — Claude Code (Opus 4.8), 2026-07-19

Worked / landed: #444 (functional-E2E: playout-build lock 409 + isLocked projection) → PR #470 merged to main (e8d73035). The last deferred lock-contention flow from #363. Added "Flow C" to scripts/e2e-functional.sh: seed a Classic Flood playout over a few short ffmpeg episodes, crank playout.days_to_build=5 so one build runs ~1s, poll GET /playouts/{id} until isLocked:true (the enqueue→dequeue race means the accepted trigger doesn't prove the lock is held), then assert PUT /playouts/{id} 409, reset 409, list-projection isLocked:true; post-build isLocked:false + PUT 200. Guarded so a mid-flight release → advisory skip, never a false red. Docs: e2e-local.md + ci-cd.md. CI green incl. Functional E2E; two cold reviews MERGEABLE @ 038703fe.

Selection (arc/gate/review/priority: high all empty — steady state): re-derived the contested priority: medium tier-5 live. Three siblings were #412, #431, #444. #431 already merged (PR #468); #350 remaining work is data-gated (no HLS cold-start samples since the v26.10.0 auto-update — still needs an operator-driven heavy tune-in); #412 deferred (CI peak-memory instrument — the 15:45 session flagged two active #406 worktrees measuring on the shared bumblebee host; didn't want to collide, per the standing #406 lesson). Picked #444 — clean, self-contained, no shared-host contention, explicitly left eligible by the 15:45 session.

Measurement note (for whoever revisits the scheduler/worker): a Classic Flood build's item count is window / item-duration, linear in playout.days_to_build, capped ~90 days (~777k items with 10s clips). Build time scales with item count (5d≈43k≈~1s, 30d≈259k≈2.5s on an M4). Going big is counter-productive for E2E: a 777k-item build enqueues 777k-row post-build gap/overlap jobs onto the single-consumer WorkerService, saturating it and starving subsequent builds (observed days=7 repeated builds intermittently returning items=0). Sized #444 at the lightest reliably-observable setting.

No new issues filed. H12 audit clean (every open issue carries a priority: label). No CI flakes this run — all 8 checks green first try (image build "skipped", expected for a non-main PR).

Queue remains: #412 (still shared-host-contended per the 15:45 note — recheck whether the #406 worktrees have wrapped), #350 (data-gated), then the priority: low backlog.

### Session — Claude Code (Opus 4.8), 2026-07-19 **Worked / landed:** **#444** (functional-E2E: playout-build lock 409 + `isLocked` projection) → **PR #470 merged** to `main` (`e8d73035`). The last deferred lock-contention flow from #363. Added "Flow C" to `scripts/e2e-functional.sh`: seed a Classic Flood playout over a few short ffmpeg episodes, crank `playout.days_to_build=5` so one build runs ~1s, **poll `GET /playouts/{id}` until `isLocked:true`** (the enqueue→dequeue race means the accepted trigger doesn't prove the lock is held), then assert `PUT /playouts/{id}` 409, reset 409, list-projection `isLocked:true`; post-build `isLocked:false` + `PUT` 200. Guarded so a mid-flight release → advisory skip, never a false red. Docs: `e2e-local.md` + `ci-cd.md`. CI green incl. `Functional E2E`; two cold reviews MERGEABLE @ `038703fe`. **Selection (arc/gate/review/`priority: high` all empty — steady state):** re-derived the contested `priority: medium` tier-5 live. Three siblings were #412, #431, #444. #431 already merged (PR #468); **#350** remaining work is data-gated (no `HLS cold-start` samples since the v26.10.0 auto-update — still needs an operator-driven heavy tune-in); **#412** deferred (CI peak-memory instrument — the 15:45 session flagged two active `#406` worktrees measuring on the shared bumblebee host; didn't want to collide, per the standing #406 lesson). Picked **#444** — clean, self-contained, no shared-host contention, explicitly left eligible by the 15:45 session. **Measurement note (for whoever revisits the scheduler/worker):** a Classic Flood build's item count is `window / item-duration`, linear in `playout.days_to_build`, capped ~90 days (~777k items with 10s clips). Build time scales with item count (5d≈43k≈~1s, 30d≈259k≈2.5s **on an M4**). **Going big is counter-productive** for E2E: a 777k-item build enqueues 777k-row post-build gap/overlap jobs onto the *single-consumer* WorkerService, saturating it and starving subsequent builds (observed days=7 repeated builds intermittently returning `items=0`). Sized #444 at the lightest reliably-observable setting. **No new issues filed.** H12 audit clean (every open issue carries a `priority:` label). No CI flakes this run — all 8 checks green first try (image build "skipped", expected for a non-`main` PR). **Queue remains:** #412 (still shared-host-contended per the 15:45 note — recheck whether the #406 worktrees have wrapped), #350 (data-gated), then the `priority: low` backlog.
Author
Owner

Session (Claude Code, Opus) — 2026-07-19

Worked: #469 (user-directed pickup, not queue selection) — CI Formatting-job speedup. CLOSED, merged PR #471.

  • Switched the format CI job + Husky pre-commit hook to dotnet format whitespace . --folder (skips the MSBuild/Roslyn workspace load; dropped restore/NuGet-cache steps). ~480s → ~0.5s, identical command. Coverage proven unchanged (whitespace + charset/BOM; full gate never enforced the style pass either). Docs: ci-cd.md + decisions.md.
  • Cold adversarial review: MERGEABLE. Sole CI red was the decisions.md append-only known flake (verified my diff is purely additive); operator reran that job → all green → merged.

New issues filed: none. H12 audit: nothing to label.

Recommended next: back to the priority: backlog cascade (arc/gate tiers empty as expected in maintenance mode).

### Session (Claude Code, Opus) — 2026-07-19 **Worked:** #469 (user-directed pickup, not queue selection) — CI Formatting-job speedup. **CLOSED**, merged PR #471. - Switched the `format` CI job + Husky pre-commit hook to `dotnet format whitespace . --folder` (skips the MSBuild/Roslyn workspace load; dropped restore/NuGet-cache steps). ~480s → ~0.5s, identical command. Coverage proven unchanged (whitespace + charset/BOM; full gate never enforced the style pass either). Docs: ci-cd.md + decisions.md. - Cold adversarial review: MERGEABLE. Sole CI red was the `decisions.md append-only` known flake (verified my diff is purely additive); operator reran that job → all green → merged. **New issues filed:** none. **H12 audit:** nothing to label. **Recommended next:** back to the `priority:` backlog cascade (arc/gate tiers empty as expected in maintenance mode).
Author
Owner

Session comment — 2026-07-19 (Claude Code orchestrator, Opus 4.8)

Claimed: #350 (HLS cold-start latency) — selected by scripts/select-queue.sh: lowest-numbered priority: medium with deps clear; both open-milestone candidates ineligible (#383 umbrella, #395 blocked-by open #381). Claim released; #350 stays OPEN.

Outcome: measurement-only session, no code change. Executed the measurement #350 has been blocked on since #446 shipped, and it falsified the issue's own hypotheses.

What happened

  • Prod was running a 2-day-stale image. The 03:00 Komodo Global Auto Update never pulled 26.10.0, and the container had been recreated from the cached image — so it looked freshly deployed and healthy while running pre-#446 code. Zero instrumentation lines read as "nobody tuned in," not "wrong binary."
  • I initially asserted the instrumentation was live, inferring it from registry tags + a fresh container start time. Both true, both irrelevant. Corrected it on the issue rather than editing quietly. The only check that catches this class: compare the local image config digest to the registry manifest's config digest.
  • Operator-authorized the deploy → took a verified DB backup + confirmed the 3 pending migrations first (the manual compose path bypasses the stack's fail-closed backup), deployed via the stack's own compose project. Prod healthy on 26.10.0.
  • Drove 36 cold-starts across 19 channels; excluded 3 samples where the channel was serving the offline filler rather than content (verified ffmpeg input paths per sample instead of assuming). N=33.

Result

startup (spawn → playlist) is 81% of cold-start and holds 100% of the variance (med 1579ms, max 8127ms); fill is pinned at ~200ms with segments 1/1 in all 33. Therefore:

  • pre-roll buffer would add tune-in latency, not remove it
  • conditional subtitle burn-in is falsified — burn-in samples were faster (1302 vs 1907ms median), and the slowest samples all had burn-in off
  • readrate burst targets the wrong phase entirely

Same channel varies up to 7.2× across repeat tunes with an identical filter graph → driver is per-tune (NFS input open/probe, or GPU contention), not per-profile.

Filed

issue prio what
#472 medium Split startup into spawn / input-open+probe / first-GOP — the real next step for #350
#473 high Tune-in hard-fails (ffmpeg exit 8) on missing media — 717 stale episode rows across 10 removed shows; broke ch100
#474 medium Music channels dead — Music Videos library has no LibraryPath, /data/music never scanned
server-management#630 — Komodo auto-update didn't pull :prod

Triage: #473 is priority: high — it is user-visible playback breakage, not latency, and it plausibly accounts for part of the original "slow to start / buffering" report that #350 was filed on. #472 is the correct pickup for anyone continuing #350; do not attempt the optimization before it lands.

H12 audit: clean — every open issue carries a priority: label. ✓

Lesson worth carrying

"Measure before optimizing" was already the rule here and the previous session correctly applied it. The gap was one level down: verify the measurement apparatus is actually running before trusting its silence. An empty log is evidence of nothing until you have confirmed the emitter is deployed — and registry-side tag checks do not establish that.

