---

title: "Verify — "Are the claims true?"" source: "tasks/TFW-53__hl_contract_and_goal_defence/phase-c/review/verify.md"

---

Verify — "Are the claims true?"

Mindset: Auditor. The RF is a declaration, not a fact. Open files. Run commands. Compare claims against reality. Test: "If I removed the RF, would the evidence alone prove the work was done?" Min verify ratio: 0.42 (project_config.yaml → tfw.review.min_verify_ratio) RF files claimed: 9 (2 created, 7 modified) Files to verify: ⌈9 × 0.42⌉ = 4 → escalated to 9 of 9 (100%) on discrepancy D1

Verification Log

V1: .tfw/templates/review/judge.md

• RF claim: row 2 clause (a) → Purpose Check carrying its rate and its consequence reason; new Purpose Check block below the table with reference set, two invalid references, fused field, three tests, override clause, three-outcome table with routing; one Checkpoint item. +42 / −1.
• Actual: all of it, at lines 16 and 32-70, Checkpoint item at line 84. git show --numstat → 43 1 (RF says +42 / −1; the extra insertion is the blank separator line — immaterial). The block sits below the table and above ## Contradictions with KNOWLEDGE.md, so the row stays one quotable sentence per clause. Row 2(b) reads "Design soundness (4.5%): is the design itself sound against HL §7 principles" — rate intact, separately answered, {(b) answered separately} in the Evidence column.
• Match: ✅

V2: .tfw/workflows/review.md

• RF claim: line 28 → frozen baseline + rule-15 pointer; Reviewer Identity gains goals/values/north star and block authority plus not rubber stamp; Step 3's HL §7 Principles check paragraph replaced by the Purpose Check instruction; Step 4 gains a Routing block. 1,065 → 1,176 words.
• Actual: line 28 confirmed; identity at 35-38 reads "Quality guardian, not rubber stamp … goals, values and the north star are yours to defend, and they alone can ground a block"; line 87 is the replacement, and the word mapping does not appear in the file; line 102 is the Routing block naming both failure outcomes and both owner routes, closing with "Verdict vocabulary unchanged." wc -w reproduces 1,176; git show HEAD~1:… | wc -w reproduces 1,065.
• Match: ✅

V3: .tfw/templates/REVIEW.md

• RF claim: §3 row 2 realigned to judge.md; the not fit for purpose finding surfaced inside the existing ### If REJECT block, no new section.
• Actual: ten rows, row 2 = "Purpose Check — is this what we set out to do? + design soundness". The finding block is a blockquote immediately after ### If REJECT, carrying the citation-and-harm requirement, the "quality checks passed" sentence, the owner route and the contract-defect branch. No heading added. The abbreviated wording is the file's pre-existing convention — rows 6-10 are all abbreviations of judge.md's longer rows, and the file's own note asks for "ten rows, matching review/judge.md one-for-one and in the same order", which is row alignment, not wording identity.
• Match: ✅

V4: .tfw/glossary.md

• RF claim: PV Index gains priority 0; priority 1 relabelled .tfw/README.md § Values and Principles — methodology values with its reason; one note that priorities 0 and 1 may name the same file.
• Actual: all three present. Rows 2-7 byte-identical in the diff. Row 1 keeps its original content ("Core beliefs (e.g., Traces Over Code, Structural Enforcement)") and gains the byte-identical-across- projects reason. The new label resolves: .tfw/README.md:82 is ## Values and Principles. The "Who scans PV" block below survives unchanged and still resolves after renumbering.
• Match: ✅

V5: .tfw/conventions.md

• RF claim: §3 gains Project North Star with seven rules (locus, no nominated task HL, payload with mandatory non-goals, admission criteria, fallback chain, same-file rule, citation namespace); §14 gains two review-side anti-patterns; +27 / −0.
• Actual: seven numbered rules confirmed at the ### Project North Star block after rule 21; both §14 entries present. git show --numstat → 27 0. Phase A's rules 1-21 and TFW-56's §14 entry untouched — the diff hunks are pure insertions.
• Match: ✅

V6: .tfw/templates/HL.md

• RF claim: north-star header field below the contract block, list-valued, with fallback and explicit N/A; +7 / −0; contract block untouched.
• Actual: one additive hunk after the contract block's own "Add further header fields below this block, not inside it" line. The field renders {one or more designated README sections — e.g. …} or N/A — no project north star designated, followed by an explanatory blockquote naming the fallback and F21. Zero deletions.
• Match: ✅

