---

title: "Verify — "Are the claims true?"" source: "tasks/TFW-56__review_mode_removal/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 (tfw.review.min_verify_ratio) RF files claimed: 24 (20 modified + 3 deleted + 1 new evidence artifact) Files to verify: ⌈24 × 0.42⌉ = 10. Opened: 24 of 24 (100%) — escalated on discrepancy V-D1.

Verification Log

V1: .tfw/workflows/review/{code,docs,spec}.md — deleted

• RF claim: folder gone, not emptied; 15 / 12 / 12 lines.
• Actual: ls .tfw/workflows/review/ → No such file or directory. git cat-file -e HEAD:.tfw/workflows/review/code.md → does not exist in HEAD. Content recovered from 6c3c506 for the coverage audit below.
• Match: ✅

V2: .tfw/workflows/review.md

• RF claim: mode step + WAIT removed, mode-file load removed from Verify, Steps 2-8 → 1-7, Step 0 = Session Naming.
• Actual: headings at L16/52/60/78/89/100/109/116 read Step 0 Name This Session · 1 Map · 2 Verify · 3 Judge · 4 Decide · 5 Tech Debt · 6 Update Traces · 7 Knowledge Capture. Contiguous, no gaps. L65 reads "Every action in it is unconditional — verification depth is set by the ratio below, never by the kind of work under review." Only surviving "mode" is L7 "coordinator in review-locked mode" — the Role Lock, one meaning.
• Match: ✅

V3: .tfw/templates/review/judge.md

• RF claim: no Mode:, no Mode-Specific section, 10 universal rows, U2 split, S1/S2/S4 added with rates, ✅/❌/⚪ vocabulary, contrast note, two new Checkpoint items.
• Actual: grep -c "Mode" → 0. Ten table rows. Row 2 carries clauses (a) mapping integrity and (b) design soundness (4.5%). Row 8 Evidence sufficiency (16.1%), row 9 Backward compatibility (8.5%), row 10 Safety (4.0%) with "kept on consequence, not on rate" written into the row. Contrast note at L26-30. Checkpoint gains "Every ⚪ N/A carries a stated reason — no row skipped as a bare ✅?" and "Rows 7 and 8 answered separately, with different reasoning?".
• Baseline (6c3c506): 7 universal rows, Mode: {code / docs / spec} header, ## Mode-Specific Checklist section. Confirms the 7 → 10 move and the removals.
• Match: ✅

V4: .tfw/templates/review/verify.md

• RF claim: Mode: removed; new Claim & Source Checks section with the three orphaned actions, a table and a Checkpoint item.
• Actual: section present at L26-38, marked "Unconditional — every review, whatever the deliverable is. Three actions, all mandatory", wording each action and stating it feeds judge.md row 8. Table C1. Checkpoint L73 added. grep -n "Mode" → no matches.
• Match: ✅

V5: .tfw/templates/review/map.md

• RF claim: Mode: field removed.
• Actual: header carries Mindset, Test, RF, TS. No Mode:.
• Match: ✅

V6: .tfw/templates/REVIEW.md

• RF claim: Review Mode header field and mode placeholder removed; §3 realigned to ten rows matching judge.md in order; ⚪ added.
• Actual: ten rows, identical labels and order to judge.md. <!-- Add mode-specific checklist items… --> absent. §3 note states ⚪ N/A requires a stated reason.
• Baseline: 6 rows + the placeholder comment, no Evidence completeness row. Confirms RF §2 decision 5.
• Match: ✅

V7-V8: .tfw/project_config.yaml, .tfw/templates/project_config.yaml

• RF claim: default_mode removed from both; min_verify_ratio: 0.42 intact with comment; tfw.version → 1.1.0.
• Actual: project_config.yaml:59-60 → review: / min_verify_ratio: 0.42 # minimum fraction of changed files to verify (escalate to 1.0 on discrepancy). templates/project_config.yaml:63-64 → review: # ← FRAMEWORK: updated by tfw-update / min_verify_ratio: 0.42. No default_mode in either tfw.review block. tfw.version: "1.1.0" at L7. The surviving default_mode: focused at L49 / L53 is the research axis, out of scope.
• Match: ✅

V9: .tfw/workflows/config.md

• RF claim: review.default_mode row removed; one review row remains, pointer resolves.
• Actual: grep -n "review\." → single row L92 | review.min_verify_ratio | .tfw/workflows/review.md | Step 2: Verify | Min verify ratio |. Opened review.md at Step 2 — it is Verify. Pointer resolves.
• Match: ✅

