fix(876): sweep session narrative out of hooks, workflows, scripts, tests and code comments; grow the detector to the process corpus #906
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: timothy/ersatztv#906
Reference in New Issue
Block a user
Closes #876.
docs.no-session-narrativereaches every durable artifact; its detector scanned onlydocs/**/*.mdand root markdown, and nothing had ever swept the rest. The issue named four sites and called them a
floor. This PR derives the population, classifies every site with #812's three dispositions, and
grows the detector to the corpus the sweep measured as active.
The population was not 4
A whitespace-joined sweep over every tracked file OUTSIDE the detector's population — for the
detector's own PATTERNS plus the attribution / review-round class #812 found — returned 453 sites in
108 files at
fb5592971(2026-09-03). A second pass for phrasings the first list missed(hyphenated
round-N, "an earlier version", "the reviewer proved", "used to carry") added residualsin the same files. Like #812, no closure over a phrasing space is claimed: the manifests' sites were
remediated, and the detector's
--allover the grown population returns only the one docs carve-out#812 already recorded.
.claude/,.gitea/,.husky/,docker/andscripts/(non-test)scripts/tests/web/design-system/Dispositions
724 manifest rows across the five per-corpus manifests (posted as comments below; round-2 rows that
restate a round-1 site are marked as such): SEVER 467 · CUT 49 · KEEP 208 (carve-out, citation,
fixture, false-positive — each labelled with its reason). The rule applied throughout: the who-benefits
test; sever the attribution, keep the finding; a code comment's issue/PR number is a citation and stays;
a hook comment's rejected design stays with its harm; a test docstring's "round N found X" becomes X.
Twenty-six vitest titles carrying review-round or severity labels or a reviewer attribution (call sites whose title changed, each changed title line walked back to its
it(/it.each(...)(anchor) were renamed after proving nothing filters orreferences a title.
The detector grows to the process corpus
scripts/check-doc-narrative.pynow scans every tracked file under.claude/,.gitea/,.husky/and
scripts/regardless of extension, minus itself and its own test (exempt by NAME — their fixturesare the phrasings);
--alllists tracked REGULAR files, so a tracked symlink (this repo's cross-repo.claude/skills/jellyfinlink) is neither scanned under its own name nor counted, and neither is a gitlink.scripts/tests/fixtures/is exempt as test data: it holds copies of decision records, so the records' own exemption reasoning
applies, and it is what keeps the record's "no scanned path deeper than three slashes" measurement true
(re-measured: 179 files, max depth 3, and now asserted by the oracle test so the claim cannot go stale silently). The #812 argument for keeping
docs/superpowers/**in scope runs the other wayhere:
--diffsees only ADDED lines, and 287 of the 453 sites were under 30 days old — this iswhere narrative is being added, so the advisory nudge has reach. Density agrees: 56 line-mode
PATTERNS hits over the 113 regular files the predicate admits (120 tracked entries minus the two by-name exemptions, the fixtures directory and a symlink), against 9 over 66 docs files before #812.
web/and C# stay outon the same measurement: 3 of the 74 PATTERNS-matching sites, across ~4,600 files. The predicate did
not grow — PATTERNS matched 74 of 453 sites, and widening the word list to the attribution class is
the treadmill the withdrawn parity test ran on; the record says so.
Proof: the population oracle is restated over path segments with the new arms; the synthetic cross
product gains the process heads,
tests/hooksseconds, the exempt basenames and.py/.sh/.yml;the fixture test witnesses a skill, a hook, a test and the exempt file's twin one directory over, and a
symlink fixture witnesses that a link to a file outside the population is neither scanned nor counted.
Ten mutants each redden the suite: keep the TAB git appends to a
+++filename containing a space (a pre-existing silent drop in--diff, fixed here with a positive control), test the mode per index row instead of per path (a conflicted path is otherwise scanned three times, and a symlink/regular add/add conflict opens the symlink), drop the process arm, drop the by-name exemption, exempt bystartswithinstead of by name (a.py.bakaxis in the cross product is what sees it), dropscripts/,add
web/, drop the fixtures exemption, list only*.mdin--all, drop the symlink filter.Code is unchanged by construction
Per file type, HEAD vs
origin/main: Python modules AST-equal with docstrings stripped, except#lines inside the embedded fixture programs (string literals) of three test modules; workflows differ
only in
#lines insiderun:block scalars; shell, C#, TypeScript and jq equal with comment linesstripped. The stated exceptions: the detector and its test, the 26 vitest titles, two registry
notestrings in
pageSizeCallSites.guard.test.ts, and the prose fields ofmutation_manifest.py(noclause=orreplacement=touched).Verification
scripts/tests: 1569 passed, 3 skipped (final head);ruff check+format --checkon all 39 changed Python files;
bash -non every changed shell file;yaml.safe_loadon everychanged workflow; BOM unchanged on every touched
.cs.web/:eslint,tsc -b,vitest run— 121 files / 1319 tests green.check-doc-narrative.py --all: 179 files, 1 warning (the #812 carve-out);--diff origin/main:0 warnings.