V7: .tfw/compilable_contract.md

• RF claim: §2 pattern table only; P{N} corrected from KNOWLEDGE.md §0 to HL §7 Principles row (task-local); PP{N} and NS{N} added; +3 / −1.
• Actual: lines 59-61 exactly as claimed; git show --numstat → 4 2, of which the §2 table accounts for 3 1 and the remaining 1 1 is the CRLF/line-ending normalisation the repo warns about — no other content line changed. Line 65's second stale KNOWLEDGE.md §0 reference survives, as RF §6 obs. 2 says.
• Match: ✅

V8: phase-c/evidence/EV__phase-c__goal_defence_in_review.md

• RF claim: environment header, 13 evidence rows, verdict, the review.md word ledger as Exhibit A.
• Actual: all present. Every gate command in E1-E12 was re-run by this reviewer and every output reproduces (see Commands Executed). The two disclosed divergences (E1 clause (b), E3 recovery form) are stated in the rows and repeated in the verdict block rather than buried. Exhibit A's removal column is genuinely empty and the arithmetic closes: 18 + 34 + 11 + 48 = 111, and 1,065 + 111 = 1,176.
• Match: ✅

V9: phase-c/evidence/purpose_check_replay.md

• RF claim: the Purpose Check as shipped, replayed against nine reviews with every citation-and-harm field filled; 5 of 6 non-approve on the rejected corpus, 0 of 3 on the sound one.
• Actual: nine rows present, each with the fused field filled, the three tests answered and an outcome. Fourteen quoted clauses and quantitative claims were traced to git history (see Claim & Source Checks) — twelve are verbatim-correct. One row's classification does not survive its own source: 49/A (D1). One quotation is a paraphrase presented inside quotation marks (D2).
• Match: ⚠️ partial — the corpus, the citations and the arithmetic hold; row 49/A's conclusion does not

Commands Executed

#	Command	Result
1	grep -rc "mapping integrity" .tfw/templates/review/judge.md .tfw/workflows/review.md	judge.md:0, review.md:0 — reproduces AC-1
2	grep -c "^| [0-9]" .tfw/templates/review/judge.md	10 — reproduces the ten-row count
3	grep -n "frozen baseline\|not fit for purpose" .tfw/workflows/review.md	lines 28, 87, 102 — reproduces AC-4
4	wc -w .tfw/workflows/review.md / git show HEAD~1:… \| wc -w	1,176 / 1,065 — reproduces AC-5 exactly
5	grep -n "P{N}\|NS{N}\|PP{N}" .tfw/compilable_contract.md	lines 59, 60, 61 (+ line 81, the resolution list — RF §6 obs. 3)
6	git show --numstat HEAD -- .tfw/conventions.md	27 0 — reproduces AC-10
7	grep -rn "git log --format" .tfw/ --include="*.md"	two hits: conventions.md:77 and templates/HL.md:10 — reproduces AC-3's disclosed counterexample
8	python -m pytest docs/scripts/ -q	68 passed in 32.81s — reproduces the RF's 68
9	git diff e8ee76e HEAD -- …/[HL-TFW-53](../../HL-TFW-53__hl_contract_and_goal_defence.md)*.md	HL changes since the baseline are the status field, the Step 3/Step 4 stale-pointer correction, the deliverable renumber 4→5-8, one §8 dependency row, S38 and the A14 row — all in free sections or non-substantive under conventions.md §3 rules 6/7. No unlogged frozen-section edit.
10	git status --short	README.md modified and uncommitted, carrying both this phase's RF link and the TFW-55 row — matches RF §2 decision 8

Claim & Source Checks

Escalated well past the mandated 2-3: the replay is AC-11's entire evidence base and every row of it is a quotation from a reverted task's contract, so each was traced to the commit the replay names.