V10: .tfw/conventions.md

• RF claim: §14 anti-pattern added; Review subfolder entry needed no change.
• Actual: L496 carries the anti-pattern, worded to cover both formulations and permitting retention on consequence provided the reason is written into the row. L262-264 Review subfolder entry describes stage files with no mode vocabulary. L466 "Mode files loaded at Step 2" remains — reported, see V-D2.
• Match: ✅

V11: .tfw/glossary.md

• RF claim: Reviewer heading disambiguated; entry describes one universal 10-row checklist; Principles Check pointer Step 4 → Step 3.
• Actual: L127 ### Reviewer (AI — coordinator under the reviewer Role Lock); L128 describes "one universal 10-row Judge checklist — every row asked in every review, with explicit ⚪ N/A". L142 points at review.md Step 3, which is Judge. L157 Session Naming → review.md Step 0, now true.
• Match: ✅

V12-V13: .tfw/VERSION, .tfw/CHANGELOG.md

• RF claim: 1.1.0; new entry; ### Removed names the key.
• Actual: VERSION = 1.1.0. ## [1.1.0] — 2026-08-13 inserted below ## [Unreleased] and above ## [1.0.0]. ### Removed L21 names tfw.review.default_mode with upgrade instructions ("this key is now inert… delete it from your tfw.review block") and states min_verify_ratio is unchanged. D42's revocation recorded at L22. No prior entry altered.
• Match: ✅

V14-V19: the six adapter and entry-point copies

• RF claim: all six clean of mode vocabulary; five diff runs empty; Codex skill no longer names a WAIT gate.
• Actual: re-ran the five diffs independently — .claude/commands/tfw-review.md, .agent/workflows/tfw-review.md, .claude/commands/tfw-config.md, .agent/workflows/tfw-config.md, .agents/skills/tfw-review/SKILL.md all identical to their sources. Read .tfw/adapters/codex/skills/tfw-review/SKILL.md in full: the line now reads "Follow every gate in the workflow exactly as it requires, including each stage self-check gate before advancing" — the four gates that actually remain. No WAIT-gate reference.
• Match: ✅

V20-V21: TECH_DEBT.md, README.md

• RF claim: TD-106 closed with the reason; Task Board → 🟢 RF with ONB link.
• Actual: TD-106 status ✅ Closed, reason recorded ("deleted, not annotated"), naming the renumbering carried out in the same task. README L304 shows 🟢 RF with HL / research / TS / ONB / RF links present.
• Match: ✅

V22: evidence/EV__TFW-56__review_mode_removal.md

• RF claim: 16 rows, 9 VERIFIED / 0 DEFERRED / 0 BLOCKED / 7 N/A.
• Actual: 16 rows, counts confirmed. Environment header complete. Every VERIFIED row carries a command with output or a named-file reading. The 7 N/A each quote the TS's own Evidence: field verbatim — checked against TS §5, all seven quotes are accurate.
• Match: ✅

V23: coverage audit against the deleted files (AC-3, AC-4)

• RF claim: eight mode checklist rows and eight distinct verify actions, each with a stated home.
• Actual: recovered all three mode files from 6c3c506 and counted independently. Checklist rows — 8: code Code quality · Test coverage · Security · Breaking changes; docs Content quality · Source verification; spec Analytical quality · Source attribution. Matches the RF table exactly. Verify actions — 10 entries, 8 distinct: the ratio action appears in all three files (3 → 1); code adds re-run build/test, AC×DoD cross-check, test-file check; docs adds structure-matches-spec and spot-check claims; spec adds citations-traceable and data-vs-primary-source. Matches the RF table exactly, including action 5 (structure matches spec), a fourth docs action neither HL nor TS enumerated and which the executor accounted for anyway under DoF-1. Each stated destination opened and confirmed present in the shipped file.
• Match: ✅

V24: git — history and attribution

• RF claim: 3 deletions, no framework file created, no task REVIEW file or past CHANGELOG entry edited; deletions swept into fbdf443.
• Actual: git show --stat fbdf443 contains the three review/ deletions alongside TFW-53/B's own three files — the sweep is real and exactly as described. 68a8be8 touches 23 files; the only tasks/ paths are TFW-56's own ONB, RF and EV. grep -rl "Review Mode" tasks/ now returns 43, not the RF's 41 — the two extra are this task's own RF and EV, written after the count. No pre-existing REVIEW file appears in either commit.
• Match: ✅