decisions_validate --base origin/main: OK; catalog regenerated.ddbe9d811— Codex (cross-family, read-only; 2 MEDIUM / 2 LOW / 1 NIT) andan Opus reviewer in an isolated worktree that read every deleted line (1 HIGH / 5 MEDIUM / 10 LOW /
3 NIT; no lost knowledge, every finding about the claims and the detector). All closed in the amended
head; round 2 verified the closures (see the
Review-verdict:comment).Docs updated
docs/decisions/records/docs/no-session-narrative.md(mechanics:,signals:, body —Decisions-Edit: yes),docs/decisions/README.md(regenerated),docs/ci-cd.md(the docs-reminder population).🤖 Generated with Claude Code
https://claude.ai/code/session_01PEcBoFw7ctrf3Nb7R7x7wk
Per-site manifest — hooks, workflows, scripts (non-test)
Line numbers are from the pre-edit tree (
fb5592971);## Round 2rows come from the wider phrasing sweep. Rows marked as duplicates restate a round-1 site.manifest_AB — narrative sweep, slice AB (hooks, workflows, docker, scripts minus scripts/tests)
Line numbers are from
slice_AB.tsv(pre-edit).adjacentrows are sites found in the ±10-linecomment block around a listed site and handled under the same rules.
reasonstring; runtime session, and a displayed stringaskmessage naming the review to performCLAUDE.mdexemption kept(ersatztv#849 round 3)is a provenance pointer into the issue thread(#849 round 4)citation keptmain)evil); the count structures the recordjq -esafe-by-accident rejection and the reasoning the rule exists to retire(ersatztv#763, round 2)is a provenance pointerCreatorID == 0) without the draft chronology(ersatztv#849, round 2); the surrounding text is the finding, not chronology::warning::with its three concrete defects; the round count is the withdrawal reason(ersatztv#849, round 2)provenance pointer(ersatztv#849 round 3)provenance pointerstate = successtest kept as "one branch too narrow", with the pending-path harm(ersatztv#849 round 3)provenance pointer(#629 round 5)provenance pointernojob/norunid/1default kept with the silent-PASS harmreport; runtime sessionexec(violated eight lines below a single statement)denyharmif ! fgating plus the measured fail-open it producestrap … EXITkept with the leaked-directory consequenceCounts
15 files edited; 11 slice files untouched (all-KEEP).
Verification
Comment-only check: of the changed lines in these 15 files, every one is either a
#comment lineor Python module-docstring prose (
check-doc-narrative.py,decisions_validate.py,mcp_smoke.py). No code, no YAML keys/values, no regexes, no asserted strings changed.Notes on the two judgement calls the dispatch flagged
pretooluse-merge-consent.sh:504("It was used first, with the list consulted only on a 404, andcold review found what that left behind: …") — SEVERED to
… and that design left a false-open behind:. Everything after the colon is byte-identical: the precedence argument, themainvsm*example, "The hardened path was dead code and the unhardened one was live."check-doc-narrative.pyPATTERNS and WARNING — untouched (KEEP-fixture). The only edit in thatfile is the module docstring's "WHY PYTHON AND NOT SHELL" paragraph, which keeps the rejected bash
parser and all four of its defects.
git status(C#, web/, docs/, scripts/tests/, and the decision recorditself) are other agents' slices and were not touched here.
Round 2
Scope: the whole of every file in the slice, plus five files newly owned here
(
.gitea/workflows/ci-image.yml,.gitea/workflows/pr-checks.yml,scripts/check-required-contexts.sh,scripts/e2e-functional.sh,scripts/lib/h10-reviewers.sh).29 of the 67 rows in
round2_AB.tsvduplicate sites already graded in round 1 and are skipped(rows 1, 2, 3, 6, 9, 10, 18, 19, 23, 25, 27, 28, 29, 30, 32, 34, 35, 37, 40, 41, 43, 44, 49, 51, 52,
54, 55, 56, 57 — line numbers shifted by round-1 edits in a few cases). The 38 new rows follow,
then the
adjacentsites the whole-file sweep turned up.echo "Changed files in this PR:"::error::stating the commit-the-artifacts ruleecho "Changed .cs files in this PR:"::error::stating the fix-as-you-touch ruleelsebranch ends with a successfulecho"; the false-open kept in fullcommand not found, the dead PROTECTED branch and the three blind tests kept(ersatztv#849 round 3)citation ten lines above(ersatztv#706, round-3 review)— issue-anchored locator, no discovery verbecho "Changed files in this PR:"::warning::stating the parity-doc rule--no-mergessilently skipped the trailer::notice::stating what the job does and does not assertpytest.ini… disarms the entire gate"; both disarm shapes and the harm keptRound 2 counts
Round 2 edited 8 files:
pretooluse-agent-model.sh,docker-build.yml,pr-checks.yml,review-verdict.yml,check-required-contexts.sh,ci-step-ran.sh,hook-fire-log.sh,lib/h10-reviewers.sh.ci-image.ymlande2e-functional.share all-KEEP and untouched.Round 2 rule applied for
round-NcitationsSevered where the phrase carries a discovery verb or no issue anchor
(
(ersatztv#698, found in round-4 review)→(ersatztv#698, round 4);(round-5 review)cut;"until round-2 review" cut). Kept where it is an issue-anchored locator into a thread a reader would
open (
(ersatztv#849 round 3),(ersatztv#706, round-3 review),(ersatztv#763, round 2),(#629 round 5)), and kept wholesale injq-preflight.sh, where the round-1/2/3 sequence IS therecurrence argument the block exists to make.
Round 2 verification
Cumulative
git diff --statfor my files (rounds 1 + 2)Comment-only check re-run over all 19: 25 changed lines are not
#comments, and all 25 are Pythonmodule-docstring prose in
check-doc-narrative.py,decisions_validate.pyandmcp_smoke.py—unchanged from round 1, i.e. round 2 touched no non-
#line at all. No code, YAML key/value, regex,identifier or asserted string was modified.