## Session comment — 2026-07-19 (Claude Code orchestrator, Opus 4.8) **Claimed:** #350 (HLS cold-start latency) — selected by `scripts/select-queue.sh`: lowest-numbered `priority: medium` with deps clear; both open-milestone candidates ineligible (#383 umbrella, #395 blocked-by open #381). Claim released; #350 stays OPEN. **Outcome:** measurement-only session, **no code change**. Executed the measurement #350 has been blocked on since #446 shipped, and it **falsified the issue's own hypotheses**. ### What happened - **Prod was running a 2-day-stale image.** The 03:00 Komodo Global Auto Update never pulled `26.10.0`, and the container had been recreated from the *cached* image — so it looked freshly deployed and healthy while running pre-#446 code. Zero instrumentation lines read as "nobody tuned in," not "wrong binary." - I initially asserted the instrumentation *was* live, inferring it from registry tags + a fresh container start time. Both true, both irrelevant. **Corrected it on the issue** rather than editing quietly. The only check that catches this class: compare the **local image config digest** to the registry manifest's config digest. - Operator-authorized the deploy → took a verified DB backup + confirmed the 3 pending migrations first (the manual compose path bypasses the stack's fail-closed backup), deployed via the stack's own compose project. Prod healthy on `26.10.0`. - Drove **36 cold-starts** across 19 channels; excluded 3 samples where the channel was serving the *offline filler* rather than content (verified ffmpeg input paths per sample instead of assuming). **N=33.** ### Result `startup` (spawn → playlist) is **81% of cold-start and holds 100% of the variance** (med 1579ms, max 8127ms); `fill` is pinned at ~200ms with `segments 1/1` in all 33. Therefore: - **pre-roll buffer** would *add* tune-in latency, not remove it - **conditional subtitle burn-in** is falsified — burn-in samples were *faster* (1302 vs 1907ms median), and the slowest samples all had burn-in off - **readrate burst** targets the wrong phase entirely Same channel varies up to **7.2×** across repeat tunes with an identical filter graph → driver is per-tune (NFS input open/probe, or GPU contention), not per-profile. ### Filed | issue | prio | what | |---|---|---| | #472 | medium | Split `startup` into spawn / input-open+probe / first-GOP — the real next step for #350 | | #473 | **high** | Tune-in hard-fails (ffmpeg exit 8) on missing media — 717 stale episode rows across 10 removed shows; broke ch100 | | #474 | medium | Music channels dead — `Music Videos` library has no LibraryPath, `/data/music` never scanned | | server-management#630 | — | Komodo auto-update didn't pull `:prod` | Triage: #473 is `priority: high` — it is user-visible playback breakage, not latency, and it plausibly accounts for part of the original "slow to start / buffering" report that #350 was filed on. #472 is the correct pickup for anyone continuing #350; **do not attempt the optimization before it lands.** **H12 audit:** clean — every open issue carries a `priority:` label. ✓ ### Lesson worth carrying "Measure before optimizing" was already the rule here and the previous session correctly applied it. The gap was one level down: **verify the measurement apparatus is actually running before trusting its silence.** An empty log is evidence of nothing until you have confirmed the emitter is deployed — and registry-side tag checks do not establish that.
Author
Owner

Session — Claude Code orchestrator (2026-07-19 eve)

Worked: #412 (CI: true peak-anon instrument) — DONE, merged PR #475 to main (all CI green; cold review MERGEABLE @ c5369b1d; all 4 Done-when ticked).

What shipped: scripts/ci-peak-anon.sh + two test-job steps sampling the cgroup anon high-water mark (headline), replacing the cache-inflated memory.peak reading. Validated on bumblebee. Docs: ci-cd.md rewritten + decisions.md entry.

Compiler-server A/B verdict (for server-management#604): swap-off sampled peak-anon, n=2 — OFF (CI config) ~5.84 GiB consistent; ON (defaults) 6.3–7.6 GiB + ~3 GiB resident VBCSCompiler. #406's "budget loosens well under 6 GiB" premise is NOT supported (OFF is already ~6 GiB for the build phase alone; the test job adds test+coverage). Size the cap off the now-live test-job peak-anon on the next main run, not the build-only A/B.

Selection note (for the next session): the cheap selector ranked #350 above #412, but I skipped it: #350's remaining work (data-driven cold-start optimization) was gated on prod cold-start measurements, and at claim time the running :prod container was still on the 2026-07-16 image (pre-#446), so no data existed. That gate has since cleared — prod was redeployed to v26.10.0 ~mid-session (image built 2026-07-18, container recreated ~18:15) and 9 HLS cold-start log lines are now present on prod. So #350 is workable now: grep prod logs for HLS cold-start to get the real setup/startup/fill split + subtitleBurnIn/hwaccel flags, then land the targeted transcode-pipeline fix. Recommend it as a strong next pickup (priority: medium, deps clear).

Housekeeping: no new issues filed → H12 audit clean (all open issues carry a priority: label). #412 closed, in-progress removed. Measurement container torn down; host healthy throughout (12–13 GiB available, prod healthy).

### Session — Claude Code orchestrator (2026-07-19 eve) **Worked:** #412 (CI: true peak-anon instrument) — **DONE, merged PR #475** to `main` (all CI green; cold review MERGEABLE @ c5369b1d; all 4 Done-when ticked). **What shipped:** `scripts/ci-peak-anon.sh` + two `test`-job steps sampling the cgroup `anon` high-water mark (headline), replacing the cache-inflated `memory.peak` reading. Validated on bumblebee. Docs: `ci-cd.md` rewritten + `decisions.md` entry. **Compiler-server A/B verdict (for server-management#604):** swap-off sampled peak-anon, n=2 — OFF (CI config) ~5.84 GiB consistent; ON (defaults) 6.3–7.6 GiB + ~3 GiB resident VBCSCompiler. **#406's "budget loosens well under 6 GiB" premise is NOT supported** (OFF is already ~6 GiB for the build phase alone; the test job adds test+coverage). Size the cap off the now-live test-job peak-anon on the next `main` run, not the build-only A/B. **Selection note (for the next session):** the cheap selector ranked **#350** above #412, but I skipped it: #350's remaining work (data-driven cold-start optimization) was gated on prod cold-start measurements, and at claim time the running `:prod` container was still on the **2026-07-16 image** (pre-#446), so no data existed. **That gate has since cleared** — prod was redeployed to **v26.10.0** ~mid-session (image built 2026-07-18, container recreated ~18:15) and **9 `HLS cold-start` log lines are now present** on prod. So **#350 is workable now**: grep prod logs for `HLS cold-start` to get the real `setup/startup/fill` split + `subtitleBurnIn`/`hwaccel` flags, then land the targeted transcode-pipeline fix. Recommend it as a strong next pickup (priority: medium, deps clear). **Housekeeping:** no new issues filed → H12 audit clean (all open issues carry a `priority:` label). #412 closed, `in-progress` removed. Measurement container torn down; host healthy throughout (12–13 GiB available, prod healthy).
Author
Owner

Session (Claude Code / Opus 4.8, orchestrator) — 2026-07-19

Claimed & shipped: #476 (scanner: FileNotFound doesn't cascade removed show → seasons/episodes), priority: high. Selected via scripts/select-queue.sh (top of priority: high tier; #395 correctly skipped as blocked-by open #381). Not bundled with #478 (its high sibling) — the bodies explicitly separate the classes and share no code surface.

Outcome: MERGED — PR #481 → main 0d704255. CI fully green; cold-context adversarial review MERGEABLE (no correctness/safety bugs); 6/6 Done-when.

  • Root cause: season/episode sweeps live inside the per-parent scan loops, which only visit parents present in the incoming server list → a removed show/season is never visited → descendants keep RemoteOnly (which PlayoutBuilder doesn't skip). Also fixed the sibling gap (removed season, show present).
  • Fix: two provider-agnostic cascade repo helpers keyed on base-table Season.ShowId/Episode.SeasonId, driven by the scanner. Shared abstract base → Jellyfin/Plex/Emby fixed at once; Movie/OtherVideo flat (no gap).
  • Tests: Scanner.Tests (orchestration, non-vacuous) + Integration (real cascade SQL, sibling-untouched + empty-input negative controls).
  • Deferred: prod rescan of the 717 stale rows = post-deploy operator acceptance (ersatztv-test after :latest, prod after a version tag). Mechanism proven by the integration test.

Docs: none required (scanner bug fix; no convention/route/API/decision change). H12 audit: clean. New issues filed: none.

Recommended next (verify live, don't trust this line): top of priority: high is now #478 (channels 102/107 VAAPI hwupload -22 tune failures) — but note it may be a host/driver issue belonging in server-management, so scope it before committing to a code fix; then the priority: medium tier (#350, #472, #474).

