RF — TFW-53 / Phase D: Glossary, Adapters and Version¶

Date: 2026-08-14 · second pass 2026-08-18 Author: Executor (Claude Code) Status: 🟢 RF — Complete, corrected after review Parent HL: HL-TFW-53 — 🔒 FROZEN TS: TS Phase D — approved 2026-08-13, amended 2026-08-14 after ONB ONB: ONB Phase D — Q1 → (a), Q2 → (a), Q3 → (a); R1-R5 approved, R6 overruled REVIEW: REVIEW Phase D — 🔄 REVISE, four items Covers: frozen DoD 30–33 · corrective pass AC-8 to AC-12 Ships: TFW v1.2.0

Second pass, 2026-08-18. The reviewer was right on all four items and I reproduced each before touching anything: the total-file count, the -U0 hunk count, the substitution count, and the Human-Only Test failures. No framework file was re-edited except AC-11's two links — no glossary article changed, no adapter was re-copied, the release stands. AC-1 to AC-7 keep their functional outcomes, which the reviewer confirmed. This pass repairs the trace, not the work. Every correction is marked ⟳ so the first-pass claim and its replacement are both readable.

1. What Was Done¶

New Files — six, all trace, none product ⟳¶

⟳ Corrected 2026-08-18 (AC-8). The first pass listed five files here and omitted the RF itself. Under the owner's ruling of 2026-08-18 — the scope budget counts product files, not task artifacts — all six are trace and none spends the budget. conventions.md §6's own rationale is "agent maintains full context of changed files", and a saved grep capture carries no context to hold.

File	Description
`phase-d/RF__phase-d__glossary_adapters_version.md`	This file. Omitted from the first pass's own list
`phase-d/evidence/EV__phase-d__glossary_adapters_version.md`	Structured evidence — 19 rows, verdict recomputed in §5
`phase-d/evidence/drift-check-before.txt`	The `config.md` drift check run verbatim before the sync: 14 `DRIFT:` lines
`phase-d/evidence/direction-check.txt`	Per-pair source-only / copy-only counts, run before any copy
`phase-d/evidence/drift-check-after.txt`	The same check after the sync: no output
`phase-d/evidence/ac-gates.txt`	Every AC gate command with its output

Modified Files — the six framework files (AC-7 requires a justification per file)¶

git diff --stat, framework files only:

 .tfw/compilable_contract.md    |  4 ++--
 .tfw/conventions.md            |  4 ++--
 .tfw/glossary.md               | 36 ++++++++++++++++++++++++++++++++++--
 .tfw/templates/HL.md           |  8 ++++----
 .tfw/templates/review/judge.md |  6 +++---
 .tfw/workflows/review.md       |  4 ++--
 6 files changed, 47 insertions(+), 15 deletions(-)

File	±	Justification — every hunk is a substitution, a named debt fix, or an addition
`.tfw/glossary.md`	+34 / 2 changed	Addition: ten articles — nine under a new `## Contract and Purpose Defence` grouping, `Result Visualization` between `Value Flow` and `Findings Map`. Two substitutions: PV Index priority 0 `frozen baseline` → `contract baseline`; Knowledge Gate `Phase 0` → `Step 2` (TD-163)
`.tfw/templates/HL.md`	4 changed	One debt fix: line 10, the inline recovery command → a pointer to `conventions.md` §3 rule 15 (TD-164). Three substitutions: the north-star field label and its explainer (:18, :20) gain `Project`; :22 `frozen baseline` → `contract baseline`
`.tfw/templates/review/judge.md`	3 changed	Three substitutions, all in the Purpose Check block: the hybrid "committed frozen baseline" → `contract baseline` (:34), the fallback chain (:35), the Checkpoint row (:84)
`.tfw/compilable_contract.md`	2 changed	Two debt fixes: `PP{N}` and `NS{N}` added to the Resolution rules (TD-165); `KNOWLEDGE.md` §0 removed from "Where references appear" (TD-167)
`.tfw/conventions.md`	2 changed	Two substitutions: :94 "frozen contract baseline" → "contract baseline"; :108 the north-star fallback rule
`.tfw/workflows/review.md`	2 changed	Two substitutions: :28 the context-loading line, :87 the Purpose Check paragraph. Word-neutral — 1,176 before, 1,176 after
`.tfw/workflows/plan.md`	0	Untouched. The TS listed it as MODIFY, but it carries no occurrence of either retired form. Deviation recorded in §3