Grand totals (rounds 1 + 2)
Per-site manifest — scripts/tests part 1
Line numbers are from the pre-edit tree (
fb5592971);## Round 2rows come from the wider phrasing sweep. Rows marked as duplicates restate a round-1 site.Manifest — slice C1 (
scripts/tests/, 113 listed sites + 24 adjacent)Line numbers are the ones in
slice_C1.tsv(pre-edit).adjacentrows carry the pre-edit linenumber found by a broader marker sweep over the same 15 files.
"$SHA"-twice defeat stays; only the finder is removed#742kept as the provenance pointer; "first draft of" removedersatztv#803kept; "round 2" and "found by hand while re-reading the diff" droppedexit 1defeat kept#751kept as the pointer#751survives in the next paragraph""resolution and the greened-rejection harm keptdocs.no-session-narrativeif false/ exit 5 / still-prints-unreadablemeasurement keptbranch_nameshort-circuit and both payload shapes kept.husky/pre-commit:7mention-vs-invocation trap kept verbatimdependency-scan.ymlkeptAdjacent rows (found beyond the slice, same files, same rules)
adjacentadjacentfiles?page=1&limit=50evasion keptadjacent[]leaves the type gate unexercised — kept as the harmadjacentadjacentposted is Nonereasoning and the FORGED-verdict inversion keptadjacentadjacentjq - e '.'token split keptadjacentadjacentadjacentersatztv#751kept as the pointeradjacentadjacentadjacentadjacentadjacentadjacentadjacentadjacentadjacentadjacentrecords/README.mdhole keptadjacentadjacentadjacentadjacentadjacentadjacentCounts
Listed sites (113): SEVER 87 · CUT 8 · KEEP-carveout 10 · KEEP-false-positive 6 · KEEP-fixture 0
Adjacent rows (26 rows covering 24 edited sites + KEEP groups): SEVER 19 · KEEP-carveout 6 · KEEP 1
Total edits applied: 111 replacement operations across 15 files (including 8 pure re-flows of
paragraphs a sever shortened).
Verification
AST equality proof (comments/docstrings only — no code, assertion, regex or identifier changed):
The single difference is ONE string constant — the embedded
ghstub script at line ~950. Itsdiff was inspected line by line: every changed line inside it begins with
#(a comment in theembedded script). No executable line of the stub changed.
(exact figures reproduced in the final report; run after the last reflow)
Residual markers deliberately left
cold review|round N|earlier version|first version|first attempt|turned outstill match 8 lines inthese files. Each is a labelled KEEP above: 4 are fixture behaviour ("fails the first attempt of
each logical read"), 1 is the mutant-precision statement about text an earlier commit shipped, and 3
are carve-outs (a shipped helper's defect, a tested-and-not-a-defect report, and the six-round
non-convergence evidence the active
docs.no-session-narrativerecord itself cites).Round 2
Scope: the WHOLE of each of the 15 round-1 files plus the two newly owned
(
test_ci_image_paths_pin_agreement.py,test_mutation_harness.py), with the wider phrasing set.Rows that duplicate round 1 — skipped
test_decisions_validate.py:661,:1154;test_pr_changed_files.py:3,:1913,:2410,:2542,:3596,:5229,:5977,:6851were already dispositioned in round 1 (all KEEP-carveout orKEEP-false-positive) and are unchanged.
test_pr_changed_files.py:478and:4042are assertionmessage strings, not comments — KEEP-fixture, not touched.
test_guard_populations_derive_from_git.py:401is "in this process", a false positive of thein this PRpattern — KEEP-false-positive.New round-2 dispositions
command not found/ dead PROTECTED branch / three tests green anyway all keptstate = successis one branch too narrow" keeps bothFurther same-class sites met in round 2 (not on either list)
adjacentadjacentadjacent<= p95-><survivor stays as the evidenceadjacentadjacentadjacentadjacentadjacentadjacentRound 2 counts
Listed rows (34, covering 32 sites): SEVER 15 · CUT 2 · KEEP-carveout 2 · KEEP-false-positive 2 ·
KEEP-fixture 2 · duplicates skipped 11.
Further sites met: SEVER 7 · KEEP-carveout 1 · KEEP 1.
Round 2 edits applied: 24 replacements plus 8 paragraph re-flows, across 8 files.
Round 2 verification
AST equality (docstrings stripped,
ast.dumpcompared againstHEAD:<file>):The one difference is the same embedded
ghstub string constant as in round 1; every changed lineinside it begins with
#.test_ci_image_paths_pin_agreement.pyis absent from the stat because its only site is a KEEP.Residual after round 2 — 30 lines, all labelled KEEP
The wide pattern set still matches 30 lines across the 17 files. Every one is a KEEP above:
21 are
used to <verb>describing the SHIPPED code/workflow/validator's prior behaviour (thetrap-and-consequence carve-out, not a draft); 4 describe fixture retry behaviour ("fails the first
attempt of each logical read"); 1 is a quoted phrase inside a methodological argument
(
test_pr_changed_files.py:1913); 1 is an assertion MESSAGE string (:4042); 1 is themutant-precision statement (
:5372); 1 is "in this process"; andtest_decisions_validate.py:83is left because its
previouslycannot be shown still-true or not-true without re-deriving thevalidator, and the brief forbids fabricating a replacement.