### Session (Claude Code / Opus 4.8, orchestrator) — 2026-07-19 **Claimed & shipped: #476** (scanner: `FileNotFound` doesn't cascade removed show → seasons/episodes), `priority: high`. Selected via `scripts/select-queue.sh` (top of `priority: high` tier; #395 correctly skipped as blocked-by open #381). Not bundled with #478 (its high sibling) — the bodies explicitly separate the classes and share no code surface. **Outcome: MERGED** — PR #481 → main `0d704255`. CI fully green; cold-context adversarial review **MERGEABLE** (no correctness/safety bugs); 6/6 Done-when. - **Root cause**: season/episode sweeps live inside the per-parent scan loops, which only visit parents present in the incoming server list → a removed show/season is never visited → descendants keep `RemoteOnly` (which PlayoutBuilder doesn't skip). Also fixed the sibling gap (removed season, show present). - **Fix**: two provider-agnostic cascade repo helpers keyed on base-table `Season.ShowId`/`Episode.SeasonId`, driven by the scanner. Shared abstract base → Jellyfin/Plex/Emby fixed at once; Movie/OtherVideo flat (no gap). - **Tests**: Scanner.Tests (orchestration, non-vacuous) + Integration (real cascade SQL, sibling-untouched + empty-input negative controls). - **Deferred**: prod rescan of the 717 stale rows = post-deploy operator acceptance (ersatztv-test after `:latest`, prod after a version tag). Mechanism proven by the integration test. **Docs**: none required (scanner bug fix; no convention/route/API/decision change). **H12 audit**: clean. **New issues filed**: none. **Recommended next** (verify live, don't trust this line): top of `priority: high` is now **#478** (channels 102/107 VAAPI hwupload -22 tune failures) — but note it may be a host/driver issue belonging in server-management, so scope it before committing to a code fix; then the `priority: medium` tier (#350, #472, #474).
Author
Owner

Session 2026-07-19 (Claude Code orchestrator, Opus)

Claimed & worked: #478 (priority: high, bug) — channels 102/107 hwupload -22 / exit 234 on prod. Selected via cheap selector (top of priority: high after #473 closed); arc/gate/review tiers empty as expected.

Outcome: CLOSED — diagnosed as a self-resolved host VAAPI transient, not an ErsatzTV bug.

  • Captured the exact prod ffmpeg argv from /proc and reproduced it verbatim on the real GPU (standalone, faithfully with seek+logo+hls, 4× concurrent, and an 8-channel burst) — all succeed. ETV's generated filtergraph is correct.
  • Root cause: transient VAAPI surface-allocation failure on the freshly re-initialized phoenix iGPU render node during the post-deploy tune burst (container recreated 20:15:11 → ~15 channels cold-started at once). Hit only the 10-bit channels (larger p010 surfaces); 8 failures 20:17–20:27; self-resolved by ~20:28. Verified 102/107 tune cleanly now, incl. concurrently.
  • No code change; no adversarial review (empty diff, host-side disposition — stated exemption on the issue).

Filed: timothy/server-management#631 — host-side follow-up (verify whether the post-recreate VAAPI transient reproduces on the next ersatztv container recreate; candidate mitigations: DRM fd handoff, GPU warm-up/readiness gate, stagger post-restart cold-starts). No new ErsatzTV issues.

Deferred (not filed, by operator decision): optional ETV resilience enhancement — software-encode fallback after N consecutive VAAPI hwupload failures.

Housekeeping: H12 audit clean (all open issues carry a priority: label). Worktree removed (no commits). Note for future selectors: the Gitea milestone milestones= name filter still no-ops on :/+ names — filter locally.

### Session 2026-07-19 (Claude Code orchestrator, Opus) **Claimed & worked:** #478 (`priority: high`, bug) — channels 102/107 `hwupload -22` / exit 234 on prod. Selected via cheap selector (top of `priority: high` after #473 closed); arc/gate/review tiers empty as expected. **Outcome: CLOSED — diagnosed as a self-resolved host VAAPI transient, not an ErsatzTV bug.** - Captured the exact prod ffmpeg argv from `/proc` and reproduced it **verbatim** on the real GPU (standalone, faithfully with seek+logo+hls, 4× concurrent, and an 8-channel burst) — all succeed. ETV's generated filtergraph is correct. - Root cause: transient VAAPI surface-allocation failure on the freshly re-initialized phoenix iGPU render node during the post-deploy tune burst (container recreated 20:15:11 → ~15 channels cold-started at once). Hit only the 10-bit channels (larger `p010` surfaces); 8 failures 20:17–20:27; self-resolved by ~20:28. Verified 102/107 tune cleanly now, incl. concurrently. - No code change; no adversarial review (empty diff, host-side disposition — stated exemption on the issue). **Filed:** `timothy/server-management#631` — host-side follow-up (verify whether the post-recreate VAAPI transient reproduces on the next `ersatztv` container recreate; candidate mitigations: DRM fd handoff, GPU warm-up/readiness gate, stagger post-restart cold-starts). No new ErsatzTV issues. **Deferred (not filed, by operator decision):** optional ETV resilience enhancement — software-encode fallback after N consecutive VAAPI `hwupload` failures. **Housekeeping:** H12 audit clean (all open issues carry a `priority:` label). Worktree removed (no commits). Note for future selectors: the Gitea milestone `milestones=` name filter still no-ops on `:`/`+` names — filter locally.
Author
Owner

Session — 2026-07-19 (Claude Code orchestrator, Opus 4.8)

Selector (scripts/select-queue.sh): shortlist #350, #474, #477, #480, #22. Resolved the flags:

  • #350 skipped — worked earlier today; its actionable successor #472 (split the HLS startup instrumentation) is already in-progress. #350 is now a tracking issue awaiting #472 data ("don't optimize yet").
  • #474 surfaced to operator — primary fix is a prod config/content change (add /data/music LibraryPath), explicitly deferred by its author to operator/media-management scope. Not a code-session pickup.

Claimed + shipped: #477 (data-loss-adjacent scanner anti-nuke guard). Merged via PR #483 → main @ 55be64bd.

  • Shared MediaServerReconciliationGuard: skips + logs the file-not-found sweep when a successful fetch returns zero items while items exist (was nuking the whole library → EmptyTrash permanent delete).
  • Deliberately overrides #476's degenerate empty-incoming-flags case (documented in decisions.md); #476 cascade preserved for partial deletions, its test rewritten accordingly.
  • Independent cold-context adversarial review: MERGEABLE. 1481 scanner tests green; non-vacuity proven by negative control. CI green.

Filed this session (both priority: low, triaged):

  • #484 — deferred ratio-threshold + projection-failure detection for the scanner guard (cross-layer; not urgent).
  • #485 — .gitignore core entry silently ignores new files under any Core/ dir on macOS (bit this PR; force-added). Build-hygiene bug.

H12 audit: clean (all open issues carry priority: labels).

Recommended next (verify live): open priority: medium backlog — #474 (needs operator config first), #480 (external-JSON unprobed stream, fresh #479 context), #477's follow-up #484. Deterministic selector will re-rank.

### Session — 2026-07-19 (Claude Code orchestrator, Opus 4.8) **Selector (`scripts/select-queue.sh`):** shortlist #350, #474, #477, #480, #22. Resolved the flags: - **#350** skipped — worked earlier today; its actionable successor **#472** (split the HLS `startup` instrumentation) is already `in-progress`. #350 is now a tracking issue awaiting #472 data ("don't optimize yet"). - **#474** surfaced to operator — primary fix is a prod config/content change (add `/data/music` LibraryPath), explicitly deferred by its author to operator/media-management scope. Not a code-session pickup. **Claimed + shipped: #477** (data-loss-adjacent scanner anti-nuke guard). Merged via **PR #483** → `main` @ 55be64bd. - Shared `MediaServerReconciliationGuard`: skips + logs the file-not-found sweep when a successful fetch returns zero items while items exist (was nuking the whole library → EmptyTrash permanent delete). - Deliberately overrides #476's degenerate empty-incoming-flags case (documented in decisions.md); #476 cascade preserved for partial deletions, its test rewritten accordingly. - Independent cold-context adversarial review: **MERGEABLE**. 1481 scanner tests green; non-vacuity proven by negative control. CI green. **Filed this session (both `priority: low`, triaged):** - **#484** — deferred ratio-threshold + projection-failure detection for the scanner guard (cross-layer; not urgent). - **#485** — `.gitignore` `core` entry silently ignores new files under any `Core/` dir on macOS (bit this PR; force-added). Build-hygiene bug. **H12 audit:** clean (all open issues carry `priority:` labels). **Recommended next (verify live):** open `priority: medium` backlog — #474 (needs operator config first), #480 (external-JSON unprobed stream, fresh #479 context), #477's follow-up #484. Deterministic selector will re-rank.
Author
Owner

Session 2026-07-19 (Claude Code orchestrator) — #472 shipped

Picked up: #472 (Split HLS cold-start startup phase). Selector (scripts/select-queue.sh) ranked #350 top of priority: medium, but #350's own wrap comment names #472 as "the actual next step" — so #350 is the container and #472 was the pickup. Verified live: unclaimed, no comments, deps clear.

Landed: PR #482, merged, CI green @ 37674d65. Log-only telemetry; no API/SPA/route/convention change, so no doc updates were owed. H12 audit clean (every open issue carries a priority: label).

Worth carrying forward

1. A ## Done-when box can encode a circular dependency, and the merge gate will then deny forever. #472's last box was "fresh samples collected on prod" — unsatisfiable before merge, since prod only runs the code after a release. I reworded the issue into pre-merge criteria + a post-release item owned by the parent (#350), with the rationale recorded in the body. Any future instrumentation/measurement issue should split its checklist the same way — otherwise the gate is structurally unsatisfiable and the temptation is to tick a box for work not done.

2. Re-reviewing the fix commit paid off twice, and both catches were the same class. Four cold adversarial passes, each over code the previous hadn't seen. Passes 2 and 3 each found a comment whose stated rationale didn't match the mechanism — and each of those had landed inside the fix for the previous one. Specifically: I justified a stale-playlist guard with "Run warns but doesn't delete", when FolderMustBeEmpty actually wipes the folder before the worker spawns (the real path is EmptyFolder swallowing a failure into a warning); and I named a test for precedence over ThreeWay, a case impossible by construction. On an instrumentation change this is not a nit — a wrong rationale produces confident misreadings of the data later. I verified both reviewer claims against source myself rather than accepting them.

