---

title: "Verify — "Are the claims true?"" source: "tasks/TFW-60__conflict_resistant_shared_workspace/phase-aa/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: 33 — 1 created, 7 moved, 25 modified Files to verify: ⌈33 × 0.42⌉ = 14 Discrepancies found → escalated to 100%. All 33 files opened or diffed; the whole change set (git diff f14f744~1 HEAD, 49 paths including this task's own artifacts) was enumerated and reconciled against the census.

Verification Log

V1: .tfw/scripts/gen_index.py · migrate_board.py · their two test files — AC-1, AC-3, AC-4, AC-5, AC-8, AC-9

• RF claim: moved with git mv, history follows; marker-based root resolution; --check {index,tasks,project}; board locator and committed-revision default on one code path; unmatched directories reported; key-naming validator; ASCII messages; tolerant streams; +33 tests
• Actual: all four at .tfw/scripts/; docs/scripts/ retains only gen_docs.py, test_gen_docs.py, test_integration.py. git diff --name-status records R063/R069/R075/R065 — renames, not copy-and-delete. find_project_root (L75–100) walks upward for .tfw/, skips any candidate containing .upstream, and raises SystemExit with no fallback when no root is found. git grep 'parents\[2\]' over the payload returns only a docstring explaining the old answer and two PROJECT_ROOT constants inside test files. read_board (L648) is one function taking board, revision, working_tree; the default path is git show {revision}:{board} and a non-zero return refuses, naming --working-tree and printing the revision it tried. set(gen_index.CHECKS) == {"index","tasks","project"}; "--validate" and "--doctor" appear nowhere as argparse strings
• Match: ✅

V2: .tfw/migrations/2.0.0.md — AC-2, AC-4, AC-6, AC-8

• RF claim: the one created file, written for a project that is not this one; ordering constraint stated where a reader would violate it; names build.*, the quiescence rule and the task_containers decision; tells the operator an unresolved directory may be renamed by hand; --check project as the last step
• Actual: read in full. No [workspace, tasks], no 7,505 references, no 666 files — every path and command is one a receiving project has. "The order" states migrate, then generate, then remove the board at the head and repeats the consequence at each step, with step 5 carrying "the one unrecoverable mistake in this procedure." § "Before you start" carries quiescence with its measured reason (the source board rewritten three times during the first real migration). § "What the manifest will tell you about odd directories" says the tool will not rename and a person may. § "build.* in your config keeps pointing at the old path" is present. Step 7 is --check project
• Match: ✅

V3: .tfw/conventions.md · .tfw/glossary.md — AC-1, AC-2, AC-4, AC-14

• RF claim: payload paths; four new canon sections
• Actual: diffed. §2 gains three payload entries (scripts/gen_index.py, scripts/migrate_board.py, migrations/{major}.md). Four canon sections added and verified in place: some artifacts legitimately have no journal event (§4, after the immutability rule); a directory the grammar does not match is reported, never described (§4 Discovery); Where the tooling lives (§4); A major release ships a migration guide (§4). §5 gains the two-act UNDECLARED table. Glossary's UNDECLARED, Topic File and Portfolio index entries updated
• Match: ✅

V4: .tfw/templates/status.md — AC-5, AC-14 · DISCREPANCY

• RF claim: "quoted example + worked example" (§1); E20 "the template's example quotes its prose values"; E21 "a complete worked example … checked against the real validator". AC-14 fourth bullet: the canon must stop reading the UNDECLARED prohibition as absolute
• Actual: the quoting rule, the invalid/valid pair and the worked example are all present and correct, and test_the_status_template_examples_parse_and_validate drives both through gen_index.validate_status. But line 92 still reads: "Normalizing such a value to a declared one is prohibited." — the identical sentence the same commit rewrote in glossary.md, verified by diffing f14f744~1:.tfw/glossary.md against HEAD (old text: "Normalizing such a value to a declared one is prohibited"). The payload now states two different rules about the same act. Separately, line 78 reads "the four keys that are never prose — id, lifecycle, owner, authority, created, updated" — six keys are listed
• Match: ⚠️ partial — AC-5's own bullets hold; the AC-14 sweep left its third copy standing, in the file a person hand-authoring the carrier reads