Per-site manifest — scripts/tests part 2
Line numbers are from the pre-edit tree (
fb5592971);## Round 2rows come from the wider phrasing sweep. Rows marked as duplicates restate a round-1 site.Manifest — slice C2 (113 sites, 20 files under
scripts/tests/)Line numbers are the slice's (pre-edit).
edited= whether the file text changed at that site.scripts/tests/test_hook_fire_log.py
:-fallback mechanism kept.:left this file green".-evs-efargument kept.mktemp" → "also reachable by stubbingmktemp"; trigger kept.TypeErrortrap and its "broken, not stricter" harm kept.cwdpasses identically for every shape, crashes included.adjacentcwdis being ignored") is stated in the same sentence.adjacentadjacentscripts/tests/test_merge_consent_required_check.py
rule_name; the AUTO-GRANT harm kept.release/26.4path-separator 404 read as "unprotected" is the trap that retired the encoding test.*)to setbp_code=200left the suite green".scripts/tests/test_ci_dropped_step_guard.py
build's two payloads moved intoenv:, which brought it into the ban" — the do-not-narrow instruction kept.GITHUB_RUN_ATTEMPT=1, a domain term, inside a displayed assertion message.false-blindness fact is already recorded at GATE_VALUES_IN_THE_WILD.adjacentLC_ALL,${{x}}spacing, red blocks every merge) kept.adjacentmarker_pathadjacentscripts/tests/test_image_build_delegates_the_spa_suite.py
npm tselector gap kept.npm t -- --runsweep escape kept.run: >fold and the swallowed-suite consequence kept.adjacentadjacentadjacentscripts/tests/test_ci_release_path_scan_job.py
always()finding and "Nor isneeds:…" kept.adjacentadjacentadjacent$GITHUB_ENVfourth tier, its measurement, and the ban-rather-than-model rule.scripts/tests/mutation_manifest.py
Checked before editing:
test_mutation_harness.pyasserts only thatwhyis non-empty and thatUNDECLAREDvalues exceed 60 characters. Noclause,replacement,targetorUNDECLAREDvaluewas touched.
scripts/tests/test_check_doc_narrative.py
NARRATIVEconstant — the detector's sample input.truth()'s predicate — executable code, not prose.docs/case.md.scripts/tests/test_post_review_verdict.py
Review-verdict: MERGEABLEwith no status" — the false-consent harm kept.adjacentcheck-review-verdict.sh"; the undisclosed-loss finding kept.adjacentscripts/tests/test_worktree_ownership_guard.py
$CLAUDE_PROJECT_DIR… worked only while the hooks read that variable" — no-op-control harm kept.scripts/tests/test_ci_image_pin_population.py
container:block.MARKED_JOBSfor exactly this and then ships the same shape" — the self-criticism kept.scripts/tests/test_optional_request_members.py
ArtworkContentTypeModelmiss; the pair is the evidence for deriving the guard._resolveresolves and NOT the universal 'anywhere in the request graph'" — the false universal kept as a caution.DELETEcounterexample kept.scripts/tests/hook_fire_isolation.py
scripts/tests/conftest.py
scripts/tests/test_docs_only_detector_clone_depth.py
--depth=200is inert" — the no-production-consumer statement kept.scripts/tests/test_merge_consent_base_change.py
scripts/tests/test_review_verdict_vocabulary.py
scripts/tests/test_ci_toolchain_image_resolves.py
scripts/tests/test_merge_consent_exemption.py
scripts/tests/test_migration_equivalence.py
docs/decisions/… is the wrong base"; both wrong bases and the 105-link measurement kept.scripts/tests/test_workflow_persist_credentials.py
Counts
Slice sites (113):
Adjacent sites found beyond the slice (16 rows; 2 of them outside ±10 but in a file I own and
carrying a phrase I had just severed elsewhere in the same file —
…spa_suite.py:837,…scan_job.py:54— plus 5 rows listed for completeness and NOT edited):Verification
ruff checkandruff format --checkwere run from the worktree root on every edited file, and onceover the whole directory:
Change-shape proof — every changed line in my 20 files is a comment or docstring line:
(the one line the filter does not classify begins
1 —, a docstring continuation of "gives pytestexit / 1 — a real test failure"). No code, assertion, regex, identifier, fixture literal or
displayed message was changed.
git diff --statfor my 20 files:Round 2 — wider phrasing sweep (44 listed sites, whole-file scope)
Duplicates of round 1:
test_check_doc_narrative.py:466(already KEEP-fixture) is the only exactrepeat and is unchanged. Four sites I had dispositioned KEEP in round 1 are REVISED to SEVER here
under the wider scope, because each restates cleanly as a hazard that is true today with nothing
lost:
test_ci_dropped_step_guard.py:812,test_hook_fire_log.py:2538,test_merge_consent_required_check.py:667,test_optional_request_members.py:13/15.test_merge_consent_required_check.py:706was out of every ±10 window in round 1 and is handled now.Policy applied consistently in round 2, stated so the KEEPs are legible: history of THE ARTIFACT'S
OWN TEXT is always severed or cut; history of THE SUBJECT UNDER TEST is severed to a present-tense
hazard when that stays true and loses nothing, and KEPT when restating would drop a removed design's
record or assert something false about the current system.