3. CI cost, stated plainly. Three runs. The second push superseded a live run I could not cancel (REST/MCP cancel still 404 on 1.25.4) — it drained holding a slot. The third waited for green before landing, so nothing was orphaned. Lesson reinforced: when a fix is comment-only, wait for the current run to settle before pushing — it costs wall-clock, not a runner slot.

4. The instrumentation does NOT answer #472's headline question, and the close comments say so. -loglevel error -nostats means a healthy FFmpeg is silent on stderr, so input-open/probe can't be separated from encoder-init without changing the FFmpeg command. ffmpegInit still contains both rival hypotheses; only the ErsatzTV-prep vs FFmpeg-time boundary is new. #472 explicitly accepted that two-way outcome. Flagged prominently on both #472 and #350 so nobody reads a large ffmpegInit as evidence for either candidate.

State

  • #472 closed, in-progress removed, worktree cleaned.
  • #350 stays open — updated with what the new data will and won't answer, plus a reminder to verify the deployed image by digest before trusting samples or an empty grep (that trap cost the last session real time).
  • No new issues filed.

Deferred

Prod sampling (user decision this session): needs a release carrying #472 to reach prod. It remains the gate on any cold-start optimization — the three originally-planned optimizations are still falsified, and nothing should be built off the old #350 numbers.

Recommended next

Nothing arc-level. Falls through to the priority: cascade — re-derive live; don't trust this line.

## Session 2026-07-19 (Claude Code orchestrator) — #472 shipped **Picked up:** #472 (Split HLS cold-start `startup` phase). Selector (`scripts/select-queue.sh`) ranked #350 top of `priority: medium`, but #350's own wrap comment names #472 as "the actual next step" — so #350 is the container and #472 was the pickup. Verified live: unclaimed, no comments, deps clear. **Landed:** PR #482, merged, CI green @ `37674d65`. Log-only telemetry; no API/SPA/route/convention change, so no doc updates were owed. H12 audit clean (every open issue carries a `priority:` label). ### Worth carrying forward **1. A `## Done-when` box can encode a circular dependency, and the merge gate will then deny forever.** #472's last box was "fresh samples collected on prod" — unsatisfiable before merge, since prod only runs the code after a release. I reworded the issue into **pre-merge criteria** + a **post-release item owned by the parent (#350)**, with the rationale recorded in the body. *Any future instrumentation/measurement issue should split its checklist the same way* — otherwise the gate is structurally unsatisfiable and the temptation is to tick a box for work not done. **2. Re-reviewing the fix commit paid off twice, and both catches were the same class.** Four cold adversarial passes, each over code the previous hadn't seen. Passes 2 and 3 each found a **comment whose stated rationale didn't match the mechanism** — and each of those had landed *inside the fix for the previous one*. Specifically: I justified a stale-playlist guard with "Run warns but doesn't delete", when `FolderMustBeEmpty` actually wipes the folder before the worker spawns (the real path is `EmptyFolder` swallowing a failure into a warning); and I named a test for precedence over `ThreeWay`, a case impossible by construction. **On an instrumentation change this is not a nit** — a wrong rationale produces confident misreadings of the data later. I verified both reviewer claims against source myself rather than accepting them. **3. CI cost, stated plainly.** Three runs. The second push superseded a live run I could not cancel (REST/MCP cancel still 404 on 1.25.4) — it drained holding a slot. The third waited for green before landing, so nothing was orphaned. Lesson reinforced: when a fix is comment-only, **wait for the current run to settle before pushing** — it costs wall-clock, not a runner slot. **4. The instrumentation does NOT answer #472's headline question, and the close comments say so.** `-loglevel error -nostats` means a healthy FFmpeg is silent on stderr, so input-open/probe can't be separated from encoder-init without changing the FFmpeg command. `ffmpegInit` still contains both rival hypotheses; only the ErsatzTV-prep vs FFmpeg-time boundary is new. #472 explicitly accepted that two-way outcome. Flagged prominently on both #472 and #350 so nobody reads a large `ffmpegInit` as evidence for either candidate. ### State - #472 **closed**, `in-progress` removed, worktree cleaned. - #350 stays **open** — updated with what the new data will and won't answer, plus a reminder to verify the deployed image **by digest** before trusting samples or an empty `grep` (that trap cost the last session real time). - **No new issues filed.** ### Deferred Prod sampling (user decision this session): needs a release carrying #472 to reach prod. It remains the gate on any cold-start optimization — the three originally-planned optimizations are still falsified, and nothing should be built off the old #350 numbers. ### Recommended next Nothing arc-level. Falls through to the `priority:` cascade — re-derive live; don't trust this line.
Author
Owner

Session — 2026-07-20 (Claude Code, Opus 4.8 orchestrator)

Worked: #480 — external-JSON playout channels handed ffmpeg an unprobed remote-stream URL (#473 class survived there). Merged: PR #486 (CI green @ 4a9502cb). #480 CLOSED.

Selection (maintenance mode, no issue named): cheap selector ranked #395 as WINNER with deps=clear — WRONG. GET /issues/395/dependencies shows it is blocked by open #381 (its own Done-when confirms: "Blocked-by #381 landed — green before AND after"). Re-deriving the whole tier myself (per the standing selector-distrust lesson) also confirmed the milestone tier had only #395 (blocked), review/priority:high tiers empty. Fell through to priority:medium: #350 (deploy-gated — its next step needs #472 released+deployed to prod then re-measured, not a code PR), #474 (self-declared config-not-code, operator lane), #480 (clean code bug). Operator picked #480.

Selector-hardening note for next session: the packet's deps field remains untrustworthy — verify GET /issues/{n}/dependencies on every candidate you'd claim, and read the issue's own Done-when for a Blocked-by line. #395's title literally said "gated on #381 goldens" yet the packet said clear.

Review: independent cold-context adversarial review → MERGEABLE (control flow, cancellation, DI, 1-min-clamp claim all verified; 2 cosmetic nits, non-blocking). Review-verdict: MERGEABLE @ 4a9502cb posted on PR.

Docs: docs/decisions.md #480 entry (append-only; closes the #473 scope gap by cross-reference).

Filed: none. H12 audit: clean (every open issue has a priority label).

Still-blocked / not-picked, for visibility: #395 (blocked by open #381), #350 (deploy-gated on #472 reaching prod — an operator/infra step), #474 (config-not-code; add /data/music LibraryPath + scan on prod — server-management lane).

### Session — 2026-07-20 (Claude Code, Opus 4.8 orchestrator) **Worked:** #480 — external-JSON playout channels handed ffmpeg an unprobed remote-stream URL (#473 class survived there). **Merged: PR #486** (CI green @ `4a9502cb`). #480 CLOSED. **Selection (maintenance mode, no issue named):** cheap selector ranked **#395 as WINNER with `deps=clear` — WRONG**. `GET /issues/395/dependencies` shows it is **blocked by open #381** (its own Done-when confirms: "Blocked-by #381 landed — green before AND after"). Re-deriving the whole tier myself (per the standing selector-distrust lesson) also confirmed the milestone tier had only #395 (blocked), review/priority:high tiers empty. Fell through to priority:medium: #350 (deploy-gated — its next step needs #472 released+deployed to prod then re-measured, not a code PR), #474 (self-declared config-not-code, operator lane), #480 (clean code bug). Operator picked #480. **Selector-hardening note for next session:** the packet's `deps` field remains untrustworthy — verify `GET /issues/{n}/dependencies` on every candidate you'd claim, and read the issue's own Done-when for a `Blocked-by` line. #395's title literally said "gated on #381 goldens" yet the packet said clear. **Review:** independent cold-context adversarial review → MERGEABLE (control flow, cancellation, DI, 1-min-clamp claim all verified; 2 cosmetic nits, non-blocking). `Review-verdict: MERGEABLE @ 4a9502cb` posted on PR. **Docs:** `docs/decisions.md` #480 entry (append-only; closes the #473 scope gap by cross-reference). **Filed:** none. **H12 audit:** clean (every open issue has a priority label). **Still-blocked / not-picked, for visibility:** #395 (blocked by open #381), #350 (deploy-gated on #472 reaching prod — an operator/infra step), #474 (config-not-code; add `/data/music` LibraryPath + scan on prod — server-management lane).
Author
Owner

Session — #488 (Claude Code / Opus 4.8 orchestrator)

Claimed & closed: #488 — JellyfinMusicVideoLibraryScanner crashed with ArgumentNullException on the first item of every music-video scan. Selector (scripts/select-queue.sh) ranked it top of priority: high, deps clear.

Fix (PR #492, merged a696f06d): LibraryRepository.GetOrAddFolder now resolves the folder from the DB by (LibraryPathId, Path) instead of the caller's LibraryPath.LibraryFolders nav collection, which is null on the Jellyfin remote path. Removes the implicit eager-load contract for all nine callers; documented on the interface; docs/decisions.md entry. Backend-only (no API/SPA/migration).

Tests (both non-vacuous — reverting reproduces the ArgumentNullException): LibraryRepositoryTests (repo-level: null-nav create + idempotent re-scan + ParentId persist) and JellyfinMusicVideoLibraryScannerTests (end-to-end scan with the real repos + in-memory SQLite, asserting Artist + MusicVideo rows — a deliberate deviation from the mock-everything house style, since a mocked repo can't exhibit this bug).

Review: cold-context adversarial → MERGEABLE, no Blocker/High/Medium. All 6 Done-when boxes ticked with evidence; H10 verdict posted at head.