Commands Executed

#	Command	Result
1	python -m pytest docs/scripts/ -q	68 passed in 36.97s — re-run independently, matches RF §4
2	grep -rn "code / docs / spec\|default_mode: code\|Review Mode\|review/{code" .tfw/ .claude/ .agent/ .agents/ --exclude=CHANGELOG.md	zero matches, exit 1 — the AC-12 gate re-run independently
3	ls .tfw/workflows/review/	No such file or directory
4	grep -rniE "review mode\|review-mode\|default_mode\|mode-specific\|mode file" .tfw/ .claude/ .agent/ .agents/	hits only in the research axis, the new CHANGELOG entry, and conventions.md:466 — no review-mode residue
5	diff × 5 across the adapter copies	all empty
6	grep -rn "review\.md.*Step [0-9]\|review\.md:[0-9]" repo-wide	every pointer resolves; no stale step reference anywhere (DoF-3 clear)
7	git show --stat fbdf443 / 68a8be8	attribution anomaly confirmed exactly as the RF describes it

Claim & Source Checks

Applying the section this task ships, on its own delivery. Three claims chosen by how much the result rests on them.

#	Claim / citation checked	Where it appears	Traces to	Holds?
C1	"E2 marked VERIFIED… RESTRICT is not exercised by this corpus" — finding 1 of the S1-vs-U7 dry-run, the exhibit AC-2 passes or fails on	EV §E3	TFW-53/phase-a/evidence/EV__phase-a…md:71 — verbatim: "RESTRICT is not exercised by this corpus… recorded as a coverage gap"	✅
C2	"E13's conclusion was superseded inside the same pass by E15: no --grep form can be subject-only"	EV §E3	Same file L185 — "The finding that decided it: no --grep form can be subject-only." E13 does ship a --grep form and E15 does supersede it	✅
C3	"E11 scored 4/4 while the budget and cut-order property is absent from both the rule and this check"	EV §E3	Same file L129 — "Not tested: the earlier budget and cut-order property is absent from both the rule and this check"	✅
C4	"Both attachments resolve — baseline_recovery.txt and classification_exercise.md"	EV §E3	ls TFW-53/phase-a/evidence/ — both present	✅
C5	"41 files under tasks/ still carry Review Mode", including TFW-53/A's REVIEW header	EV §E8	Count is now 43 (this task's own RF and EV added the string after the measurement); the named header > **Review Mode**: spec is present at L6 of that REVIEW file	✅ (count drift explained, claim intact)
C6	"git log -p shows tfw.version moving in lockstep with .tfw/VERSION on every release since 0.8.5"	RF §2 decision 1	Both fields read 1.1.0 today; the historical lockstep is the executor's stated method and is plausible, but I did not re-run the full log -p. Partially verified	⚠️
C7	Measured firing rates (16.1% / 8.5% / 4.5% / 4.0%) carried inside the promoted rows	judge.md rows 2, 8, 9, 10	RES iter1 E1/E2 (637 rows / 203 reviews / 3 installs) and HL §3 coverage table — the numbers in the shipped rows match the amended §3 table exactly	✅

Discrepancies Found

V-D1 — RF §3 AC-8 states a command it did not run. The RF writes "grep -rn "review mode" .tfw/ → 0 matches". Run exactly as written, that command returns two hits, both in .tfw/CHANGELOG.md (L17 and L23 of the new 1.1.0 entry). The EV file records the correct command — grep -rn "review mode" .tfw/ --exclude=CHANGELOG.md — so the work was done properly and the RF dropped the flag when summarising.

Substance is unaffected. TS AC-8's gate is not "zero matches"; it is "every remaining hit means the same thing". Both hits are the new changelog entry describing the axis that was removed — exactly one meaning, and the changelog is the one file DoD-15 forbids rewriting. AC-8 passes on its own criterion. Severity: Low, a reporting inaccuracy in a summary line, not a coverage failure. Escalated verification to 100% on finding it; nothing further surfaced.

V-D2 — conventions.md:466 remains, by decision, not by oversight. "Mode files loaded at Step 2, not at start" is inherited from D42 and now describes only the research axis. It is true as written (research/base.md Step 2 loads focused/deep), out of AC-8's scope, and correctly reported in RF §2 decision 6 and Observation 6 instead of being silently edited. Not a discrepancy against the TS — recorded so the review does not read as having missed it. → tech debt.