cwd=$HOMEmechanism kept whole.os.fspathreason and the green-on-deletion measurement kept.os.fspathalone leavesos.path.joinraising on mixing str with bytes" — the trap kept as a rule._anchored) instead of when it shipped.docs/case.md. Duplicate of round 1.marker_pathdefaulting tonojob/norunid/1uses names every run shares"; the silent-PASS harm kept.TOOLCHAIN_JOBS/BARE_RUNNER_JOBS) and their #789 replacement.builditself is wrong on two counts"; both counts kept.os.fspathreason and removed-rather-than-proved decision kept.startswith("[")test here is dead code"; theraise-poisoning measurement kept.vitestis a SELECTOR"; all four short-goings kept...finding was an anchor subversion"; both surviving anchor mutants kept.endswithbranch below would make the assertion unfalsifiable" — the mechanism kept.branch_protections/{name}lets a base likerelease/26.4inject a path separator and 404".docs/tree would make this a one-shot"; the whole one-shot harm kept.attempt 1; both live silent drops and the column layout unchanged.attempt 2; theArtworkContentTypeModelmiss and its reachability kept.then/elsein the conjunctive list … is the polarity error"; the mutual-exclusion reason kept.total_count), the measured ORDER, and the unproducible conclusion.Further same-class sites met during the whole-file sweep (not on either list)
${…}/$NAMEtest" is the finding and stays.$CLAUDE_PROJECT_DIR… reads as harmless"; the two-roots confident-table harm kept.Confirmed-KEEP residuals the sweep re-surfaced and did not touch:
test_ci_dropped_step_guard.py:761(
GITHUB_RUN_ATTEMPTdomain term in a displayed message),hook_fire_isolation.py:56and:112(pytest-session referent),
test_check_doc_narrative.py:26/:44/:466(detector fixtures andpredicate),
mutation_manifest.py:418/:419(clause=/replacement=mutation strings),test_post_review_verdict.pyH10_REVIEWERS/"reviewer" occurrences andtest_review_verdict_vocabulary.py:55(domain terms),test_merge_consent_exemption.py:285("pin this property"),
test_workflow_persist_credentials.py:138(rule in a displayed message).Also untouched:
test_merge_consent_required_check.py:82's "this PR's sibling record" — the record'skey could not be resolved from the catalog, so naming it would be fabrication (brief: KEEP and say why).
Round 2 counts
Listed sites (44):
Further same-class sites found beyond the list (9): SEVER 7, CUT 2.
Combined across both rounds: 157 listed sites + 25 unlisted = 182 dispositioned;
SEVER 138, CUT 23, KEEP 21.
Round 2 verification
Change-shape proof, re-run over my 20 files after round 2 — 511 changed lines, all comment or
docstring, the same single unclassifiable line (a docstring continuation beginning
1 —):git diff --statfor my 20 files after round 2:Per-site manifest — web/
Line numbers are from the pre-edit tree (
fb5592971);## Round 2rows come from the wider phrasing sweep. Rows marked as duplicates restate a round-1 site.Manifest — slice D (
web/), ersatztv #876 narrative sweepWorktree
/Users/timothy/ersatztv-wt-876. 108 slice rows over 31 files; comment-only edits.Line numbers are the SLICE's (pre-edit).
edited= yes/no.Slice-specific rule applied to test TITLE strings (
it/describe/it.eachargument): gradedKEEP-fixture, not renamed — rename proposals are listed in a separate section below.declare moduleblock silently overwrites a schema); dropped the attribution.Complete<>off the wrapper, retype the test fixture) fully kept; only "Cold review demonstrated" removed..test.-only form is false; the dated measurement ("measured on 4.1.9") stays on the next line.&/`(#650 follow-up round 4); the rejected plant-a-file verification and why it pins nothing are untouched.(#650 follow-up round 3 MEDIUM finding)→(#650 follow-up); issue pointer kept, review chronology dropped.b90f8a3b, these 4 prove nothing) and the pointer to the fixture test; dropped "where the review asked for it" and "this round's".it.eachTITLE string (displayed test output). Rename proposed below, not applied.(#819 round 6)→(#819); the "not derived fromtrackedon both sides" finding is untouched.b90f8a3b; "a round-3 review found" dropped. Which fixtures are contract vs regression pins is kept in full.b90f8a3b"; the verified fixture list is unchanged.(#650 follow-up round 3 MEDIUM finding)→(#650 follow-up).