V5: .tfw/workflows/update.md — AC-2, AC-6, AC-7, AC-10

• RF claim: Step 3 guide routing, Step 3a pristine-tag diff, the two decisions, the team/ step, local-source form, Step 8 self-check; 1165 words under the 1200 ceiling
• Actual: read in full. Step 3 routes to .tfw/.upstream/.tfw/migrations/{major}.md and states a major release without one is incomplete. Step 3a's loop is git -C {source} … with the comment "In the SOURCE tree, not yours" and the measured three-merges-to-zero reason. Step 3b carries task_containers as a two-option decision that "does not exist before 2.0.0, so there is nothing to preserve", initial_seq as a key to delete, and "Preserved does not mean correct. build.* is yours and is never overwritten." § "Create team/" says before the first durable write, this update's own commit included. Step 0 accepts a local working tree and requires git status --porcelain -- .tfw/ to print nothing. Step 8 is --check project. wc -w → 1165
• Match: ✅

V6: .tfw/workflows/plan.md — AC-12

• RF claim: Step 0 removed; naming is step 3 of Step 4, after the identifier exists
• Actual: diffed. ## Step 0: Name This Session deleted. Step 4 item 3 is a numbered step, names Coordinator | {ID} with | Phase {X} when given one, explains why it is not step 0, and says "Repeat this step whenever the slug changes." Step 3's "Ask clarifying questions / 🛑 WAIT" still precedes Step 4, so the question-first order the owner wanted kept is intact
• Match: ✅

V7: .tfw/workflows/init.md · .tfw/workflows/knowledge.md — AC-1, AC-7, AC-9

• RF claim: payload paths; AC-7 team/ step; --check project confirmation
• Actual: diffed. init.md step 3 creates team/ together with its first profile with the reason stated, renumbering 3–7 to 4–8; new step 8 is --check project. Steps writing status.md and the first journal event are step 7 — after step 3, so "before the first durable write" holds on conventions.md §4's own definition of that phrase. knowledge.md:71 now names templates/knowledge/topic.md
• Match: ✅

V8: .tfw/templates/team/profile.md — AC-7

• RF claim: cut to its own job, 50 → 33 lines; cites conventions.md §4 rather than restating the identity canon
• Actual: 33 lines. Carries what team/ is, the four-key bound table, and "Create this file BEFORE the first durable write." The identity-field canon, binding resolution and the shared-tree rule are cited, not copied: "They are not repeated here: a second copy of a rule is a second thing to keep true." No .tfw/templates/team_readme.md exists
• Match: ✅

V9: .tfw/templates/project_config.yaml · .tfw/project_config.yaml — AC-9

• RF claim: the five-line disambiguating comment deleted, not rewritten; build.verify is --check tasks; the template's verify is a real command
• Actual: diffed. The project config's build block goes 9 lines → 6; the comment beginning "NOT gen_index.py --check: that asks whether the shared index is current…" is gone, and verify is python .tfw/scripts/gen_index.py --check tasks. The template's verify placeholder is replaced by the same real command, with a new note that build.* is a preserved project section and therefore keeps a stale path silently
• Match: ✅

V10: .tfw/adapters/claude-code/CLAUDE.md.template · claude-code/README.md · antigravity/README.md — AC-11

• RF claim: TD-11 in all three, not one
• Actual: diffed. All four occurrences (template :31, claude-code README :43, antigravity README :22 and :64) now name .tfw/workflows/research/base.md. grep -rn "workflows/research\.md" over the whole tree outside tasks/ returns one hit — .tfw/CHANGELOG.md:102, the entry describing the fix
• Match: ✅

V11: docs/scripts/gen_docs.py · docs/scripts/test_integration.py — AC-1, AC-11, AC-5

• RF claim: cross-directory import bootstrap; +7 tests
• Actual: diffed. gen_docs.py replaces sys.path.insert(parent) and parent.parent.parent with a local _find_root walking upward for .tfw/, and points sys.path at {root}/.tfw/scripts absolutely because mkdocs runs the file with docs/ as its config root. test_integration.py gains six named tests: the two adapter-path checks, the self-test that proves the check fires, the copy-vs-source comparison, the status-template validator drive, and the ASCII class check — with an annotated two-entry NON_REPO_PATHS allowlist
• Match: ✅