V-D3 — the count in EV §E8 has drifted from 41 to 43. Cause identified: this task's own RF and EV files contain the string Review Mode and were written after the measurement. The claim the count supports — no pre-existing REVIEW file was modified — is verified independently by git show --name-only 68a8be8, which touches only TFW-56's own artifacts under tasks/. Not a defect.

No other discrepancies. All 24 files opened; every AC checkmark cross-checked against the actual file; the mandated grep gate, the build gate and the five parity diffs re-run independently.

Evidence Verification

#	RF Evidence ref	Artifact exists?	Matches claim?
E1	EV §E1 — folder absence, step contiguity	✅	✅ — ls re-run, headings re-counted 0-7
E2	EV §E2 — 10 rows, Mode count 0	✅	✅ — re-counted against the shipped file
E3	EV §E3 — S1-vs-U7 dry-run vs TFW-53/A	✅	✅ — all three findings traced to the source EV file, C1-C4 above. Rows 7 and 8 did produce different answers from different reasoning
E4-E6	N/A rows quoting TS Evidence fields	✅	✅ — all seven N/A quotes checked against TS §5, verbatim
E7	EV §E5 — both config excerpts, before and after	✅	✅ — re-run; min_verify_ratio comment byte-identical
E10	EV §E6 — six-adapter grep + five diffs	✅	✅ — re-run independently, all empty
E13	EV §E7 — the grep gate verbatim, with exit status	✅	✅ — re-run, zero matches, exit 1. The executor's finding that review/{code was a dead alternative is correct: the real string was always review/{mode}.md
E14	EV §E8 — history intact	✅	⚠️ count 41 → 43, explained (V-D3); the claim it supports verifies independently
E15	EV §E9 — diffstat + attribution anomaly	✅	✅ — git show --stat fbdf443 confirms the three TFW-56 deletions sitting in TFW-53/B's commit, exactly as described. Keeping this VERIFIED rather than softening it to DEFERRED was the right call: the fact is true and the disclosure is the point
E16	EV §E10 — 68 tests passed	✅	✅ — re-run: 68 passed

Total evidence items: 16 · verified: 16 · missing: 0.

Knowledge Citations Verified

HL §7.2 carries 26 citations. Sampled 8, weighted toward the ones the design rests on.

#	Artifact	Citation	Link resolves?	Item exists?
1	HL §7.2 #1	KNOWLEDGE.md D42 — review mode files	✅	✅ — L74, wording matches the citation
2	HL §7.2 #6	KNOWLEDGE.md D54 — adapter parity is behavioural	✅	✅
3	HL §7.2 #7	KNOWLEDGE.md D28 — one name, one behaviour	✅	✅
4	HL §7.2 #14	knowledge/philosophy.md F21 — explicit N/A	✅	✅
5	HL §7.2 #18	knowledge/process.md F19 — the Step 0 anomaly	✅	✅ — L26, and now historical (RF obs. 3)
6	HL §7.2 #19	TECH_DEBT.md TD-106	✅	✅ — L22, closed by this task
7	HL §7.2 #21	conventions.md §14 anti-patterns registry	✅	✅
8	HL §7.2 #23-25	External — Gawande Do-Confirm; 5-9 checklist band; LLM-judge dilution/order sensitivity	✅	✅ — sourced to RES iter1 2_gather G7, present in that file

Total citations: 26 · sampled: 8 · verified: 8 · hallucinations: 0.

Checkpoint

Self-check: - [x] Opened ≥ ⌈N × ratio⌉ files and recorded findings? 24 of 24 — 100%, escalated on V-D1. - [x] Ran at least 1 build/test command (or documented why not)? pytest docs/scripts/ → 68 passed, plus the grep gate and five parity diffs. - [x] Claim & Source Checks filled — 2-3 key claims spot-checked, every citation traced to a real artifact, data claims checked against a primary source (or explicit N/A with a reason)? Seven checks, C1-C7; C6 partially verified and marked as such. - [x] Each RF §3 (AC) checkmark verified against actual file? All twelve. - [x] KNOWLEDGE.md checked — contradictions with changes documented? D42 and the Legacy row are now stale — the RF reports both (obs. 4) and routes them to /tfw-docs. No undocumented contradiction. - [x] Knowledge Citations from HL §7.2 and ONB §7 verified (links resolve, items exist)? - Total citations: 26, verified: 8 sampled, hallucinations: 0 - [x] Evidence artifacts from RF §5 verified (files exist, claims match)? - Total evidence items: 16, verified: 16, missing: 0

Stage complete: YES