(#650 follow-up round 3 MEDIUM)→(#650 follow-up).(#650 follow-up round 3 MEDIUM)→(#650 follow-up).(#651 round 8)→(#651); relative time "until now it was only exercised indirectly" → "screen tests exercise it only indirectly".itTITLE string. Rename proposed below, not applied.#578 / #651 review round 2:→#578 / #651:; the LCG derivation and why a recomputed divisor proves nothing are untouched.(#651 round 4 HIGH-1)→(#651); the cross-namespace harm stays.#651 review round 4:→#651:.(#651 round 7)→(#651); the int32-binding harm stays.(#651 round 7)→(#651).(#651 round 6)→(#651); the container-vs-contents finding is untouched.(#651 round 8)→(#651).(#651 round 4 HIGH-1)→(#651).(#651 round 4 MEDIUM-5)→(#651).(#651 round 5 MEDIUM-2: …)→(#651: …); the request-storm harm inside the parenthesis is kept.(#651 round 3, MEDIUM)→(#651).(#651 round 5 MEDIUM-2)→(#651).(#651 round 5 MEDIUM-3)→(#651).(#651 round 5 MEDIUM-3, round 6)→(#651).(#651 round 2 BLOCKER 2, round 4 HIGH-1)→(#651).(#651 round 7)→(#651).(#651 review round 3, HIGH)→(#651); the one-gate-per-commit-path rationale is untouched.(#651 review round 2, BLOCKER 3)→(#651).(#685 review round 3 — …)→(#685 — …).(#685 review round 4)→(#685).(#685 review round 4)→(#685).(#685 review round 3)→(#685); the live-region trap is untouched.(#685 review round 4)→(#685).(#651 round 8)→(#651).#651 review round 3.→#651.; the "add the guard WITH the test that reaches it" rule is untouched.(#651 round 4 HIGH-2)→(#651).#651 round 9:→#651:.(#651 round 9)→(#651); the contradictory-label finding is untouched.(#651 round 9)→(#651).(#651 round 4 HIGH-2)→(#651).(#651 round 6)→(#651), and relative "now"; the fixture trap (a mock without an ETag exercises a force-write path) is kept.(#651 round 3 HIGH, round 5 redesign)→(#651); "prevented structurally rather than by a merge rule" kept.(#651 round 4 HIGH-2)→(#651).#651 review round 5:→#651:.(#651 round 7 HIGH)→(#651); the empty-ETag force-write harm is untouched.(#651 round 7)→(#651).#651 round 8:→#651:.(#651 round 8 — the boundary, not the site)→(#651 — the boundary, not the site).(#651 round 7 HIGH)→(#651).NOTE (#651 review round 5):→NOTE (#651):; the no-reconciliation-machinery prohibition is untouched.(#651 round 4 HIGH-2)→(#651).(#651 round 7)→(#651).(#651 round 6)→(#651).(#651 round 4 MEDIUM-4, now structural rather than guarded)→(#651 — structural rather than guarded).(Codex review, PR #242)→(PR #242)— the brief's own example: keep the number, drop the attribution.#819 round 6:→#819:; the cancellation argument is untouched.(#819 round 6)→(#819).Adjacent rows (found inside the ±10-line window of a slice site)
.spec.tsbypassisProductionSourceabove.(#685).(#685); the measured harm (rows wiped whileselectedstill held them) is kept.(#685).(PR #242), same rule as line 430.Disposition counts
Slice rows (108): SEVER 84 · CUT 2 · KEEP-carveout 3 · KEEP-fixture 12 · KEEP-false-positive 7
Adjacent rows (14): SEVER 10 · CUT 1 · KEEP-carveout 1 · KEEP-false-positive 2
Test-TITLE rename proposals (NOT applied — parent decides)
Titles are displayed test output, so per the dispatch they were left alone. None is purely
chronological; each would only lose a round/severity prefix or suffix.
pageSizeCallSites.guard.test.ts:700— dropround 4from(#650 follow-up round 4)ChannelBuilder.test.tsx:845, 1220, 1342— dropround 3;:1263— dropround 4pickers.test.tsx:310— drop theBLOCKER 3 / round 3:prefix;:341— dropround 3 HIGH:;:512— drop(round 6)pageSizeScan.test.ts:42, 74—not a round-1 regression pin→not a regression pin against the prior scannerPlaylistsScreen.test.tsx:366, 410— drop theround 9:prefixRerunCollectionsScreen.test.tsx:974, 1000— drop theround 8:prefixRerunCollectionsScreen.test.tsx:364, 390— drop(round-3 review)/, round-3 reviewPlaybackTroubleshootingScreen.test.tsx:361— drop(Codex race fix)pageSizeCallSites.guard.test.ts:195—the #685 round-3 review finding→the #685 findingSame-class sites found in my files but OUTSIDE the ±10-line window (NOT edited)
Reported rather than edited, per the brief's hard limit. Each is the same phrase class I severed
above, so several of my files are now internally inconsistent (e.g.
pageSizeScan.test.ts's headerno longer names "round 1" while three section banners below still say "round-3 MEDIUM finding").
Proposed dispositions in brackets.