#	Claim / citation checked	Where it appears	Traces to	Holds?
C1	"smaller where precision and references can replace prose"	replay 48/A	721ca15:…HL-[TFW-48](../../../TFW-48__value_first_methodology_rebaseline/)…:15 — verbatim	✅
C2	DoF-12 "adds another conceptual layer or document that duplicates an existing owner"	replay 48/A, 48/C	721ca15:…HL-[TFW-48](../../../TFW-48__value_first_methodology_rebaseline/)…:369 — verbatim	✅
C3	DoD-6 "Planning preserves user insights, product requirements, applicable Project Values, and uncertainty…"	replay 48/B	721ca15:…HL-[TFW-48](../../../TFW-48__value_first_methodology_rebaseline/)…:341 — verbatim	✅
C4	DoD-12 "Every claimed deliverable has local proof…"	replay 48/C	721ca15:…HL-[TFW-48](../../../TFW-48__value_first_methodology_rebaseline/)…:347 — verbatim	✅
C5	"transitional attention signals" (the Phase E adjacency finding)	replay 48/C	721ca15:…phase-c/REVIEW…:20 and TS…:42/97/168	✅
C6	§1 "The identity remains readable without special tooling"	replay 49/A	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:15-16 — verbatim, but the sentence continues	⚠️ see D1
C7	"This is provenance, not decoration"	replay 49/A	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:19 — verbatim	✅
C8	DoD-3 "A versioned structural validator rejects malformed or missing identity…"	replay 49/A	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:200 — verbatim	✅
C9	DoF-8 "Enforcement depends only on agent compliance prose or only on unversioned .git/ state"	replay 49/A, 49/C	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:235 — verbatim	✅
C10	DoD-4 "observable point-of-action consumer of the canonical contract"	replay 49/B	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:203 — verbatim	✅
C11	P11 "agents see the local imperative and one valid example at commit time"	replay 49/B	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:270 — verbatim	✅
C12	"3,160 lines" router	replay 49/B	721ca15:…phase-b/REVIEW…:38 — 719 + 1,066 + 1,296 + 79 = 3,160	✅
C13	"1,708 lines" validator	replay 49/A	721ca15:…phase-a/RF…:47 / REVIEW…:74	✅
C14	Phase C deliverable 1 "Replace or safely bypass the current local branch-prefix hook…"	replay 49/C	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:182 — verbatim	✅
C15	"the independent reviewer commit 1ebb680 recorded 7 of 10 Judge checks FAIL and was overwritten three commits later"	replay 49/C	1ebb680 exists — [codex/TFW-49/phase-c/reviewer] request runtime lifecycle corrections, writes phase-c/review/judge.md	✅
C16	"preserves all ten mapped principles" — offered as the shipped review's wording	replay 48/A note	721ca15:…phase-a/review/judge.md:12 reads "preserves Phase HL P1–P10"	⚠️ see D2
C17	The 48/A analysis: the shipped row mapped against a Phase HL that dropped master principles	replay 48/A note	Confirmed — 721ca15:…phase-a/HL__phase-a…:145-154 authors its own ten principles; the master HL carries thirteen, and master P7/P10/P12 are absent from the phase list	✅ substance correct
C18	§1 "One precise Markdown rule achieves this without enforcement software"	replay TFW-50	tasks/[TFW-50](../../../TFW-50__minimal_agent_commit_attribution/HL-TFW-50__minimal_agent_commit_attribution.md)…/[HL-TFW-50](../../../TFW-50__minimal_agent_commit_attribution/HL-TFW-50__minimal_agent_commit_attribution.md)…:9 — verbatim	✅
C19	§1 "Codex becomes a first-class TFW adapter with dedicated shortcut skills … matching the adapter parity already achieved"	replay 47/B	tasks/[TFW-47](../../../TFW-47__codex_adapter_shortcut_skills/HL-TFW-47__codex_adapter_shortcut_skills.md)…/[HL-TFW-47](../../../TFW-47__codex_adapter_shortcut_skills/HL-TFW-47__codex_adapter_shortcut_skills.md)…:14 — verbatim	✅
C20	The 42/A vision clauses and "know exactly what to investigate"	replay 42/A	tasks/[TFW-42](../../../TFW-42__research_cycle_restructure/HL-TFW-42__research_cycle_restructure.md)…/[HL-TFW-42](../../../TFW-42__research_cycle_restructure/HL-TFW-42__research_cycle_restructure.md)…:10 — every clause present	✅
C21	The 42/A near-miss: the guidance table the owner later removed as tautological	replay 42/A	knowledge/process.md F22 — "Generic capability guidance tables … tautological overhead — removed by user as valueless", TFW-42 session, 2026-04-30	✅
C22	"~4 blocks in 149 reviews" inside the shipped row	judge.md:16	HL §4 Phase C design note at baseline: "base rate ~4 blocks in 149 reviews". No project named in the template (F13); the row flags it as "a different corpus" against the file header's 637-row figure	✅

Discrepancies Found

D1 — Replay row 49/A: the third-outcome classification does not survive its own source ❌

The row rules TFW-49 Phase A a contract defect (the third outcome) on this reasoning:

§1 Vision promises "readable without special tooling" and "This is provenance, not decoration", while DoD-3 mandates a versioned structural validator and DoF-8 makes "enforcement depends only on agent compliance prose" a failure condition. The minimal Markdown solution the Vision implies is the solution DoF-8 forbids. Both clauses are owner-approved and cannot be jointly satisfied.

The quoted §1 sentence, read to its end at 9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:15-17:

"The identity remains readable without special tooling, while structural validation prevents quiet drift between Coordinator, Researcher, Executor, Reviewer, adapters, and repositories."