Modified Files — adapters, entry points, release, registry¶

File	Changes
`.claude/commands/tfw-{plan,review,research,init,handoff,update,knowledge}.md` × 7	Byte copies from `.tfw/workflows/`. Diffstat lines: plan 82 · init 41 · handoff 14 · review 12 · update 12 · research 6 · knowledge 2
`.agent/workflows/tfw-{plan,review,research,init,handoff,update,knowledge}.md` × 7	Identical set, identical figures
`.tfw/adapters/codex/skills/tfw-plan/SKILL.md` · `.agents/skills/tfw-plan/SKILL.md`	One word: `description` said "creates or revises approved HL/TS artifacts" — a permission `conventions.md` §3 rule 3 now forbids. Both copies stay byte-identical
`CLAUDE.md`	Two purpose cells rewritten (`/tfw-plan`, `/tfw-review`); two rows added (`/tfw-knowledge`, `/tfw-config`) per the coordinator's recorded R6 extension. No new section
`AGENTS.md`	Unchanged — checked row by row, nothing false. AC-4 directs change only on falsity
`.tfw/VERSION` · `.tfw/project_config.yaml`	`1.1.0` → `1.2.0`, in lockstep
`.tfw/CHANGELOG.md`	One `## [1.2.0]` entry covering Phases A–D, five blocks, with the TFW-54 pointer. Both `[Unreleased]` bullets folded into `### Changed`, the drift-check sentence corrected from 12 to 14 and from "not yet repaired" to repaired
`TECH_DEBT.md`	Five closures with reasons (TD-157, TD-163, TD-164, TD-165, TD-167); two re-routes out (TD-158, TD-131); one new debt (TD-170)
`README.md`	Board row only — left unstaged, see §2 D8

File accounting under the stated rule ⟳ (AC-8)¶

Counting rule, quoted: "the budget counts product files, not task artifacts. The RF, the EV file and its attachments are the trace of a change, not the change." — owner ruling 2026-08-18, TS AC-8.

Class	Count	Limit	Detail
Modified product files	28	30 · `max_modified_files` 30	27 in commit `ce30f3b` + `README.md`, modified in the working tree and deliberately unstaged
├─ of which byte `cp`	14	—	zero authored content: 7 workflows × 2 adapter folders
└─ of which authored	14	—	6 framework files · 2 Codex router copies · `CHANGELOG` · `VERSION` · `project_config` · `CLAUDE.md` · `TECH_DEBT.md` · `README.md`
New product files	0	`max_new_files` 15	none
Trace files (not budgeted)	6	—	this RF · the EV file · four command-output attachments

Total product files touched: 28 of 30. Under the limit, with the rule named beside the number.

Two figures in circulation were wrong, and I am not adopting either:

The first pass reported "28 modified, 5 new, against 30 / 15" and compared only the sub-limits. It never computed a total, which is what let a 33-file phase read as green. The reviewer was right to call that masking.
TS AC-8 says "27 modified product files"; the correct figure is 28. 27 is the commit's modified count, which omits README.md — a product file this phase changed twice (board row to 🟠 ONB, then to 🟢 RF) and will change a third time now. Being unstaged does not make it untouched. The conclusion is unaffected — 28 < 30 — and the ambiguity is filed as TD-173.