V12: .tfw/VERSION · .tfw/CHANGELOG.md · README.md — AC-14

• RF claim: 2.0.0-dirty.2, the entry, the payload path
• Actual: cat .tfw/VERSION → 2.0.0-dirty.2; project_config.yaml:7 matches. git tag -l → latest is v2.0.0-dirty; no v2.0.0-dirty.2 tag, correct per the ONB Q3 (b) ruling. The CHANGELOG entry opens by justifying -dirty.2 over 2.0.1-dirty, carries § "Why this release exists" naming what the first external update found including that the framework could not deliver its own tooling, and closes with an explicit note that the older entry's code fence is a record and the reader should follow .tfw/migrations/2.0.0.md instead. README.md:257 names .tfw/scripts/gen_index.py
• Match: ✅

V13: team/README.md — R3 template move

• RF claim: "the template link the move would have broken"
• Actual: the one changed line points at .tfw/templates/team/profile.md; the file resolves
• Match: ✅

V14: the three moved templates — R3

• RF claim: git mv, references updated, no new files, flat namespace shrinks by 3
• Actual: git diff --name-status shows R100 for journal_event.md → journal/event.md and topic_file.md → knowledge/topic.md (byte-identical). team_profile.md → team/profile.md shows as A/D across the full range only because commit 80c2ed5 rewrote it 50 → 33 lines after the move; git log --follow on the new path returns 3 commits, so history followed. No team_profile.md, journal_event.md or topic_file.md remains
• Match: ✅

V15: adapter copies — 8 workflow copies + the 11 Codex skills

• RF claim: re-synced, verified byte-identical; "all 22 workflow copies and 11 Codex skills → all identical"
• Actual: cmp over every .claude/commands/tfw-*.md and .agent/workflows/tfw-*.md against its .tfw/workflows/ source → 11 + 11 identical, 0 diff. The only unmatched name is tfw-task, a meta-workflow with no canonical file, which predates this phase. cmp over all 11 .tfw/adapters/codex/skills/*/SKILL.md against .agents/skills/ → 11 identical
• Match: ✅

V16: the never-modify classes

• RF claim: 82 historical trace artifacts, 11 provenance comments, KNOWLEDGE.md's historical rows — bytes unchanged
• Actual: git diff --name-status f14f744~1 HEAD touches no file under tasks/ other than this phase's own artifacts and this task's two journal events. KNOWLEDGE.md is not in the change set at all. git grep -l "Written by docs/scripts/migrate_board.py" returns 14 today against the census's 13 — the difference is this phase's own RF and evidence, which quote the string
• Match: ✅

Commands Executed

#	Command	Result
1	python -m pytest .tfw/scripts/ docs/scripts/ -q --collect-only	254 tests collected, no collection errors — matches RF
2	python -m pytest .tfw/scripts/ -q	158 passed, 1 skipped in 11.97 s
3	python -m pytest docs/scripts/ -q	95 passed in 248.66 s
4	(1)+(2)+(3) → full suite	253 passed, 1 skipped — matches RF exactly, against the ONB's recorded 220-passed baseline. Net +33
5	python .tfw/scripts/gen_index.py --check tasks	exit 0 — 53 tasks validate against the closed schema, and the output names what it did not check
6	python .tfw/scripts/gen_index.py --check project	exit 0 — version 2.0.0-dirty.2, 1 participant, creates in workspace, resolves across [workspace, tasks]
7	python .tfw/scripts/gen_index.py --check index	exit 0 — index up to date
8	git grep -n "docs/scripts/gen_index\|docs/scripts/migrate_board" -- . \| grep -v "^tasks/"	11 hits, identical to ac1_gate.txt line for line — 8 CHANGELOG, 1 guide, 1 deliberate test fixture, 1 KNOWLEDGE.md:22
9	wc -w .tfw/workflows/update.md	1165 — under the §11 ceiling of 1200
10	cmp sweep over 22 workflow copies + 11 Codex skills	all identical
11	git log --oneline --follow on each moved file	gen_index 9, migrate_board 6, team/profile 3, journal/event 3, knowledge/topic 5 — history follows every move
12	independent AST-free diff of parse_board across the move	42 → 47 lines whole-function; exactly two code lines differ — the signature gaining heading: str = BOARD_HEADING, and the locator using it. Every row-reading line byte-identical
13	python -m mkdocs build --config-file docs/mkdocs.yml	started; the 95-test docs/scripts/ suite that passed in (3) runs mkdocs builds internally, which is the same gate