CI note: the first run's Build & test (.NET) went red on the pre-existing flaky SPA test LibrariesScreen.test.tsx:287 (#447 family) — unrelated (0 web/ files changed; EF-migration + Functional-E2E both green). Cleared by operator selective job rerun. The #489 session has bumped that test's waitFor to 10s on feat/489-jellyfin-mixed-libraries (reaches main with #493).

New issues filed (triage verdicts):

  • #491 priority: low — LibraryFolder has no unique index on (LibraryPathId, Path); pre-existing concurrent double-insert window (not a regression). Low: only reachable under parallel scans of one library, which the scheduler doesn't do today.
  • #494 priority: medium — JellyfinMusicVideoLibraryScanner does no reconciliation (items removed from Jellyfin never removed from ErsatzTV). Medium: real data-correctness gap, latent until this path started running. #489 session added SQL-level evidence that no cross-delete exists today but implementing the fix creates the cross-kind-delete risk.

Unblocks: #474 (+ transitively #487); #489/PR #493 (its Mixed arm composes this scanner) can now validate end-to-end. Operator: prod library 15 sync can be re-enabled (shouldSyncItems: true).

H12 audit: ✓ all open issues carry a priority: label.

### Session — #488 (Claude Code / Opus 4.8 orchestrator) **Claimed & closed: #488** — `JellyfinMusicVideoLibraryScanner` crashed with `ArgumentNullException` on the first item of every music-video scan. Selector (`scripts/select-queue.sh`) ranked it top of `priority: high`, deps clear. **Fix** (PR #492, merged `a696f06d`): `LibraryRepository.GetOrAddFolder` now resolves the folder from the DB by `(LibraryPathId, Path)` instead of the caller's `LibraryPath.LibraryFolders` nav collection, which is `null` on the Jellyfin remote path. Removes the implicit eager-load contract for all nine callers; documented on the interface; `docs/decisions.md` entry. Backend-only (no API/SPA/migration). **Tests** (both non-vacuous — reverting reproduces the `ArgumentNullException`): `LibraryRepositoryTests` (repo-level: null-nav create + idempotent re-scan + ParentId persist) and `JellyfinMusicVideoLibraryScannerTests` (end-to-end scan with the **real** repos + in-memory SQLite, asserting Artist + MusicVideo rows — a deliberate deviation from the mock-everything house style, since a mocked repo can't exhibit this bug). **Review**: cold-context adversarial → MERGEABLE, no Blocker/High/Medium. All 6 Done-when boxes ticked with evidence; H10 verdict posted at head. **CI note**: the first run's `Build & test (.NET)` went red on the pre-existing flaky SPA test `LibrariesScreen.test.tsx:287` (#447 family) — unrelated (0 `web/` files changed; EF-migration + Functional-E2E both green). Cleared by operator selective job rerun. The #489 session has bumped that test's `waitFor` to 10s on `feat/489-jellyfin-mixed-libraries` (reaches main with #493). **New issues filed (triage verdicts):** - **#491** `priority: low` — `LibraryFolder` has no unique index on `(LibraryPathId, Path)`; pre-existing concurrent double-insert window (not a regression). Low: only reachable under parallel scans of one library, which the scheduler doesn't do today. - **#494** `priority: medium` — `JellyfinMusicVideoLibraryScanner` does no reconciliation (items removed from Jellyfin never removed from ErsatzTV). Medium: real data-correctness gap, latent until this path started running. #489 session added SQL-level evidence that no cross-delete exists today but implementing the fix *creates* the cross-kind-delete risk. **Unblocks**: #474 (+ transitively #487); #489/PR #493 (its Mixed arm composes this scanner) can now validate end-to-end. Operator: prod library 15 sync can be re-enabled (`shouldSyncItems: true`). H12 audit: ✓ all open issues carry a `priority:` label.
Author
Owner

Session — 2026-07-20 (Claude Code orchestrator, Opus 4.8)

Picked: #494 (priority: medium backlog) via scripts/select-queue.sh. Skipped #350 (user-directed; blocked on a prod deploy of the #482 sub-split anyway). #494's selector row carried a CLAIM? flag → resolved to historical claims only (no live claim).

Issue Outcome PR Notes
#494 CLOSED — merged #495 (squash fdef9089) JellyfinMusicVideoLibraryScanner add/update-only → added library-scoped reconciliation sweep
#496 FILED (priority: low) — Follow-up: per-library music-video identity (global GetOrAdd path dedup); the issue's deferred "option 2"

#494 in one line: the Jellyfin music-video scanner never removed items deleted upstream; added TrashMissingMusicVideos (diff incoming local paths vs FindMusicVideoPaths, hard-delete missing + clean empty artists), gated by the #477 empty-fetch guard, cross-kind safe by construction (queries join the concrete MusicVideo table). decisions.md records the identity choice.

Coordination: file-disjoint from the in-flight #489/PR #493 (mixed libraries) — verified #493 touches the dispatch/wiring, not JellyfinMusicVideoLibraryScanner.cs. Incorporated the #493 session's four warnings (esp. the global-GetOrAdd identity trap and the mixed-library empty-incoming steady state — pinned by a dedicated test).

Verification: TDD (RED 3/5 pre-fix) → GREEN 6/6; full pass Scanner.Tests 1481 + ErsatzTV.Tests 1748, 0 failures. Cold-context adversarial review MERGEABLE (no Blocker/High). CI green on fdef9089. #311 BOM strip on the scanner file (inherited).

CI note: the orphaned run on the first push (2351841f, superseded by the review-response push) showed a decisions.md append-only red — the known flake; the fresh run on fdef9089 passed it green. No action taken on the orphan.

H12 audit: clean — every open issue carries a priority: label.

Triage of new issue: #496 — priority: low (latent; requires overlapping-path multi-libraries + owner drops a shared file; documented limitation, not a regression).

Recommended next: back to the priority: cascade (arc/milestones normally empty in maintenance mode).

## Session — 2026-07-20 (Claude Code orchestrator, Opus 4.8) **Picked:** #494 (`priority: medium` backlog) via `scripts/select-queue.sh`. Skipped #350 (user-directed; blocked on a prod deploy of the #482 sub-split anyway). #494's selector row carried a `CLAIM?` flag → resolved to historical claims only (no live claim). | Issue | Outcome | PR | Notes | |---|---|---|---| | #494 | **CLOSED** — merged | #495 (squash `fdef9089`) | `JellyfinMusicVideoLibraryScanner` add/update-only → added library-scoped reconciliation sweep | | #496 | **FILED** (`priority: low`) | — | Follow-up: per-library music-video identity (global `GetOrAdd` path dedup); the issue's deferred "option 2" | **#494 in one line:** the Jellyfin music-video scanner never removed items deleted upstream; added `TrashMissingMusicVideos` (diff incoming local paths vs `FindMusicVideoPaths`, hard-delete missing + clean empty artists), gated by the #477 empty-fetch guard, cross-*kind* safe by construction (queries join the concrete `MusicVideo` table). decisions.md records the identity choice. **Coordination:** file-disjoint from the in-flight #489/PR #493 (mixed libraries) — verified #493 touches the dispatch/wiring, not `JellyfinMusicVideoLibraryScanner.cs`. Incorporated the #493 session's four warnings (esp. the global-`GetOrAdd` identity trap and the mixed-library empty-incoming steady state — pinned by a dedicated test). **Verification:** TDD (RED 3/5 pre-fix) → GREEN 6/6; full pass Scanner.Tests 1481 + ErsatzTV.Tests 1748, 0 failures. Cold-context adversarial review MERGEABLE (no Blocker/High). CI green on `fdef9089`. #311 BOM strip on the scanner file (inherited). **CI note:** the orphaned run on the first push (`2351841f`, superseded by the review-response push) showed a `decisions.md append-only` red — the known flake; the fresh run on `fdef9089` passed it green. No action taken on the orphan. **H12 audit:** clean — every open issue carries a `priority:` label. **Triage of new issue:** #496 — `priority: low` (latent; requires overlapping-path multi-libraries + owner drops a shared file; documented limitation, not a regression). **Recommended next:** back to the `priority:` cascade (arc/milestones normally empty in maintenance mode).
Author
Owner

Session — 2026-07-20 (Claude/Opus, backlog)

Worked: #497 (Jellyfin music-video metadata reconcile) — DONE, merged #499 (c4ad4a73), issue closed.

  • Fix: JellyfinMusicVideoLibraryScanner.UpdateMetadata now reconciles Genres/Tags/Studios/Artists on the update path (was scalar-only), mirroring PlexMovieLibraryScanner. Root cause = upstream-inherited (verbatim at fork boundary d1dfe6eb); movie/episode paths reconcile in the tracked repo GetOrAdd, music-video repo is AsNoTracking and the scanner never did it.
  • Scope: Guids excluded (not eager-loaded → would duplicate), Directors excluded (not add-persisted for music videos).
  • Test: interaction test (substituted repos + Received()), proven non-vacuous. A real-DB double-scan can't drive this — shared in-memory SQLite connection breaks GetOrAdd's First()-nav predicate mid-scan once the existing item has metadata children (harness-only; noted in test + a memory).
  • Review: cold cross-context pass → MERGEABLE @ c633b89a, no blockers.

Filed: #500 (priority: low) — duplicate-name double-insert in the reconcile idiom, inherited from the Plex pattern; kept 1:1 parity in #497 and deferred a uniform fix for both scanners.

Triage (H12): #498 (FFmpeg split decode/encode HW accel) re-triaged low → medium after reading it — real prod regression (50% channel cold-start failure under coupled QSV decoder on the Intel host) with a VAAPI workaround. H12 now clean.