2. Key Decisions¶

uncommitted baseline left untouched (ONB Q1 → (a)). The phrase names the absence of a baseline, not a synonym for it. Substituting contract into it produces a sentence meaning its opposite, and repairing that is a rewrite — DoF-1. The gate is read as grep -roE "[^n]committed baseline", which returns 0 both before and after. Two occurrences survive by design: conventions.md:75 and plan.md:60.
The north-star field label was renamed, not exempted (ONB Q2 → (a)). templates/HL.md:18 and :20 were the only two genuine bare-capital occurrences in the framework. A form field reading North Star while the glossary defines Project North Star teaches the wrong name at the moment the reader learns it. Cost: 2 lines, +2 words, in a template off F2's budget.
The nine contract terms got one grouping, not two (ONB R1). ## Contract and Purpose Defence after ## Artifact Types. The five contract terms and the four review-side terms are one mechanism read from two ends; 35 → 36 groupings is the minimal structural change to a file with 78 existing entries.
The AC-1 gate is case-insensitive, and this was not cosmetic. A case-sensitive grep -c "deferral confession" returns 0 across the whole framework — the term ships as **Deferral confession** at judge.md:50. The original gate would have failed a term that exists and driven a false conclusion that Phase C never shipped it. The glossary heading matches the shipped casing rather than the DoD's, so one name carries one form.
All fourteen adapter copies were re-synced, after a direction check, not on faith. Six were this task's doing (tfw-plan ×2, tfw-review ×2, tfw-research ×2); eight belong to earlier tasks (tfw-init ×2, tfw-handoff ×2, tfw-update ×2, tfw-knowledge ×2). Before copying, every pair was diffed and every copy-only line read: each is an older form of a source-only line in the same hunk. No copy held content its source lacks, so the stop condition did not fire and the copy propagated already-reviewed content only.
One word changed in the Codex tfw-plan router; tfw-review was left alone. AC-3 permits editing only statements that are now false. tfw-review's contract survives the contract mechanism intact. tfw-plan's description granted revision of an approved HL — precisely what Phase A removed. Removing approved is the minimal true edit; adding a freeze/amendment statement would have been an addition, not a correction, and is reported in §6 instead.
The [Unreleased] block was edited, not folded verbatim (ONB R4). Its closing sentence read "First run of the check found 12 drifted copies (6 workflows × 2 folders) — recorded, not yet repaired." Carried as written into 1.2.0, the release note would have been false about the release it announces, on the day it ships, in the exact number that release corrects. grep -c "12 drifted copies" now returns 0.
README.md was edited but not staged (ONB R3). The working tree carries a concurrent uncommitted TFW-55 row change. Staging README.md by path would sweep another session's work into a TFW-53 commit. The board row is written; the coordinator lands the file. TS §9 provides this fallback explicitly.
The registry re-scan the coordinator ordered found a second omission. Beyond TD-158, TD-131 points at Phase D and appears in neither TS list. Treated the same way — re-routed out with the reasoning recorded — because its own text says fixing it "changes heading depth in every HL", a structural change to a shipped template inside Phase A's section, not a naming fix.
The resolver gap was filed once, for four patterns, not four times (coordinator ruling on ONB §5.4). TD-165's stated defect is closed. TD-170 records the wider finding: gen_docs.py resolves only TD-{N} and D{N}, so P{N} and F{N} were already aspirational in the same line before NS/PP joined them.
The Codex tfw-plan router carried a false permission, and byte-parity could never have caught it. ⟳ Re-homed here from §8 S3 by AC-10 — it describes a change that was made, not an insight. The description field read "creates or revises approved HL/TS artifacts": written before the contract existed, true then, and granting exactly what conventions.md §3 rule 3 now forbids. The routers are deliberately not copies, so a diff-based parity check passes over them by design. Only a human reading a router against the mechanism catches it, which is why AC-3 states that as a separate bullet from the cp. One word removed, in both the source and the installed copy.
Review item 1 was refused, and the refusal is the coordinator's, recorded. ⟳ The reviewer asked to "consolidate or remove at least three supporting evidence files" to bring 33 under 30. Deleting proof to satisfy a counter is not available in this framework — the four attachments are the raw output of the gates AC-3 and AC-5 turn on. The coordinator refused it as written and ruled the counting subject instead (§1, AC-8). I record the refusal here rather than quietly complying, because a reviewer reading the next RF must be able to see that an instruction was declined, and by whom.
The first pass's 12 hunks came from a command run over a different file set than the claim describes. ⟳ The ledger names six files — the five substitution and debt-fix targets plus glossary.md. The command I actually ran listed plan.md (0 hunks) and omitted glossary.md (4 hunks): 16 − 4 + 0 = 12. The number and the sentence were written in the same session and neither checked the other. Corrected in §3, §4 and §9; the classification underneath is unchanged, and the reviewer confirmed it independently.
The evidence verdict was recomputed, not restored (AC-12). Two rows moved because their underlying artifacts changed; seventeen were left alone. §5 carries the row-level accounting.