Claim & Source Checks

#	Claim / citation checked	Where it appears	Traces to	Holds?
C1	"253 passed, 1 skipped … Baseline before the phase: 220 passed, 1 skipped. Net +33"	RF §4	Re-run: 158+95 = 253 passed, 1 skipped, 254 collected. Baseline traced to ONB § "Baseline measurements taken before any edit" → pytest docs/scripts/ -q → 220 passed, 1 skipped	✅
C2	"the row parser is untouched, measured rather than asserted … exactly one line changed — the locator"	RF §3 AC-3, EV E13, ac3_parser_untouched.txt	Independently re-diffed (command 12). The row-reading body is byte-identical; the locator became a parameter. The artifact's declared method excludes the docstring — it also excludes the def line, so the signature change that carries the parameter is not shown. Substance holds; presentation understates by one line	✅ (with note)
C3	"parents[2] would have resolved wrong in two of three"	EV E2, fixture_report.md § F1	fixture_run.txt § "AC-1's real gate" prints, per placement, what parents[2] would give against what the tool resolved. tools/ → a directory outside the project; a/b/c/d/ → two levels short; tools/tfw/ → right by luck. Two of three, as claimed	✅
C4	"0 files hand-carried, 0 framework files edited inside .tfw/"	EV E47	fixture_run.txt prints framework files edited inside .tfw/: 1 — .tfw/scripts/__pycache__/gen_index.cpython-313.pyc. Reconciled in fixture_report.md ("Python bytecode written by running the tool, not an edit"), which E47 also cites, and the guide gained a .gitignore line for it. The EV row itself carries the bare 0 with no caveat	⚠️ reconciled, but not at the row
C5	"git log --follow returns 8 commits for gen_index.py, 5 for migrate_board.py", in a file headed "Pinned at 1079020"	RF §1	At 1079020 the values are 9 and 6; at 80c2ed5 they are 8 and 5. The numbers are true, taken one commit before the declared pin. fixture_run.txt's own header likewise records source: … at 80c2ed5 for the first pass	⚠️ measured off-pin
C6	"--validate and --doctor appear nowhere in the source, asserted by a test"	EV E33	test_the_three_checks_are_one_flag_with_a_subject asserts '"--validate"' and '"--doctor"' are absent from gen_index.py — i.e. absent as argparse strings. Prose mentions survive in a comment at gen_index.py:1090 explaining the retirement and in one test docstring. The test's guarantee is narrower than E33's sentence, and correctly so	✅
C7	ONB §7 knowledge citations, PV 0–4 in full, PV 5–7 by relevance	ONB §7, HL §7.2	See the table below — 29 of 29 resolve	✅

Discrepancies Found

#	Discrepancy	Severity
D1	.tfw/templates/status.md:92 still reads "Normalizing such a value to a declared one is prohibited." — verbatim the sentence commit f14f744 rewrote in glossary.md and replaced in conventions.md §5 with the two-act rule. The payload now states two different rules about the same act, and the absolute one is in the canonical carrier template a receiving project hand-authors from. AC-14's fourth bullet exists because "the prohibition reads as absolute and projects will either strand tasks or resolve them without a trace"; RF §3 marks AC-14 ✅, and E56 cites only conventions.md and glossary.md	Medium
D2	Same file, line 78: "the four keys that are never prose — id, lifecycle, owner, authority, created, updated" — six keys are listed. In a template AC-5 rewrote specifically so a person hand-authoring the carrier gets it right	Low
D3	RF §1 heads its table "Modified — 25", a figure census.md reaches by classifying the four scripts as moves and excluding them from the modified count. The RF's own table then lists those four scripts as Modified rows, along with templates/project_config.yaml and templates/team/profile.md, which the census does not count either. Counting distinct paths in the RF's table gives ~31. No budget limit is crossed under the census's declared method, and the census was raised to the coordinator before acting — but the number in the RF is not the number of rows in the RF	Low
D3a	The TS's basis for treating the scripts as moves — "The 2,864 lines of script relocate rather than get rewritten, so they cost a move and their path constants" — did not survive execution: gen_index.py changed 507 lines and migrate_board.py 246. A coordinator estimate the work invalidated; the executor raised a census group and proceeded, which is the return-to-coordinator rule working	Low — informational
D4	Evidence pinning. RF and EV both declare "Pinned at 1079020"; RF §1's git log --follow counts are the values at 80c2ed5 (C5), and fixture_run.txt's first pass is headed at 80c2ed5 too. TS §6: "Evidence is measured at a pinned commit and never against HEAD"	Low
D5	EV E47's bare "0 framework files edited inside .tfw/" against its own artifact's "framework files edited inside .tfw/: 1" (C4). Reconciled one file away, not at the row	Low
D6	RF §4 / E60 claim runtime ASCII is enforced "as a class". test_every_runtime_message_is_ascii is a line scanner — it toggles on print(/SystemExit( and resets on a line ending in ), so a message built into a variable and printed later escapes it. It caught five real occurrences and is worth having; the claim is stronger than the mechanism	Low