§1 asks for both properties in one sentence — human-readable identity and structural validation. It does not imply a Markdown-only solution; it names validation as the companion of readability. DoD-3's validator is therefore the §1 clause discharged, not a contradiction of it, and DoF-8 (which forbids enforcement resting only on prose or only on unversioned .git/ state) is satisfied by any versioned check. There is no pair of clauses here that cannot be jointly satisfied. The reference set is not internally inconsistent, so the third outcome is not the correct outcome for this row.

The replay's own supporting argument points the same way once tested. It offers TFW-50 as "confirmation from outside this replay" — TFW-50 shipped "one precise Markdown rule … without enforcement software" and was approved. But TFW-50 ran under its own HL; a prose-only rule would have failed TFW-49's DoF-8 outright. What TFW-50 confirms is that the owner's preference changed between two contracts, not that one contract contradicted itself. And the row itself concedes the excess reading is "arguable at most … the clause asked for one" — which is the honest reading, and it lands on ✅ aligned, exactly where research iteration 2 put it before amendment A6 existed.

Four consequences, sized:

#	What it touches	Effect
1	AC-11's count	5 of 6 → 4 of 6 non-approve (48/A, 48/C, 49/B, 49/C). Pass condition is "≥1 non-approve" → AC-11 still passes, and the 0-of-3 sound-corpus result is untouched
2	RF §3 AC-12 "Exercised for real: replay row 49/A"	Unsupported. AC-12's gate is textual ("read the block; routing target is the owner; reachable without leaving judge.md") and passes independently, and its Evidence field says a replay row is "expected", not required → AC-12 still passes on its gate, but the RF sentence overclaims
3	The shipped judge.md precedent line	This is the material one. Every install now reads "one rejected task's approved §1 promised 'readable without special tooling' while its approved DoD required a versioned structural validator" as the canonical illustration of a contract defect. The illustration is built on a half-sentence, so it teaches reviewers to read surface tension as unsatisfiability — the opposite of the discipline the same block demands three paragraphs above ("a citation that resolves but is irrelevant fails the row")
4	RF §8 S2 and the KNW pipeline	S2 ("the check found its own predecessor's defect while being validated on it") has no basis left, and the derived implication (the replay is the check's regression suite) rests on it. S2 is High-confidence material headed for KNOWLEDGE.md at KNW — approving it writes a false fact into project memory

Escalation triggered: verification raised from 4 files to 9 of 9.

D2 — A paraphrase inside quotation marks ⚠️

The 48/A note attributes to the shipped review the words "preserves all ten mapped principles". The source reads "preserves Phase HL P1–P10" (721ca15:…phase-a/review/judge.md:12). The substance is right — and C17 confirms the deeper claim independently — but the marks say verbatim and the words are not. Low severity on its own; it shares a shape with D1, and in the phase whose thesis is alignment must be cited, not asserted the form is part of the deliverable.

D3 — HL §4 Phase C deliverable 2 no longer describes what shipped ⚠️ (coordinator-side)

The frozen deliverable reads "Locus: a designated section of the root README.md". The shipped conventions.md rule 1 reads "designated section(s) of a README" and permits more than one location, on the owner's ONB Q2 ruling, recorded at HL §11 S38.

The widening is legitimate: conventions.md §3 rule 6 makes deliverable lists inside an approved phase free, and the tripwire clears — no §5 or §6 item as it stands is breached (DoD-18 and DoD-19 name no locus, DoF-14 is satisfied). But the deliverable text itself was never refined to match, so the HL now describes a narrower rule than the framework ships. Not the executor's to fix — the role lock forbids it, and this reviewer's too. → coordinator.

D4 — The adapter copy of review.md is stale, and this review is the proof ⚠️ (known, in flight)

This review session was invoked through .claude/commands/tfw-review.md, which still carries "Reviewer Identity: Quality guardian." and the deleted "HL §7 Principles check" paragraph. The canonical .tfw/workflows/review.md was followed instead, per AGENTS.md. Out of scope by design (Phase D owns the sync, TS §2), already recorded as TD-157 and predicted in ONB §5 risk 4 — logged because the window is now demonstrably live, not theoretical, and because a reviewer who trusted the adapter copy would have run the retired check.