completeAnnotationScan.ts:86— "(the #819 round-5 finding recorded in the pageSize guard)" [SEVER →(the #819 finding …)]completeAnnotationScan.ts:247— "An earlier version threw on intersections only … Review measured that false by execution: …" [SEVER — keep the measured shapes, drop "An earlier version"/"Review measured"; this is the source-side mirror of the test comment I severed atcompleteAnnotationScan.test.ts:146]libraryBrowse.test.ts:159— "The reviewer proved this helper was dead code to the suite: deleting its clamp, or deleting…" [SEVER — keep the deadness proof, drop "The reviewer proved"]pageSizeCallSites.guard.test.ts:68— "an input class a text-level scanner previously got wrong" [SEVER → "gets wrong"]pageSizeCallSites.guard.test.ts:792— "An earlier version calledisScannableSourceFileNameon this side too … MEASURED: …" [SEVER — keep the cancellation harm and both measurements]pageSizeCallSites.guard.test.ts:1130— "the anti-vacuityexpect(...)this comment used to enforce is deleted DELIBERATELY" [KEEP-carveout — a deliberate deletion plus the condition to re-add it]pageSizeScan.test.ts:179, 210, 234— section banners// ---- round-3 MEDIUM finding: … ----[SEVER → dropround-3 MEDIUM finding:]pageSizeScan.ts:22— "A round-3 review found the AST version still had its own … false negatives" [SEVER → "The AST version still had its own …"]builder/libraryBrowse.ts:176, 187— "the exact window earlier rounds kept losing" / "the page-ordering defect earlier rounds kept chasing one interleaving at a time" [SEVER — keep the defect class, drop "earlier rounds"]pickers.test.tsx:401— "the round-3 re-query guard keyed on query alone, which SUPPRESSED this request" [SEVER — keep the rejected keyed-on-query-alone guard and its harm]pickers.test.tsx:469— "re-asking would be the round-3 defect" [SEVER → "would be the same defect"]ScheduleItemInspector.tsx:371,FillerPresetsScreen.tsx:668— "Out-of-list injection (round-3 review F2)" [SEVER → "Out-of-list injection (#651 F2)" or drop the parenthesis]SchedulesScreen.tsx:62— "(round-3 review F3)" [SEVER, same]CollectionsScreen.tsx:253— "This screen used to keep a SECOND copy of the min-query check here — the #685 review proved…" [SEVER — keep the rejected duplicate check and its harm]CollectionsScreen.tsx:287— "(#685 — merged.slice(0, 50) used to drop up to 100 of…)" [KEEP-carveout — measured harm of a removed clamp]CollectionsScreen.test.tsx:616— "the previously-ticked row" [KEEP-false-positive — domain sequence inside one test]RerunCollectionsScreen.test.tsx:623— "The round-4 false-conflict hole: …" [SEVER — keep the hole, drop "round-4"]RerunCollectionsScreen.test.tsx:787— "so this passes pre-#651-round-7 too" [SEVER — keep why the test exists, drop the round]PlaybackTroubleshootingScreen.test.tsx:385— "it is ACCEPTED (previously discarded as a stale result)" [KEEP-carveout — states the behaviour change the test pins]PlaylistsScreen.tsx:63,RerunCollectionsScreen.tsx:38— "the truncation hint they used to carry is gone with the truncation" [KEEP-false-positive — domain statement about the current code]PlexSourceScreen.test.tsx:175,libraryBrowse.test.ts:93— "used to claim …" / "used to compile with&&live" [KEEP-carveout — product-defect descriptions the tests pin]Verification
.ts/.tsxonly — no build run (comment-only edits, per the brief).Only comment lines changed.
git diff -U0 -- web/filtered for lines not starting with//,*,/*or{/*yields exactly five hits, all continuation lines of two JSX{/* … */}block comments inCollectionsScreen.tsx. No added line exceeds 118 columns.git status --porcelain -- web/lists exactly these 28 files, all inside the slice's 31-file set.The three untouched slice files are
LocalLibraryEditScreen.tsx,PlaybackTroubleshootingScreen.test.tsxandPlexSourceScreen.test.tsx(all-KEEP).Round 2 — whole-file scope, plus the 5 newly owned files
Scope: the WHOLE of all 36 owned files (31 from slice D +
web/src/api/trakt.ts,web/src/mediaSources/useDirtyGuard.ts,web/src/screens/PlayoutsScreen.tsx,PlayoutsScreen.test.tsx,SchedulesScreen.test.tsx). Sources: my Round 1 out-of-window list,round2_D_web.tsv, and my own whole-file re-sweep with a widened pattern set (round[ -]N,rounds N,an earlier,the reviewer,previously,originally,used to keep/enforce/sit/…,first fix/version,second review,in this PR, severity labels).Skipped as duplicates (one line, as asked):
round2_D_web.tsvrows 8, 11, 12, 17-20, 24, 25,28, 33, 34, 37, 38, 41, 42 are the test titles already renamed in the title pass below, and rows
32, 35, 36, 43 were already graded in Round 1.
Test-TITLE renames (decision 1) — applied
Reference check first: nothing in
scripts/,.gitea/,.husky/,web/package.json,web/vite.config.tsordocs/filters or names any of these titles.web/package.json's script isa bare
"test": "vitest",vite.config.tsselects by FILE glob only, and there is no-t/--testNamePatternanywhere in the repo. Grepping each title's distinctive fragment repo-widereturned only its own test file, every time.
13 of the 14 proposals applied, across 17 title-string lines (four titles are
' + 'concatenations spanning two lines). Each change is the only edit on its line.
(#650 follow-up round 4)→(#650 follow-up).(#650 follow-up).BLOCKER 3 / round 3:,round 3 HIGH:and(round 6)labels; descriptions unchanged.not a round-1 regression pin→not a regression pin; the body comment now names the prior scanner.round 9:prefix.round 8:prefix.(round-3 review)and, round-3 review;(seqRef guard)kept.(Codex race fix); the behaviour description carries the test.Proposal 14 was NOT a title — correction to Round 1.
pageSizeCallSites.guard.test.ts:195is aREGISTRY
notestring (prose data, asserted only asentry.note.length > 20, never rendered in afailure message). It is handled as a prose SEVER below instead.
Applied edits (decision 2)
(the #819 round-5 finding …)→(the #819 finding …).round-3 MEDIUM finding:, rule width preserved by padding the trailing dashes.(round-3 review F2)→(#644 follow-up F2); issue MEASURED fromgit log -S(commitdaedf003efix(644)), not guessed.(#644 follow-up F2).(round-3 review F3)→(#644 follow-up F3), matching the#644 follow-up F4two lines above.pickers.tsxgives it.usableEtag's trim"; it stays a regression guard.(#651 rounds 6 and 7)→(#651).KEEP rows (43 sites, no edit)
Rock && RollPRODUCT regression the test pins.BLOCKER 2:/BLOCKER 3:tests, which were NOT renamed — stripping them breaks a live cross-reference.describeTITLE strings; rename candidates.F1is a finding id used as a defect name inSchedulesScreen.tsxtoo.merged.slice(0, 50)dropping up to 100 rows is the measured harm of the removed clamp.Round 2 counts
Comment/prose sites: SEVER 25 · CUT 1 applied; KEEP 43 (carveout 10 · citation 6 · fixture 13 · false-positive 14).