Escalation: D1 triggered 100% verification. All 33 RF-claimed files were opened or diffed, and the full 49-path change set was enumerated against census.md. Nothing further surfaced: the remaining files match their claims exactly.

Evidence Verification

#	RF Evidence ref	Artifact exists?	Matches claim?
E1–E5	AC-1 — fixture_run.txt, ac1_gate.txt, git show --stat f14f744	✅	✅ — ac1_gate.txt's 11 classified hits reproduce line for line at HEAD (command 8)
E6–E10	AC-2 — .tfw/migrations/2.0.0.md, fixture_run.txt steps 1–7	✅	✅ — guide read in full (V2); Step 3 route fires in the transcript
E11–E13	AC-3 — fixture_run.txt, ac3_parser_untouched.txt	✅	✅ — independently re-diffed (C2); both board attempts quoted, refusal names relocation first
E14–E18	AC-4 — fixture_manifest.md, fixture_index.md, fixture_run.txt	✅	✅ — Unresolved inputs — 2, no Backlog section, backlog idea, never started absent
E19–E22	AC-5 — ac5_validator.txt, .tfw/templates/status.md	✅	✅ for the AC's own bullets — before/after for all five keys is quoted verbatim. See D1/D2 for what the same file still carries
E23–E24	AC-6 — update.md, fixture_run.txt	✅	✅ — both decisions recorded as made, with reasons
E25–E27	AC-7 — init.md, update.md, .tfw/templates/team/profile.md, fixture_run.txt	✅	✅ — 33-line profile template verified; "wrote team/saubakirov.md from the payload template, no file hand-carried"
E28–E32	AC-8 — named tests, read_board(), fixture_run.txt	✅	✅ — read_board read directly (V1); board source: git show HEAD:tasks/README.md printed on every run
E33–E38	AC-9 — six named tests, project_config.yaml diff	✅	✅ — three subjects re-run here (commands 5–7); comment deletion confirmed 9 lines → 6
E39–E41	AC-10 — update.md Step 3a / Step 0, fixture_run.txt	✅	✅ — "CUSTOMIZED ×2, customized files: 2" of ~38
E42–E44	AC-11 — three named tests	✅	✅ — re-verified independently (V10, V15): 4 route fixes, 22+11 copies identical
E45–E46	AC-12 — .tfw/workflows/plan.md	✅	✅ — diffed (V6)
E47–E50	AC-13 half one — fixture_report.md, fixture_run.txt	✅	⚠️ — the run is real and complete; E47's 0 edited is contradicted by its own transcript's 1 and reconciled only in the sibling artifact (D5)
E51	AC-13 half two — DEFERRED	—	✅ correctly deferred. The blocker is named, the closing artifact is named, and TS §7 makes any other answer a rejection. This is the honest outcome
E52–E57	AC-14 — VERSION, project_config.yaml, CHANGELOG.md, conventions.md, glossary.md	✅	⚠️ — E52, E54, E55, E57 hold as verified. E56's canon claim holds for the two files it cites and leaves a third contradicting copy in the payload (D1)
E53	AC-14 tag — N/A	—	✅ — git tag -l confirms no v2.0.0-dirty.2. Correctly out of executor scope per ONB Q3 (b)
E58–E60	budget · DoF	✅	✅ for E58 (census re-derived) and E59 (phases/ → no matches). E60's "as a class" overstates a heuristic (D6)

Totals: 60 evidence items, 60 artifacts resolve, 55 match their claims cleanly, 4 carry the notes above (E19–E22 via D1/D2, E47 via D5, E56 via D1, E60 via D6), 1 correctly DEFERRED, 1 correctly N/A. No evidence artifact is missing, and no green signal was found sitting over a red one — the one place that risk was live, KNOWLEDGE.md:22, is named in AC-1's own gate text so the RF could not report around it.