Evidence Verification

#	RF Evidence ref	Artifact exists?	Matches claim?
E1	AC-1 gates in EV	✅	✅ — both greps and the row count reproduce; the clause-(b) divergence is disclosed in the row itself
E2	purpose_check_replay.md §5	✅	✅ — passing and failing forms both present; the failing form is rejected by the row's own quoted wording
E3	recovery-command gate	✅	✅ — two hits reproduce, and the row states the gate does not literally clear rather than reporting a pass
E4	review.md line greps	✅	✅ — lines 28, 87, 102; verdict set unchanged at line 100
E5	word ledger, Exhibit A	✅	✅ — 1,065/1,176 reproduce; ledger arithmetic closes; removal column empty by construction
E6	PV Index + conventions.md §3	✅	✅ — rows 2-7 byte-identical; seven rules present; new label resolves to a real heading
E7	templates/HL.md diff	✅	✅ — one additive hunk, zero deletions, below the contract block
E8	REVIEW.md row alignment	✅	✅ — ten rows, row 2 realigned, no new section
E9	compilable_contract.md numstat	✅	✅ — table-only change confirmed line by line
E10	conventions.md numstat 27 0	✅	✅ — reproduces exactly
E11	the replay	✅	⚠️ — 8 of 9 rows hold; row 49/A's classification fails on its own source (D1). AC-11's pass condition survives at 4 of 6
E12	third outcome + routing	✅	⚠️ — the block, the routing and the reachability all hold; "exercised for real: replay row 49/A" does not
E13	build gate	✅	✅ — 68 tests re-run and pass; the changed files are docs-pipeline sources and no warning names them

Total evidence items: 13, verified: 11, partial: 2 (E11, E12), missing: 0.

On the RF's 13/13 VERIFIED verdict: eleven rows are clean, and the two divergences the RF does disclose (E1, E3) are disclosed well — named in the row, repeated in the verdict, not downgraded to look tidy. The defect is not concealment; it is that E11 and E12 rest on a reading of one source that the source does not support, and nothing in the evidence file re-tested it.

Knowledge Citations Verified

ONB §7 carries 26 items from HL §7.2 plus six new ones (N1-N6). Every referenced record was checked for existence in its named file.

#	Artifact	Citation	Link resolves?	Item exists?
1	HL §7.2 / ONB §7	KNOWLEDGE.md D19, D20, D23, D24, D31, D43, D46, D49, D54, D55	✅	✅ all ten
2	ONB §7 N1, N3	KNOWLEDGE.md D61, D53	✅	✅ both
3	HL §7.2 / ONB §7	knowledge/philosophy.md F4, F13, F21, F22, F25	✅	✅ all five
4	ONB §7 N4	knowledge/philosophy.md F24	✅	✅
5	HL §7.2 / ONB §7	knowledge/process.md F4, F6, F11, F14, F19, F20, F25	✅	✅ all seven
6	replay 42/A	knowledge/process.md F22	✅	✅
7	HL §7.2 / ONB §7	knowledge/constraint.md F2	✅	✅
8	ONB §7 N2	knowledge/constraint.md F4	✅	✅
9	ONB §7	.tfw/README.md § Structural Enforcement, § Naming Creates Behavior, § Candor Over Flattery	✅	✅ headings at 100, 104, 88 as claimed
10	ONB §7	conventions.md §7, §15	✅	✅

Total citations: 32, verified: 32, hallucinations: 0.

KNOWLEDGE.md contradiction check. D61 (TFW-56's universal ten-row checklist, measured rates carried inside each promoted row) is respected, not contradicted: ten rows kept, and row 2(a) carries its figure inside the row in the rows 8-10 grammar. D46 (Reviewer Identity) is completed rather than contradicted — its recorded text was "Quality guardian, not rubber stamp" and only half had shipped. D43 (citation-with- link as the anti-hallucination device) is extended by the harm half, which is the SS2 gap. No contradiction found.

Checkpoint

Self-check: - [x] Opened ≥ ⌈N × ratio⌉ files and recorded findings? — 9 of 9, escalated to 100% on D1 - [x] Ran at least 1 build/test command (or documented why not)? — pytest re-run, 68 passed; ten commands logged - [x] Claim & Source Checks filled — 22 claims traced, 20 verbatim-correct, 2 flagged (D1, D2) - [x] Each RF §3 (AC) checkmark verified against actual file? — all twelve, gate commands re-run - [x] KNOWLEDGE.md checked — contradictions with changes documented? — none found; D61/D46/D43 examined explicitly - [x] Knowledge Citations from HL §7.2 and ONB §7 verified (links resolve, items exist)? - Total citations: 32, verified: 32, hallucinations: 0 - [x] Evidence artifacts from RF §5 verified (files exist, claims match)? - Total evidence items: 13, verified: 11, partial: 2, missing: 0

Stage complete: YES

---

Verify — second pass (corrective, 2026-08-13)

Scope: the corrective pass only. Commits 2642e81 (coordinator: HL §2, §9, §12 note, TD-168), 7e5311d (coordinator: TS AC-13 to AC-15 + four DoF items), bd032a5 (executor: judge.md, RF, EV, replay). Files changed: 4 — one framework file and three task artifacts. Ratio 0.42 → 2 required; all 4 opened, plus the HL and TS the coordinator changed, for 6 of 6. First-pass findings under re-test: D1 (49/A truncation), D2 (48/A quotation), D3 (HL deliverable 2).