3. Acceptance Criteria¶

AC	Gate	Result
AC-1 — vocabulary defined	`grep -ci` each term ≥ 1; `wc -w` per article	✅ 10/10 terms ≥ 1 (lowest is 1, highest 5). All ten articles 41–50 words against the p75 = 50 ceiling. Each ends with a `→` pointer; every pointer target opened and confirmed
AC-2 — one name per concept	`grep -rn` each retired form → 0; `wc -w` both workflows	✅ `frozen baseline` 8 → 0. `committed baseline` excluding the negation 0 → 0. Bare `North Star` 2 → 0 (one line-wrap artefact remains and is recorded). `plan.md` 1,195 → 1,195, `review.md` 1,176 → 1,176
AC-3 — adapter parity	drift check → no output; `diff` per copy → empty	✅ 14 `DRIFT:` lines → silent, exit 0. All 22 pairs diff to 0 lines. Direction verified per pair before copying. The six/eight split is stated in §2 D5. Codex routers checked; one false statement found and fixed
AC-4 — entry points	read both tables against the shipped workflows	✅ Both `CLAUDE.md` cells rewritten and read back against `plan.md` Steps 4/6c/6d and `review.md` Step 3. Two rows added under the recorded extension. `AGENTS.md` unchanged — nothing false. No new section in either file
AC-5 — release shippable	`cat VERSION`; `grep -c` config; `grep -c [TFW-54](../../TFW-54__agent_team_mode/)`	✅ `1.2.0` · `1` · `1`. One `[1.2.0]` entry, five blocks, Phases A–D. `[Unreleased]` folded and corrected
AC-6 — five debts, TD-164 first	open each line; `pytest docs/scripts/`	✅ TD-164 closed first, by replacement not rewording. TD-163 (glossary half), TD-165, TD-167 closed; TD-157 closed by AC-3. Each records the reason. 68 passed, unchanged from the pre-phase baseline
AC-7 — no unasked content change	`git diff` reviewed hunk by hunk	✅ ⟳ Ledger corrected 2026-08-18, see AC-9. `git diff -U0 ce30f3b^ ce30f3b` over the six ledger files returns 16 hunk headers (14 at default context), classifying as 10 substitution hunks carrying 11 substituted lines, 4 named debt fixes and 2 addition sites. The first pass said "twelve hunks … eight substitutions"; both figures were wrong. The classification is unaffected, and the reviewer confirmed it: no hunk is a rewritten sentence, no section was added, no template gained a field
AC-8 — budget counted under a stated rule	the recount, with the rule quoted beside it	✅ 28 modified product files of 30, 0 new product, 6 trace files listed separately. The rule is quoted in §1 from the owner's 2026-08-18 ruling. I do not adopt the TS's own figure of 27 — it omits `README.md`, a product file this phase changed. Ambiguity filed as TD-173
AC-9 — the ledger reproduces	run the named command, compare	✅ `git diff -U0 ce30f3b^ ce30f3b -- <six files>` → 16 hunk headers: `conventions.md` 2 · `glossary.md` 4 · `templates/HL.md` 4 · `judge.md` 2 · `review.md` 2 · `compilable_contract.md` 2. Full reconciliation in §9. AC-7 asked for `--stat`; both are now reported, each labelled with the command that produced it
AC-10 — §7/§8 carry knowledge, not computation	read §7 and §8; confirm each re-homed item is at its destination	✅ §8 is `No strategic insights.` with its reason. §7 keeps FC3 only. S3 → §2 D11, S4 → §6 obs. 8, S2's operative half → §6 obs. 9, FC5 → §6 obs. 10. S1, FC1, FC2 and FC4 dropped as duplication of rulings the ONB and TS already carry. Template contradiction filed as TD-174
AC-11 — broken entry-point links	`grep -rn "PROJECT_CONFIG" CLAUDE.md AGENTS.md` → no matches	✅ Both `CLAUDE.md` links now read `.tfw/project_config.yaml`, and the target exists at that path. Gate returns nothing. The repo-wide sweep the AC required was run — every other occurrence is history that must not be rewritten, or the rule itself; enumerated in §6 obs. 11. TD-172 closed
AC-12 — verdict recomputed	read the verdict against its rows	✅ 17/19 → 19/19, stated as a recomputation: E17 and E19 moved because their artifacts changed, and five rows were added for this pass. No row upgraded without an underlying change. Row-level accounting in §5