Test titles: 13 proposals applied over 17 title-string lines; 1 proposal reclassified as a registry note.
Further rename candidates (still NOT applied — outside decision 1's 14)
Titles carrying only a review SEVERITY label, no round:
pickers.test.tsx:257(BLOCKER 2:),:293(BLOCKER 3:— referenced bypickers.tsx:321, so rename both together or neither),:385(HIGH-1:),:426(MEDIUM-5:),:473/:541(MEDIUM-3:);PlaylistsScreen.test.tsx:324andRerunCollectionsScreen.test.tsx:508(HIGH-2:);SchedulesScreen.test.tsx:404(— review fixes (F1/F3/F4/F5)) and:605((#242 review follow-up)).Round 2 verification
Every changed line in
git diff -U0 -- web/is one of: a//or*comment line (431), a JSXblock-comment continuation (4), a test-TITLE string (17), or one of the 2 REGISTRY
noteprosestrings. Enumerated exhaustively — the 21 added lines that are not
//-prefixed are exactly those17 + 2 + 2. No added line exceeds 118 columns.
git status --porcelain -- web/lists 29 files, all inside the 36-file owned set. Seven owned filesare untouched (all-KEEP):
LocalLibraryEditScreen.tsx,PlexSourceScreen.test.tsx,trakt.ts,useDirtyGuard.ts,PlayoutsScreen.tsx,PlayoutsScreen.test.tsx,SchedulesScreen.test.tsx.Per-site manifest — C# and design-system
Line numbers are from the pre-edit tree (
fb5592971);## Round 2rows come from the wider phrasing sweep. Rows marked as duplicates restate a round-1 site.Manifest — slice_EF (ersatztv#876 narrative sweep)
(#70, found by adversarial review of PR #402)→(#70, PR #402); kept both issue/PR numbers per brief instruction.GetOrAddFoldercalls in the same test), not session chronology.Adjacent (found beyond the slice's grep hits, same rule applied)
Counts by disposition
Verification
All 19 slice sites are in
.csfiles except site 19 (.md). Per the brief's.ts/.tsx/.csrule, no build/run needed for comment-only edits.head -c3 <file> | xxd -p) run before and after editing on all 15.csfiles in the slice: none startedefbbbfbefore, none startedefbbbfafter (all began757369=using, matching.csfile starts). No BOM was introduced.git diff --statfor the edited files (5 of 16 slice files touched: 4.cs+ 1.md):git difffor those 5 files (manually reviewed): every changed hunk is inside a///XML-doc comment, a//line comment, or markdown prose — no code, attribute, string literal, YAML key, or identifier was touched.git add/commit/stash/checkoutrun in the worktree.Round 2 (round2_EF.tsv — whole-file scope, wider phrasing sweep)
ShouldBemirror) kept with its concrete harm (silent drift, ersatztv#779/#754); the rejected design IS the actionable content.Full-file rescan (whole-file scope per round-2 instructions)
Re-grepped every file in round2_EF.tsv for the wider phrasing set (
round-N, "an earlier version", "used to carry/keep", "review proved/measured", "in this PR", plus the round-1 phrase set) across the ENTIRE file, not just ±10 lines. No additional same-class sites were found beyond the 24 listed rows above (one hit,WatermarkSelectorDecoResolutionTests.cs:468, is the same "previously yielded two" site already logged asadjacent/KEEP-false-positive in round 1).Round 2 counts by disposition
Round 2 verification
head -c3 <file> | xxd -p) run before and after on the 2.csfiles touched this round (GetSearchFieldValuesHandler.cs,GetSearchFieldValuesHandlerTests.cs): both began and remained757369(using) — no BOM introduced.git diff --statfor round-2-touched files (both already appeared in round 1's file set; totals below are the FILE'S cumulative diff across both rounds):git difffor both files: every changed hunk is inside a///XML-doc comment or a//line comment; no code, attribute, string literal, or identifier was touched.git add/commit/stash/checkoutrun in the worktree.Review-verdict: MERGEABLE @
a7d91bfSix cold review rounds, two families, review-only and worktree-isolated. Codex (cross-family) rounds 1-5 through 8cadae08a: 2 MEDIUM, 2 LOW, 1 NIT in round 1, all closed; round 5 found the space-path control had been deleted by a range edit (restored byte-identical) and the title count stale (fixed). Codex round 6 on the final tree aborted on a usage limit. Opus (worktree, every deleted line read) rounds 1-6: 1 HIGH (the population growth falsified the record's depth measurement; fixtures exempted, re-measured, now asserted), 5 MEDIUM, 10 LOW, 3 NIT, all closed; round 6 on this tree MERGEABLE with 0 new findings, mutants and by-function-name test diff verified by execution. No lost rejected design, harm, measurement or trap across ~1,100 deleted prose lines. Sweep: scripts/tests 1569 passed; web 1319 passed; check-doc-narrative --all 179 files / 1 known carve-out, --diff 0 warnings; decisions_validate OK.