Verification Log — second pass

V10: .tfw/templates/review/judge.md — AC-14

• RF claim: the blockquote no longer offers the withdrawn reading as the canonical contract defect; it states the bar instead; no substitute example was hunted for; 1,167 → 1,165 words, paid inside the same block; review.md untouched.
• Actual: the old precedent blockquote is gone. In its place, lines 66-70: "Two clauses conflict only if satisfying one necessarily violates the other. Read each to the end of its sentence: a clause that qualifies itself in its second half is not in tension with the clause that discharges it. Surface tension is not inconsistency — and a contract that is coherent but wrong for the product is the purpose question above, not this one. No instance has been observed in the corpus this check was validated against." wc -w reproduces 1,165; git show HEAD~1:… | wc -w reproduces 1,167. review.md reproduces 1,176, byte-identical to the first pass.
• Match: ✅ — and the replacement is stronger than the thing it replaces. The old line gave a reader one example to pattern-match; the new one gives the discriminating test, and the test is exactly the one whose absence produced D1. A gate that teaches the error it was built from is the better artifact.

V11: .tfw/templates/review/judge.md — what the trim removed

• RF claim: "paid inside the same block" — no ledger given.
• Actual: diffed line by line. Four sites, all inside the Purpose Check block, all rhetorical emphasis: "— naming them is the point of this row" · "— the fallback chain makes that hard to write honestly" · "Three outcomes, not two. The Status column keeps…" → "Three outcomes. Status stays…" · three outcome-table cells compressed ("the filled citation-and-harm field" → "the filled field", the fused field being defined in bold two paragraphs above). No mechanism lost: the reference set, both invalid references with their reasons, the fallback chain, all five failing conditions, the three tests, the override clause and all three outcomes with their routing are intact and verified present.
• Match: ✅ substantively — see second-pass discrepancy D6 on the missing ledger.

V12: purpose_check_replay.md — AC-13

• RF claim: row 49/A re-scored with the full sentence quoted, outcome ✅ aligned, excess argued and rejected, the TFW-50 confirmation dropped, the correction recorded with its cause named.
• Actual: §0 carries a Correction of record quoting §1 to the end of the sentence and naming the cause ("a quotation ended early"), and states that the shipped check and research iteration 2 now agree on all nine rows. Row 49/A returns ✅ aligned. The excess rebuttal cites DoD-7 — "repository fixtures … across all four TFW roles and at least two agent surfaces" including "search/filter behavior" — which I traced to 9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:206-208 and is verbatim. The TFW-50 paragraph is gone. The §3 diagram, the §3 AC-11 condition table and the §4 reconciliation all carry 4 of 6.
• Match: ✅ — and the row is now the most informative in the file. It records that 49/A is aligned with its approved contract while the owner rejected the product, and routes that case to the north star (PV priority 0). That is empirical support for the HL's own deliverable weighting, which the first pass did not have.

V13: EV__phase-c…md — AC-13 propagation

• RF claim: E11 re-scored to 4 of 6 with the cause named, E12 narrowed to the textual gate, new E14 DEFERRED with the blocker named; verdict 13/14 VERIFIED, 1 DEFERRED.
• Actual: all three present and worded as claimed. E14's blocker: "no such case exists in the nine-review corpus", with the Phase B precedent for an unexercised branch cited. The verdict block gained E14 as a third disclosed divergence.
• Match: ✅ — see D5 on row ordering.

V14: RF__phase-c…md — AC-13, AC-15

• RF claim: eight sites propagated; S2 rewritten; S4 added; §6 obs. 7 records the 48/A finding.
• Actual: §2 decision 6 rewritten with the cause and a "how the error got in" paragraph naming TD-166 as its structural half · §3 AC-11 → 4 of 6, AC-12 → gate passes / branch unexercised · §5 → 13/14 with E14 explained · §8 S2 rewritten with the withdrawal marked inline, S4 added · §9 both affected diagrams corrected, each carrying a "corrected on the second pass" note · a second AC table for AC-13 to AC-15 · DoF count 16, which matches TS §7 (grep -c "^- ❌" → 16).
• Match: ✅ 8 of 8 sites

V15: [HL-TFW-53](../../HL-TFW-53__hl_contract_and_goal_defence.md)…md — coordinator, and the harder question