Definition of Failure — all nine clear:

DoF item	Status
A sentence rewritten under the consistency banner	✅ clear — hunk-by-hunk review, §1 table
`plan.md` or `review.md` crosses 1,200 words	✅ clear — 1,195 and 1,176, neither moved
Drift check still prints, or the sync reported without running it	✅ clear — run before and after, both captured verbatim
An adapter copy found ahead of its source and merged	✅ clear — direction check on all 14 pairs before copying
A glossary article restates a rule instead of defining a term	✅ clear — max 50 words, each pointing at the rule rather than reproducing it
TD-164 closed by editing the command rather than replacing it	✅ clear — the command is gone, replaced by the pointer
A re-routed debt fixed here anyway	✅ clear — TD-120/140/142/153/154/155/158/131 all untouched
`VERSION` and `tfw.version` disagree	✅ clear — both `1.2.0`
Ships without the TFW-54 pointer	✅ clear — named in the `[1.2.0]` preamble

Deviations from the TS §4 file list, all in the restrictive direction:

.tfw/workflows/plan.md — listed MODIFY, not touched. It carries no occurrence of either retired form. Touching it would have been an unasked change.
.tfw/adapters/codex/skills/tfw-review/SKILL.md and .agents/skills/tfw-review/SKILL.md — listed MODIFY, not touched. AC-3 conditions the edit on a statement being false; none is.
README.md — modified, deliberately not staged (§2 D8).

4. Verification¶

Check	Command	Result
Tests	`python -m pytest docs/scripts/`	68 passed in 55.82s. Pre-phase baseline captured at ONB time: 68 passed in 55.99s — same count, no new failures
Adapter parity	`config.md` § Drift check, verbatim	No output, exit 0 (was 14 lines)
Byte parity	`diff` per adapter pair, 22 pairs	0 diff-lines on every pair
Codex router parity	`diff` source vs installed, `tfw-plan` and `tfw-review`	Byte-identical after the edit
Word budgets	`wc -w`	`plan.md` 1,195 · `review.md` 1,176 — both unchanged, both under 1,200
Scope budget ⟳	`git diff --name-status ce30f3b^ ce30f3b` + working tree	28 modified product files of 30; 0 new product; 6 trace files, unbudgeted. Rule quoted in §1. The first pass reported "28 modified, 5 new, against 30 / 15" and computed no total — the defect the reviewer caught
AC-7 ledger ⟳	`git diff -U0 ce30f3b^ ce30f3b -- <six ledger files>`	16 hunk headers; 14 at default context. 10 substitution hunks / 11 substituted lines · 4 debt fixes · 2 additions. The first pass reported 12 and 8
Entry-point links	`grep -rn "PROJECT_CONFIG" CLAUDE.md AGENTS.md`	No matches (was 2 in `CLAUDE.md`). Target `.tfw/project_config.yaml` exists

No lint or build command is configured for markdown in this project; docs/scripts/ is the only executable gate and it is the one AC-6 names.

5. Evidence¶

Cognitive mode: Observational verification — evidence lives in the EV file, not inline.

See EV file for evidence details.