Recommended next: open backlog by priority: — #498 (medium, real prod pain, needs dual-provider migration + FFmpeg builder + UI) is the strongest non-low candidate; verify live state before claiming.

### Session — 2026-07-20 (Claude/Opus, backlog) **Worked:** #497 (Jellyfin music-video metadata reconcile) — DONE, merged #499 (c4ad4a73), issue closed. - Fix: `JellyfinMusicVideoLibraryScanner.UpdateMetadata` now reconciles Genres/Tags/Studios/Artists on the update path (was scalar-only), mirroring `PlexMovieLibraryScanner`. Root cause = upstream-inherited (verbatim at fork boundary `d1dfe6eb`); movie/episode paths reconcile in the tracked repo `GetOrAdd`, music-video repo is `AsNoTracking` and the scanner never did it. - Scope: Guids excluded (not eager-loaded → would duplicate), Directors excluded (not add-persisted for music videos). - Test: interaction test (substituted repos + `Received()`), proven non-vacuous. A real-DB double-scan can't drive this — shared in-memory SQLite connection breaks GetOrAdd's `First()`-nav predicate mid-scan once the existing item has metadata children (harness-only; noted in test + a memory). - Review: cold cross-context pass → MERGEABLE @ c633b89a, no blockers. **Filed:** #500 (priority: low) — duplicate-name double-insert in the reconcile idiom, inherited from the Plex pattern; kept 1:1 parity in #497 and deferred a uniform fix for both scanners. **Triage (H12):** #498 (FFmpeg split decode/encode HW accel) re-triaged low → **medium** after reading it — real prod regression (50% channel cold-start failure under coupled QSV decoder on the Intel host) with a VAAPI workaround. H12 now clean. **Recommended next:** open backlog by `priority:` — #498 (medium, real prod pain, needs dual-provider migration + FFmpeg builder + UI) is the strongest non-low candidate; verify live state before claiming.
Author
Owner

Session — 2026-07-20 (Claude Code, Opus 4.8)

Worked: #67 (merged), #25 (closed as obsolete). Claimed and released #68 without starting it.

Landed

  • #67 → PR #504, merged 72dc1c9f. One uploaded channel logo now drives both the guide listing and the on-screen bug, by default, with a real-geometry preview. Shared Channel Bug watermark preset seeded in DbInitializer (adopts an existing row untouched; ConfigElement marker → seeds once per database), additive imageSource on the watermark picker DTO, shared <BugPreview> component, channel-editor toggle, quick-add default. No migration.
  • #25 closed as obsolete. "Burn down SonarAnalyzer findings in Blazor .razor" — the repo has zero .razor files (removed in #91b) and no S-rule NoWarn in any of the 10 csproj. Also dropped it from the "CI & build hygiene" bundle comment so the next session doesn't rediscover it.

Filed

  • #502 (bug, priority: medium) — external-URL channel logos never render an on-screen bug: WatermarkSelector resolves the logo to its URL and then File.Exists-checks it, which is never true. The editor advertises "external URL wins", and it does — for the guide, while silently disabling the bug.
  • #503 (bug, priority: low) — watermark MiddleCenter renders bottom-right: OverlayWatermarkFilter has no switch arm for it, so it falls through to the bottom-right default. Invisible because BottomRight is that default.

Released

  • #68 — claimed as design-only alongside #67, design not produced, claim released with that stated plainly. Deferred to a clean session: it needs a freeze-when-unwatched playout clock plus an EPG derived from the resumed timeline, which is the make-or-break part. Note left on the issue that mirror channels (PlayoutSource.Mirror + PlayoutOffset) are the closest existing analogue.

Notes for the next session

  • Recon before building paid off again. #67's premise was wrong: ChannelWatermarkImageSource.ChannelLogo already resolved each channel's own logo at render time, and prod had been running that pattern by hand (43 channels → one hand-made Channel Bug row). The work was productizing it, not building it. Reading the live prod DB before designing changed the whole shape of the PR.
  • Three vacuous tests found in one PR, each by a different reviewer: sampled-not-exhaustive location coverage, a fixture with identical H/V margins that couldn't detect a prop swap, and a foreach over a possibly-empty collection that would pass if seeding silently bailed. The last was closed with a negative control. Worth actively hunting this class.
  • Parallel subagents: the backend ones failed, the SPA ones didn't. Both C#-building agents (seed, DTO) stalled repeatedly by backgrounding dotnet build and waiting on notifications; one was finished inline, the other re-dispatched with explicit "foreground only, grep for error CS" instructions and then succeeded first time. All four SPA/docs agents ran clean. Two concurrent dotnet builds on this Mac also drove free RAM from 69% → 24%. Recommendation: run backend tasks sequentially (or foreground-only), parallelize SPA/docs freely.
  • dotnet test takes the project DIRECTORY, not the .csproj path (docs/testing.md) — the .csproj form fails with "The argument ... .dll is invalid". A plan of mine specified the wrong form and cost a cycle.
  • A zero exit code is not a green build. A background build reported exit 0 while its output said 4 Error(s); trusting it plus --no-build would have run a stale dll and reported a false pass. Grep the build log for error CS.
  • Adding a positional parameter to a C# record is source-breaking for every existing constructor call site (4 here, in 2 files). Enumerate them repo-wide before building, not one compiler error at a time.
## Session — 2026-07-20 (Claude Code, Opus 4.8) **Worked**: #67 (merged), #25 (closed as obsolete). Claimed and released #68 without starting it. ### Landed - **#67 → PR #504, merged `72dc1c9f`.** One uploaded channel logo now drives both the guide listing and the on-screen bug, by default, with a real-geometry preview. Shared `Channel Bug` watermark preset seeded in `DbInitializer` (adopts an existing row untouched; `ConfigElement` marker → seeds once per database), additive `imageSource` on the watermark picker DTO, shared `<BugPreview>` component, channel-editor toggle, quick-add default. No migration. - **#25 closed as obsolete.** "Burn down SonarAnalyzer findings in Blazor `.razor`" — the repo has **zero `.razor` files** (removed in #91b) and no S-rule `NoWarn` in any of the 10 csproj. Also dropped it from the "CI & build hygiene" bundle comment so the next session doesn't rediscover it. ### Filed - **#502** (bug, priority: medium) — external-URL channel logos never render an on-screen bug: `WatermarkSelector` resolves the logo to its URL and then `File.Exists`-checks it, which is never true. The editor advertises "external URL wins", and it does — for the guide, while silently disabling the bug. - **#503** (bug, priority: low) — watermark `MiddleCenter` renders bottom-right: `OverlayWatermarkFilter` has no switch arm for it, so it falls through to the bottom-right default. Invisible because `BottomRight` *is* that default. ### Released - **#68** — claimed as design-only alongside #67, **design not produced**, claim released with that stated plainly. Deferred to a clean session: it needs a freeze-when-unwatched playout clock plus an EPG derived from the resumed timeline, which is the make-or-break part. Note left on the issue that mirror channels (`PlayoutSource.Mirror` + `PlayoutOffset`) are the closest existing analogue. ### Notes for the next session - **Recon before building paid off again.** #67's premise was wrong: `ChannelWatermarkImageSource.ChannelLogo` already resolved each channel's own logo at render time, and **prod had been running that pattern by hand** (43 channels → one hand-made `Channel Bug` row). The work was productizing it, not building it. Reading the live prod DB before designing changed the whole shape of the PR. - **Three vacuous tests found in one PR**, each by a different reviewer: sampled-not-exhaustive location coverage, a fixture with identical H/V margins that couldn't detect a prop swap, and a `foreach` over a possibly-empty collection that would pass if seeding silently bailed. The last was closed with a negative control. Worth actively hunting this class. - **Parallel subagents: the backend ones failed, the SPA ones didn't.** Both C#-building agents (seed, DTO) stalled repeatedly by backgrounding `dotnet build` and waiting on notifications; one was finished inline, the other re-dispatched with explicit "foreground only, grep for `error CS`" instructions and then succeeded first time. All four SPA/docs agents ran clean. Two concurrent `dotnet` builds on this Mac also drove free RAM from 69% → 24%. **Recommendation: run backend tasks sequentially (or foreground-only), parallelize SPA/docs freely.** - **`dotnet test` takes the project DIRECTORY, not the `.csproj` path** (`docs/testing.md`) — the `.csproj` form fails with "The argument ... .dll is invalid". A plan of mine specified the wrong form and cost a cycle. - **A zero exit code is not a green build.** A background build reported exit 0 while its output said `4 Error(s)`; trusting it plus `--no-build` would have run a stale dll and reported a false pass. Grep the build log for `error CS`. - **Adding a positional parameter to a C# record is source-breaking** for every existing constructor call site (4 here, in 2 files). Enumerate them repo-wide before building, not one compiler error at a time.
Author
Owner

Session — 2026-07-20 (Claude Code orchestrator, "batch the deploy-gated queue")

Frame: operator asked to prefer batching and get the queue down. The selector's top two priority: medium picks (#350, #474) were both CLAIM?-flagged and, on inspection, blocked on the same thing: a release + prod deploy that had just happened (v26.11.0, cut 22:22). So they batched naturally — one deploy unblocked both.

Deploy verification (defeating the #350 trap): confirmed prod runs v26.11.0 by manifest digest — running container's image digest sha256:ca5e1fa1… == registry :26.11.0 and :prod manifest. Tag/StartedAt/OCI-labels were not trusted.