Knowledge Citations Verified

Scanned PV priorities 0–4 in full and 5–7 by relevance. Master HL §7.2 carries 29 items; ONB §7 confirms all 29 and adds four of its own (N1–N4).

#	Artifact	Priority + exact citation	Link resolves?	Item exists?	Meaning matches?	Relevant to asserted application?
1	HL §7.2 #1 · ONB #1	PV 0 — README.md opening · § How It Works	✅	✅	✅ — bounded resumability without promising lossless context	✅ — AC-2's guide and AC-9's self-check are what a receiving project resumes from
2	HL §7.2 #2 · ONB #2	PV 0 — .tfw/README.md NS1, NS2 principles 3 and 5	✅ anchors id="ns1", id="ns2" present at :72, :79	✅	✅ — NS1 "another authorized person or agent can … continue without rebuilding the original conversation"; NS2 §5 "Continuation over isolated output"	✅ — a project told to run a file it does not have cannot continue. AC-1 is this clause made true off-site
3	HL §7.2 #3 · ONB #3	PV 1 — § Methodology values, Structural Enforcement	✅ anchor id="methodology-values" at :91	✅	✅ — "a rule that cannot reveal its own violation is only advice"	✅ — AC-11's path test and AC-1's depth fixture are the structural forms; parents[2] passing by coincidence is the prose-wearing-a-test's-clothes case
4	HL §7.2 #4 · ONB #4	PV 1 — § Methodology values + § Where truth belongs, one authoritative owner per truth type	✅ both anchors present	✅	✅ — "assigns one authoritative owner to each kind of truth rather than forcing every truth into one monolithic file"	✅ — the load-bearing constraint on AC-9. Verified operative: test_no_check_subject_writes_anything byte-compares the tree across all three subjects
5	HL §7.2 #5 · ONB #5	PV 1 — § Methodology values (Portability) + § Success Criteria	✅	✅	✅ — "no vendor may become the sole home of project memory"; criterion 4 "does not require the recipient to reconstruct missing work"	✅ — the phase's own clause. A payload that cannot carry its own tooling makes the recipient reconstruct it
6	HL §7.2 #6 · ONB #6	PV 2 — knowledge/philosophy.md F4	✅	✅ philosophy.md:11	✅ — structural enforcement beats state tables	✅
7	HL §7.2 #7 · ONB #7	PV 2 — F11	✅	✅ :18	✅ — avoid extra entities	✅ — and it is what R3's two withdrawals enforce
8	HL §7.2 #8 · ONB #8	PV 2 — F27	✅	✅ :34	✅	✅ — shapes --check's per-subject output
9	HL §7.2 #9 · ONB #9	PV 2 — F34	✅	✅ :41	✅	✅ — ONB Q1 is this clause: "an external project" is unusable until a named fixture is authorized
10	HL §7.2 #10 · ONB #10	PV 2 — F38	✅	✅ :45	✅ — coordinator attention is finite	✅ — eight questions batched into one file
11	HL §7.2 #11 · ONB #11	PV 3 — KNOWLEDGE.md D31, D50	✅	✅ :65, :84	✅	✅
12	HL §7.2 #12 · ONB #12	PV 3 — D37	✅	✅ :71	✅ — tfw-docs owns §1–§3, tfw-knowledge owns knowledge/*	✅ — N/A justified per row, and it is why ONB Q5 offered option (b) rather than editing KNOWLEDGE.md:22
13	HL §7.2 #13 · ONB #13	PV 3 — D43	✅	✅ :77	✅ — "Executor confirms or extends" is the cascade	✅ — ONB §7 is the executor link, filled per row
14	HL §7.2 #14 · ONB #14	PV 3 — D55, D59	✅	✅ :89, :93	✅ — D59: "declared attribution ≠ authentication", capability boundaries	✅ — commits carry [claude-code/TFW-60/phase-aa/executor] (verified in git log); D59 is exactly ONB Q3, and the tag was not cut
15	HL §7.2 #15 · ONB #15	PV 3 — D65	✅	✅ :99	✅ — reverting a result never reverts its trace	✅ — the 82 never-modify artifacts and 11 provenance comments, verified untouched (V16)
16	HL §7.2 #16 · ONB #16	PV 4 — conventions.md §§3–5	✅	✅	✅	✅ — §4's identifier rules are what collapse AC-4 to reported
17	HL §7.2 #17 · ONB #17	PV 4 — §13, §14	✅	✅	✅	✅ — git mv rather than copy-and-delete; verified as R0xx in --name-status
18	HL §7.2 #18 · ONB #18	PV 5 — knowledge/convention.md F22	✅	✅ convention.md:29	✅ — the board is a process artifact	✅ — N/A justified: Phase A retired the live board; AA only makes its location an input
19	HL §7.2 #19 · ONB #19	PV 6 — knowledge/process.md F7, F30	✅	✅ :14, :37	✅ — "Capture without an enforcement site does not change behaviour"	✅ — AC-2's routing from the step that needs it is the enforcement half
20	HL §7.2 #20 · ONB #20	PV 7 — knowledge/risk.md F1	✅	✅ risk.md:11	✅ — two sessions, one index; verbal warning 0/1	✅ — operative: the fixture is a separate clone with its own index
21	HL §7.2 #21 · ONB #21	PV 7 — knowledge/constraint.md F1, F3	✅	✅ :8, :10	✅	⚠️ — the ONB's application names team_readme.md and .tfw/scripts/README.md, both withdrawn at TS R3 after the ONB was written. The citation and meaning are right; the application is stale by design, and the R3 outcome honours F3 more strictly than the ONB's reading did
22–29	HL §7.2 #22–#29 · ONB #22–#29	RES 1 — YAML 1.2.2 · RFC 8259 · Git/rev-parse/add · Google Drive · OneDrive · Dropbox · five external systems · RES 2 — git trailers	✅ all URLs well-formed and named	✅	✅ — YAML 1.2.2's plain-scalar rule is AC-5's actual root cause and is correctly the basis of both the quoted example and the validator message	✅ — five correctly N/A with a per-row reason (JSONL not chosen; transport is TFW-61; Phase A's L3 untouched); #28 applied as counter-evidence, which is a legitimate use
N1–N4	ONB § "New items the coordinator did not cite"	conventions.md §10.4 · §11 · §10.3 · templates/journal_event.md	✅ ×3	✅ ×3	✅	⚠️ N4's path .tfw/templates/journal_event.md no longer exists — this phase moved it to templates/journal/event.md. The ONB predates the move; the item it cites (closed kind vocabulary, clock-read timestamp) exists at the new path and is what AC-14's fifth bullet writes into the canon