• RF claim (coordinator's): §2's evidence row and §9's risk row corrected; a correction of record appended below §12; A6's row not rewritten; TD-168 closed by refining deliverable 2.
• Actual: §2's row now carries the strikethrough and the withdrawal inline, with the surviving claim stated. §9's probability reads Unmeasured (was Medium) with zero instances recorded. The §12 note is appended below the table and A6's row is byte-identical to the baseline. Deliverable 2 now reads "designated section or sections of a README" with the rule-6 reasoning inline.
• Contract discipline re-checked, because a frozen section was touched. §2 is free (header: "Free: §2 · §7.2 · §8 · §9 · §10 · §11"), §9 is free, §12 is append-only and was appended to. Deliverable 2 sits in frozen §4, and conventions.md §3 rule 6 makes deliverable lists inside an already-approved phase free; the tripwire clears, since no §5 or §6 item as it stands is breached by permitting more than one location. git diff e8ee76e HEAD on the HL shows no edit to §1, §3, §5, §6 or §7.
• Match: ✅ — D3 closed. The refusal to rewrite A6's row is the correct call and the harder one: it keeps the log honest about what was believed when the amendment was ruled.

V16: TS__phase-c…md — coordinator

• RF claim (coordinator's): AC-13 to AC-15 added with gates; the scope of the pass stated as a negative first; four DoF items added.
• Actual: all present. The "What is not reopened" paragraph names the mechanism, the corpus, the seven framework files and AC-1 to AC-12 explicitly. Four new DoF items, including "judge.md grows on a corrective pass, or a corrected quotation is fixed without naming why it was wrong".
• Match: ✅ — and AC-13's bullet list names all eight propagation sites, which is why none was missed.

Commands Executed — second pass

#	Command	Result
11	wc -w .tfw/templates/review/judge.md / git show HEAD~1:… \| wc -w	1,165 / 1,167 — AC-14's no-growth gate reproduces
12	wc -w .tfw/workflows/review.md	1,176 — untouched, as AC-14 requires
13	grep -rc "mapping integrity" …judge.md …review.md	0 / 0 — still clear after the trim
14	grep -c "^| [0-9]" .tfw/templates/review/judge.md	10 — still ten rows
15	grep -c "^- ❌" TS §7	16 — matches RF's "all sixteen clear"
16	git diff e8ee76e HEAD -- …[HL-TFW-53](../../HL-TFW-53__hl_contract_and_goal_defence.md)*.md	frozen §1/§3/§5/§6/§7 untouched; §4 deliverable 2 refined under rule 6 with reasoning inline; A6's row byte-identical
17	python -m pytest docs/scripts/ -q	68 passed in 58.66s
18	python -m mkdocs build -f docs/mkdocs.yml	built in 51.92s, no new warning attributable to the four changed files
19	for f in templates/review/*.md templates/*.md workflows/review.md; do wc -w; done	map 159 · verify 618 · judge 1,165 · REVIEW 679 · TS 533 · RF 593 · review.md 1,176 — see D4

Claim & Source Checks — second pass

#	Claim / citation checked	Where it appears	Traces to	Holds?
C23	§1 quoted to the end: "readable without special tooling, while structural validation prevents quiet drift between Coordinator, Researcher, Executor, Reviewer, adapters, and repositories"	replay §0, row 49/A; HL §2, §12 note	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:15-17 — verbatim, full sentence	✅
C24	DoF-8 "Enforcement depends only on agent compliance prose or only on unversioned .git/ state"	replay §0	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:235-236 — verbatim	✅
C25	DoD-7 "repository fixtures … across all four TFW roles and at least two agent surfaces", incl. "search/filter behavior" — the new excess rebuttal	replay row 49/A	9e19a4f:…HL-[TFW-49](../../../TFW-49__agent_commit_identity_and_attribution/)…:206-208 — verbatim	✅
C26	"preserves all ten mapped principles" — the executor's counter-claim that it is verbatim	replay 48/A note; RF §6 obs. 7	721ca15:…phase-a/REVIEW__phase-a__method_kernel.md:51 (§4 Verdict) — verbatim	✅ executor is right
C27	"Phase HL P1–P10 all pass through their mapped ACs" — the sentence the first pass offered as the true source	replay 48/A note; RF §6 obs. 7	721ca15:…phase-a/REVIEW__phase-a__method_kernel.md:38 (Judge row 2) — verbatim, different sentence, same file	✅ both exist
C28	"the same verdict research iteration 2 recorded" for 49/A	replay row 49/A, §0	research/iter2/ replay recorded 49/A as passes — consistent with the first pass's own §0 statement of the divergence, which is now withdrawn as an artifact of the truncation	✅

Discrepancies Found — second pass

D1 — CLOSED. Corrected, propagated, and the cause named

Every one of the four consequences the first pass sized is discharged: AC-11 recounts to 4 of 6 (condition ≥1 holds), AC-12 is restated as gate passes, branch unexercised, the shipped illustration is replaced by the bar, and S2 is rewritten to what survives. The correction is recorded as a correction in five artifacts — replay §0 and §1, HL §2 and §9, the §12 note, EV E11/E14, RF §2 decision 6 — with the cause stated each time. Nothing was silently repaired.