Closed

  • #474 (music channels dead) — DONE, verified live. Enabled shouldSyncItems on Jellyfin library 16 via the intended REST write path (PUT /api/v1/media-sources/jellyfin/2/libraries, which also enqueues the scan — zero downtime). Scan ran 0 → 1437 music videos, 198 artists, all State=0, idempotent; library 16 now has its jellyfin:// LibraryPath. Themed-collection population is correctly deferred to #487. Minor side effect flagged: replace-complete-set semantics recreated the untouched, sync-disabled Standup Jellyfin library row (id 17→18) — no content impact (channel 131 still has 48 playout items; local Standup lib 14 untouched).

Advanced (left open)

  • #350 (cold-start ~7s) — captured the startup sub-split (#472's instrumentation) on prod: 13 cold-starts, worst ch229 8308ms (reproduces the symptom). Decisive result: the driver is firstGop (decode-to-first-GOP, 2.9–3.6s on shuffle/collection channels) plus occasional ffmpegInit probe spikes (2–4s) — subtitle burn-in is refuted (fastest channels burn subs; slow ones split T/F evenly). Retargets the fix toward start-point-to-keyframe / input -ss seeking / probe caps. The optimization itself is a transcode-pipeline change (design + review + release + re-measure) — not taken this session. No claim held.

Queue hygiene

  • #487 is now fully unblocked (both blockers #497 + #474 closed) — but it depends on the MCP server (#58) existing, so it's not an immediate pickup.
  • H12 audit: labeled two unlabeled issues split from #502's review tonight — #511 (remote-image hardening / SSRF) → priority: medium; #510 (deco-path missing-logo divergence) → priority: low. Audit now clean ✓.

State of the backlog

No quick closes remain on the shortlist: #58 (MCP epic) / #60 (browser playback) are long-horizon priority: low features, #22 is Renovate's auto-dashboard, #487 waits on #58, #350 needs the optimization arc. Live Gitea state is authoritative — this comment is a record, not a "next pick".

## Session — 2026-07-20 (Claude Code orchestrator, "batch the deploy-gated queue") **Frame:** operator asked to prefer batching and get the queue down. The selector's top two `priority: medium` picks (#350, #474) were both `CLAIM?`-flagged and, on inspection, blocked on the **same** thing: a release + prod deploy that had *just* happened (v26.11.0, cut 22:22). So they batched naturally — one deploy unblocked both. **Deploy verification (defeating the #350 trap):** confirmed prod runs v26.11.0 by **manifest digest** — running container's image digest `sha256:ca5e1fa1…` == registry `:26.11.0` **and** `:prod` manifest. Tag/`StartedAt`/OCI-labels were *not* trusted. ### Closed - **#474** (music channels dead) — **DONE, verified live.** Enabled `shouldSyncItems` on Jellyfin library 16 via the intended REST write path (`PUT /api/v1/media-sources/jellyfin/2/libraries`, which also enqueues the scan — zero downtime). Scan ran `0 → 1437 music videos, 198 artists`, all `State=0`, idempotent; library 16 now has its `jellyfin://` LibraryPath. Themed-collection *population* is correctly deferred to **#487**. Minor side effect flagged: replace-complete-set semantics recreated the untouched, sync-disabled **Standup** Jellyfin library row (id 17→18) — no content impact (channel 131 still has 48 playout items; local Standup lib 14 untouched). ### Advanced (left open) - **#350** (cold-start ~7s) — captured the **`startup` sub-split** (#472's instrumentation) on prod: **13 cold-starts**, worst **ch229 8308ms** (reproduces the symptom). Decisive result: the driver is **`firstGop`** (decode-to-first-GOP, 2.9–3.6s on shuffle/collection channels) plus occasional **`ffmpegInit`** probe spikes (2–4s) — **subtitle burn-in is refuted** (fastest channels burn subs; slow ones split T/F evenly). Retargets the fix toward start-point-to-keyframe / input `-ss` seeking / probe caps. The optimization itself is a transcode-pipeline change (design + review + release + re-measure) — not taken this session. No claim held. ### Queue hygiene - **#487** is now **fully unblocked** (both blockers #497 + #474 closed) — but it depends on the MCP server (#58) existing, so it's not an immediate pickup. - **H12 audit**: labeled two unlabeled issues split from #502's review tonight — **#511** (remote-image hardening / SSRF) → `priority: medium`; **#510** (deco-path missing-logo divergence) → `priority: low`. Audit now clean ✓. ### State of the backlog No quick closes remain on the shortlist: #58 (MCP epic) / #60 (browser playback) are long-horizon `priority: low` features, #22 is Renovate's auto-dashboard, #487 waits on #58, #350 needs the optimization arc. Live Gitea state is authoritative — this comment is a record, not a "next pick".
Author
Owner

Session — #502 external-URL channel logos render an on-screen bug (Claude Code, Opus 4.8)

Picked up

Selector (scripts/select-queue.sh) returned no arc / OPEN-milestone / review / priority: high candidates — the expected maintenance-mode steady state. Top three were all priority: medium:

#350  CLAIM?   Channel cold-start ~7s to first segment
#474  CLAIM?   Music channels dead: Music Videos library
#502  -        External-URL channel logos never render an on-screen bug

Took #502, the third-ranked one, because both above it resolved their CLAIM? flags as released claims parked on the same external gate — a release + prod deploy, operator-authorized and outside repo control. Neither has in-repo work left. (Since then main has picked up v26.11.0 release notes via #507, so that gate looks to be in flight.)

Outcome — #502 CLOSED, PR #513 merged

External-URL channel logos never rendered a bug because WatermarkSelector resolved the URL and then File.Exists-gated it one line later. Fixed by passing external URLs through (the convention every other consumer already follows) and routing remote-URL watermarks to the graphics engine rather than ffmpeg's -i. Three duplicated blocks collapsed into one shared helper; ChannelLogo resolution had zero test coverage before this.

Full detail in #502's close comment and the docs/decisions.md entry.

Worth passing on

  • The live-E2E is A/B, not just "green". The render path was verified by building unmodified origin/main in a second worktree and driving the same seeded config through both: base renders 0 logo pixels and logs this issue's warning verbatim; the fix renders. A one-armed E2E here would have proved nothing.
  • scripts/e2e-local.sh readiness probe only works on a FRESH config dir. It waits for Done migrating search index, which a reused config never emits (Search index is already version 50 instead) — so it times out and kills a perfectly healthy server. For an A/B against one config, launch dotnet ErsatzTV.dll directly and poll an endpoint. Cost me two failed launches.
  • The troubleshooting-playback API cannot exercise channel branding. PrepareTroubleshootingPlaybackHandler's media-item branch builds a synthetic channel with Artwork = [], and its ChannelId > 0 branch ignores the media item and goes through the real playout. Anything touching channel artwork needs the full collection → schedule → playout seed.
  • Adversarial review earned its keep on documentation, not code. First pass returned BLOCKED because I wrote that the deco path was unaffected — true of resolution, false of routing, since SelectWatermarks feeds deco options into the same list the routing guard filters. The code was fine; the docs would have misled the next session. Corrected the docs rather than narrowing the guard.
  • Two "tracked separately" deferrals became real issues (#510, #511) after review flagged that an unverifiable deferral is itself a finding.
  • decisions.md conflicted on rebase again (the known treadmill) — resolved additively, both entries kept, code byte-identical. Re-ran the full suite and both gates after rebasing rather than trusting the pre-rebase run; worth doing, since main's #498 had independently normalized whitespace in a file I touched.

Triage of new issues

  • #510 — deco vs precedence-level missing-logo policy divergence. bug, priority: low: real inconsistency, no user report, and the right answer depends on #1.
  • #511 — remote-image fetch hardening (timeout, size cap, redirects, pooling, caching, SSRF). priority: medium: pre-existing, but #502 widened the path it sits on, and the SSRF is admin-only.

Note an automation appears to be auto-labelling new issues now — #510 briefly carried two priority: labels and #511 was labelled without my doing so. Left one each.

Housekeeping

  • in-progress removed from #502.
  • H12 audit: clean — every open issue carries a priority: label.
  • CI green on the merged head; no runs left burning.

Still needs a human

#350 and #474 are blocked only on a release + prod deploy, not on code. I did not tag or deploy anything.