Totals: 29 HL §7.2 citations + 4 ONB additions = 33. Resolved: 33. Semantically verified: 33. Irrelevant: 0. Hallucinated: 0. Two application notes (#21, N4) are both artifacts of the ONB being written before TS R3 and before the template move — historical, not defects in the result.

Checkpoint

Self-check: - [x] Opened ≥ ⌈N × ratio⌉ files and recorded findings? — 100%, escalated on D1: all 33 claimed files plus the full 49-path change set - [x] Ran at least 1 build/test command? — 13 commands, including the full suite, all three --check subjects, the AC-1 gate grep, the cmp sweep and an independent parse_board diff - [x] Claim & Source Checks filled — C1–C7: seven claims spot-checked by re-derivation, every citation traced, the test counts and the parents[2] numbers checked against primary output rather than against the RF's summary - [x] Each RF §3 (AC) checkmark verified against actual file? — 15 rows, all traced - [x] KNOWLEDGE.md checked — contradictions with changes documented? — KNOWLEDGE.md is untouched by this phase, correctly; :22's stale path is a known, declared exception named inside AC-1's own gate and filed as RF §6 observation 1 for /tfw-docs - [x] Knowledge Citations from HL §7.2 and ONB §7 verified? - Total: 33, resolved: 33, semantically verified: 33, irrelevant: 0, hallucinated: 0 - [x] Evidence artifacts from RF §5 verified? - Total evidence items: 60, verified: 58 clean + 4 with notes (overlapping), missing: 0, deferred: 1, N/A: 1

Stage complete: YES