D2 — WITHDRAWN. This was my error, not the executor's ❌→✅

The first pass claimed "preserves all ten mapped principles" was a paraphrase inside quotation marks. It is not. It is verbatim at REVIEW__phase-a__method_kernel.md:51. I compared it against phase-a/review/judge.md:12 — the stage file, a different document — found different wording there, and called the quotation inaccurate without checking the file the quotation came from. Both sentences the executor now cites are real, at the lines given (C26, C27).

The executor refused the fix, verified both sources, added line numbers to each half, and filed RF §6 obs. 7 saying so — including that TS AC-15's second bullet "asks for a fix that would have introduced the defect it was written to remove". That is the correct handling of a wrong instruction from a reviewer, and it is the same discipline the phase ships: the claim was checked against the source rather than obeyed.

D3 — CLOSED. HL deliverable 2 refined under rule 6, reasoning inline, TD-168 closed (V15)

D4 — judge.md is now the largest artifact of its class in the repository ⚠️ (new, Low)

Artifact	Words	Note
templates/review/map.md	159
templates/TS.md	533
templates/RF.md	593
templates/review/verify.md	618
templates/REVIEW.md	679
templates/review/judge.md	1,165	639 before this phase — +82% in one phase
workflows/review.md	1,176	F2 hard threshold 1,200

The Judge template is now 1.7× the next-largest template and effectively the same weight as the workflow it serves, while being the artifact a reviewer fills in every review. This is not a violation: frozen DoD-28 directs the mechanism into judge.md precisely because "a template is not a workflow" and is off F2's budget, and HL §4 states that choice explicitly. The coordinator also installed the right rule the moment it mattered — AC-14 forbids growth on a corrective pass, and the executor met it at −2 words.

What is new is the measurement. TD-141 records that the attention budget is numeric for workflows only, and that the classes with no number are where growth lands; TD-140 measured templates/HL.md at +76% in one phase as the first instance. This is the second, larger instance, in the template with the highest fill frequency in the framework. → appended to TD-141 as evidence, not filed as a new defect.

D5 — EV row order runs E1…E12, E14, E13 ⚠️ (new, Low)

E14 was inserted above E13 rather than appended. Evidence rows are addressed by identifier — RF §4 still reads "AC gates reproduced with their commands and outputs in EV E1-E13" while E14 exists outside that range. Cosmetic, no claim affected. → /tfw-docs, on the TD-146 precedent.

D6 — the judge.md trim has no ledger ⚠️ (new, Low)

AC-14 required no-net-growth and did not require a ledger, so this is not a gate miss. But AC-5 required one for review.md for exactly this risk — a trim that quietly removes the sole statement of a mechanism — and the asymmetry means a reviewer must diff the file to learn what was cut. I diffed it (V11): four rhetorical removals, nothing load-bearing. Recorded so the next corrective pass on a template carries the two lines.

Evidence Verification — second pass

#	RF Evidence ref	Artifact exists?	Matches claim?
E11	replay, re-scored	✅	✅ — 4 of 6 reproduces in all five places it appears; cause named
E12	third outcome, textual gate	✅	✅ — narrowed honestly from the first pass's overclaim
E14	third outcome unexercised	✅	✅ — DEFERRED with the blocker named; the status is auditable, which is the point of using one
E13	build gate	✅	✅ — re-run: 68 passed, site builds

Total evidence items: 14, verified: 13, deferred: 1 (E14, blocker named), missing: 0. The 13/14 VERIFIED, 1 DEFERRED verdict is now accurate.

Checkpoint — second pass

Self-check: - [x] Opened ≥ ⌈N × ratio⌉ files? — 6 of 6 (4 changed + HL + TS), ratio required 2 - [x] Ran at least 1 build/test command? — pytest 68 passed, mkdocs built; nine commands logged - [x] Claim & Source Checks filled? — 6 new claims traced, all 6 hold, including the one that overturns my own first-pass finding - [x] Each RF §3 (AC) checkmark verified? — AC-13, AC-14, AC-15 against their gates; AC-1 to AC-12 re-confirmed unchanged - [x] KNOWLEDGE.md checked? — no new contradiction; S2's rewrite removes the false fact before KNW - [x] Knowledge Citations verified? — 32 of 32 from the first pass unchanged; 6 new source checks added - [x] Evidence artifacts verified? — 14 items, 13 verified, 1 deferred with its blocker named

Stage complete: YES