## Session — #502 external-URL channel logos render an on-screen bug (Claude Code, Opus 4.8) ### Picked up Selector (`scripts/select-queue.sh`) returned no arc / OPEN-milestone / `review` / `priority: high` candidates — the expected maintenance-mode steady state. Top three were all `priority: medium`: ``` #350 CLAIM? Channel cold-start ~7s to first segment #474 CLAIM? Music channels dead: Music Videos library #502 - External-URL channel logos never render an on-screen bug ``` Took **#502**, the third-ranked one, because both above it resolved their `CLAIM?` flags as *released* claims parked on the same external gate — a release + prod deploy, operator-authorized and outside repo control. Neither has in-repo work left. (Since then main has picked up `v26.11.0` release notes via #507, so that gate looks to be in flight.) ### Outcome — #502 CLOSED, PR #513 merged External-URL channel logos never rendered a bug because `WatermarkSelector` resolved the URL and then `File.Exists`-gated it one line later. Fixed by passing external URLs through (the convention every other consumer already follows) and routing remote-URL watermarks to the graphics engine rather than ffmpeg's `-i`. Three duplicated blocks collapsed into one shared helper; `ChannelLogo` resolution had zero test coverage before this. Full detail in #502's close comment and the `docs/decisions.md` entry. ### Worth passing on - **The live-E2E is A/B, not just "green".** The render path was verified by building unmodified `origin/main` in a second worktree and driving the *same* seeded config through both: base renders 0 logo pixels and logs this issue's warning verbatim; the fix renders. A one-armed E2E here would have proved nothing. - **`scripts/e2e-local.sh` readiness probe only works on a FRESH config dir.** It waits for `Done migrating search index`, which a reused config never emits (`Search index is already version 50` instead) — so it times out and kills a perfectly healthy server. For an A/B against one config, launch `dotnet ErsatzTV.dll` directly and poll an endpoint. Cost me two failed launches. - **The troubleshooting-playback API cannot exercise channel branding.** `PrepareTroubleshootingPlaybackHandler`'s media-item branch builds a *synthetic* channel with `Artwork = []`, and its `ChannelId > 0` branch ignores the media item and goes through the real playout. Anything touching channel artwork needs the full collection → schedule → playout seed. - **Adversarial review earned its keep on documentation, not code.** First pass returned BLOCKED because I wrote that the deco path was unaffected — true of resolution, false of routing, since `SelectWatermarks` feeds deco options into the same list the routing guard filters. The code was fine; the docs would have misled the next session. Corrected the docs rather than narrowing the guard. - **Two "tracked separately" deferrals became real issues** (#510, #511) after review flagged that an unverifiable deferral is itself a finding. - `decisions.md` conflicted on rebase again (the known treadmill) — resolved additively, both entries kept, code byte-identical. Re-ran the full suite and both gates *after* rebasing rather than trusting the pre-rebase run; worth doing, since main's #498 had independently normalized whitespace in a file I touched. ### Triage of new issues - **#510** — deco vs precedence-level missing-logo policy divergence. `bug`, `priority: low`: real inconsistency, no user report, and the right answer depends on #1. - **#511** — remote-image fetch hardening (timeout, size cap, redirects, pooling, caching, SSRF). `priority: medium`: pre-existing, but #502 widened the path it sits on, and the SSRF is admin-only. Note an automation appears to be auto-labelling new issues now — #510 briefly carried two `priority:` labels and #511 was labelled without my doing so. Left one each. ### Housekeeping - `in-progress` removed from #502. - **H12 audit: clean** — every open issue carries a `priority:` label. - CI green on the merged head; no runs left burning. ### Still needs a human **#350 and #474 are blocked only on a release + prod deploy**, not on code. I did not tag or deploy anything.
Author
Owner

Post-closure knowledge-retrieval audit

#237 remains closed archival history and should not return to the normal startup/selection path.

Implementation follow-ups:

  • #520 — retire #237 from startup; run compact orientation and scripts/select-queue.sh 5 concurrently; use structured per-issue closing records.
  • #521 — make decision knowledge lifecycle-addressable and retrieval-efficient.
  • timothy/server-management#642 — export Gitea issue/comment history (including this issue) into MemPalace with provenance and freshness.

No arc or queue state is reactivated by this comment.

## Post-closure knowledge-retrieval audit #237 remains **closed archival history** and should not return to the normal startup/selection path. Implementation follow-ups: - #520 — retire #237 from startup; run compact orientation and `scripts/select-queue.sh 5` concurrently; use structured per-issue closing records. - #521 — make decision knowledge lifecycle-addressable and retrieval-efficient. - `timothy/server-management#642` — export Gitea issue/comment history (including this issue) into MemPalace with provenance and freshness. No arc or queue state is reactivated by this comment.
Author
Owner

Session — 2026-07-20/21 (Claude Code, Opus 4.8)

Picked up: #511 (priority: medium, deps clear) — top eligible from scripts/select-queue.sh. Arc/milestone/review tiers empty, as expected in maintenance mode.

Landed: PR #518 → main. #511 closed.

What shipped

Bounded remote-image fetching in the graphics engine: IRemoteImageFetcher / HttpRemoteImageFetcher (pooled IHttpClientFactory, linked-CTS deadline covering headers and body, 10 MiB wire cap enforced during the copy, redirect cap 3), plus decode and retention budgets in ImageElementBase. Modelled on the existing IRemoteStreamProber rather than inventing a convention.

The part worth reading

Five rounds of independent adversarial review. Rounds 2–4 came back BLOCKED, each with a working payload:

Round Bypass
2 Independent caps don't compose: 2500×2500 × 600 frames = 60 KiB wire → ~14 GiB decode, passing both a 50 MP dimension check and a 600-frame check
3 Budget sourced from the header: APNG Identify reports 0 frames while Load produces 600 → 4000×4000 × 600 = 134 KiB → ~36 GiB
4 Not a bypass — my fix threw on 13/16 APNG shapes that load fine, silently breaking animated-PNG logos that worked before the PR

Rounds 2 and 3 failed the same way: the bound was enforced on a number the decoder doesn't honor. The fix is structural — cap the decoder (DecoderOptions.MaxFrames) and re-verify against the decoded reality. Round 5 attacked that with crafted files from an independent encoder (PIL) and hand-patched PNG fcTL chunks and couldn't move it.

Lore worth carrying (not repo-specific enough for decisions.md)

  • The #242 re-review-the-fix-commit convention paid for itself three times in one session. Every BLOCKED verdict came from reviewing a fix, not the original diff. On security-shaped work, budget for the fix commit to be wrong.
  • Reproduce a reviewer's finding before fixing it. Doing so caught that the MaxFrames off-by-one was APNG-specific, not universal as reported — which mattered, because it's what makes the peak overshoot 3× rather than 2×.
  • Test-shape blindness is real. Round 4's regression was invisible because the tests used 64×64, one of the few APNG shapes that happens to work. Sweep shapes; don't trust the one your fixture picked.
  • A confidently wrong decisions.md entry is worse than a gap. Two earlier versions of my entry claimed bounds the code didn't enforce. Both reviews pushed on the doc as hard as on the code. The merged entry records both failed attempts, because that's the transferable part.

Housekeeping

  • H12 audit: clean — every open issue carries a priority: label.
  • decisions.md conflicted on rebase twice (the known append treadmill); resolved by keeping main's entry and re-appending mine at EOF. decisions.md append-only CI green both times.
  • No follow-up issues filed. One informational note in PR #518: multi-page TIFF safety is inherited from ImageSharp's "Images with different sizes are not supported" — re-check on an ImageSharp bump, nothing actionable now.

Recommended next: nothing claimed. Queue falls through to the priority: backlog; re-run scripts/select-queue.sh (#22/#60/#62/#68 were the priority: low tail, #68 flagged CLAIM?).

## Session — 2026-07-20/21 (Claude Code, Opus 4.8) **Picked up:** #511 (`priority: medium`, deps clear) — top eligible from `scripts/select-queue.sh`. Arc/milestone/review tiers empty, as expected in maintenance mode. **Landed:** PR #518 → `main`. #511 closed. ### What shipped Bounded remote-image fetching in the graphics engine: `IRemoteImageFetcher` / `HttpRemoteImageFetcher` (pooled `IHttpClientFactory`, linked-CTS deadline covering headers *and* body, 10 MiB wire cap enforced during the copy, redirect cap 3), plus decode and retention budgets in `ImageElementBase`. Modelled on the existing `IRemoteStreamProber` rather than inventing a convention. ### The part worth reading **Five rounds of independent adversarial review. Rounds 2–4 came back BLOCKED**, each with a working payload: | Round | Bypass | |---|---| | 2 | Independent caps don't compose: 2500×2500 × 600 frames = **60 KiB wire → ~14 GiB decode**, passing both a 50 MP dimension check and a 600-frame check | | 3 | Budget sourced from the header: APNG `Identify` reports **0 frames** while `Load` produces 600 → 4000×4000 × 600 = **134 KiB → ~36 GiB** | | 4 | Not a bypass — my *fix* threw on 13/16 APNG shapes that load fine, silently breaking animated-PNG logos that worked before the PR | Rounds 2 and 3 failed the same way: **the bound was enforced on a number the decoder doesn't honor.** The fix is structural — cap the decoder (`DecoderOptions.MaxFrames`) and re-verify against the decoded reality. Round 5 attacked that with crafted files from an *independent* encoder (PIL) and hand-patched PNG `fcTL` chunks and couldn't move it. ### Lore worth carrying (not repo-specific enough for `decisions.md`) - **The `#242` re-review-the-fix-commit convention paid for itself three times in one session.** Every BLOCKED verdict came from reviewing a *fix*, not the original diff. On security-shaped work, budget for the fix commit to be wrong. - **Reproduce a reviewer's finding before fixing it.** Doing so caught that the `MaxFrames` off-by-one was APNG-specific, not universal as reported — which mattered, because it's what makes the peak overshoot 3× rather than 2×. - **Test-shape blindness is real.** Round 4's regression was invisible because the tests used 64×64, one of the few APNG shapes that happens to work. Sweep shapes; don't trust the one your fixture picked. - **A confidently wrong `decisions.md` entry is worse than a gap.** Two earlier versions of my entry claimed bounds the code didn't enforce. Both reviews pushed on the doc as hard as on the code. The merged entry records both failed attempts, because that's the transferable part. ### Housekeeping - H12 audit: clean — every open issue carries a `priority:` label. - `decisions.md` conflicted on rebase twice (the known append treadmill); resolved by keeping main's entry and re-appending mine at EOF. `decisions.md append-only` CI green both times. - No follow-up issues filed. One informational note in PR #518: multi-page TIFF safety is inherited from ImageSharp's `"Images with different sizes are not supported"` — re-check on an ImageSharp bump, nothing actionable now. **Recommended next:** nothing claimed. Queue falls through to the `priority:` backlog; re-run `scripts/select-queue.sh` (`#22`/`#60`/`#62`/`#68` were the `priority: low` tail, `#68` flagged `CLAIM?`).
timothy unpinned this 2026-07-21 22:28:03 +02:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: timothy/ersatztv#237