Evidence verdict: 26/26 VERIFIED, 0 DEFERRED, 0 BLOCKED, 0 N/A — ⟳ recomputed 2026-08-18, not restored (AC-12)

Rows	First pass	Now	Why they moved
E17 (AC-7 ledger)	VERIFIED on a count that does not reproduce	VERIFIED on the reproduced count	The artifact changed. The row now names the exact command, its 16-header output and the per-file breakdown. Not the old claim upgraded — the old claim replaced
E19 (scope budget)	VERIFIED against sub-limits only	VERIFIED against the total, under a quoted rule	The artifact changed. The row now carries the counting rule, the 28-of-30 product total and the trace list
E17b, E19b, E20–E24	—	VERIFIED (seven new)	Added for the corrective pass: E17b and E19b record why the two replaced rows were wrong; E20–E24 cover AC-8 through AC-12. Row count 19 → 26
E1–E16, E18	VERIFIED	VERIFIED, untouched	The reviewer independently established all seventeen. Nothing was re-run to inflate a count

The reviewer's figure of 17 of 19 was correct at REVISE time. It is superseded by artifacts that changed, not by argument.

6. Observations (out-of-scope, not modified)¶

#	File	Line(s)	Type	Description
1	`.tfw/workflows/plan.md`	Step 2	duplication	TD-163's second half, left open on the coordinator's R5 ruling. Step 2 restates the three Knowledge Gate modes (`hard` stop + justification / `soft` reminder / `off` skip) that `glossary.md` also states — ~50 words. Removing it is content deletion in Phase B's file, which HL §7.1 forbids this phase. It is the single cheapest cut available to a file at 1,195 of 1,200, and TFW-57 needs exactly this kind of concrete candidate
2	`docs/scripts/gen_docs.py`	`resolve_references()`, `add_table_anchors()`	missing-test	Four of the six anchor-link patterns the contract declares are not implemented. The resolver links `TD-{N}` and `D{N}` only; `P{N}` and `F{N}` get table anchors but no link resolution; `NS{N}` and `PP{N}` get neither. An `NS3` or `P7` reference renders as plain text and the build emits no warning. Filed as TD-170, one debt for all four, per the coordinator's ruling on ONB §5.4. Extending the script was explicitly out of scope
3	`TECH_DEBT.md`	TD-131	todo	Second row found by the ordered re-scan. It points at "Phase D terminology/consistency pass" and appears in neither TS list — the same omission as TD-158. Re-routed out and the reasoning recorded in the row: its own text says fixing it "changes heading depth in every HL", so it is a structural change to a shipped template inside Phase A's section, not a naming fix
4	`TECH_DEBT.md`	TD-133	todo	An orphaned row. It routes to "Phase C, owns the `NS{n}` / `PP{n}` namespace work" — a phase that closed on 2026-08-13. Its stated defect (`P{N}` resolving to the removed `KNOWLEDGE.md` §0) no longer reproduces: the §2 pattern table now reads "HL §7 Principles row (task-local)". The residual — that `P{N}` has no resolution target — is subsumed by TD-170. Left for reviewer triage rather than closed here, since it is in neither TS list
5	`CLAUDE.md`	51, 53	naming	⟳ Reported in the first pass, fixed in the second. Two references to `.tfw/PROJECT_CONFIG.yaml`, a filename D48 retired in April 2026. Broken on any case-sensitive filesystem; Windows masks it. The review routed it to TD-172; the owner ruled it fixed here on 2026-08-18, widening AC-4's limit to "two rows and two link corrections, nothing else". Both links now read `.tfw/project_config.yaml`. TD-172 closed
6	`.tfw/adapters/codex/skills/tfw-{plan,review}/SKILL.md`	Contract bullets	todo	Neither router mentions the freeze, the amendment channel or the Purpose Check. No statement is false — both point at the canonical workflow, which carries the rules — so AC-3's "edit only what is false" correctly leaves them alone. Recording it because the two routers are the Codex surface's only always-loaded contract, and a reader who never opens `plan.md` learns nothing about the mechanism this release exists to ship
7	`.tfw/glossary.md`	268	style	The PV Index closing note wraps "Project North Star" across a line break, so `> North Star.` stands alone on line 268 and any naive `grep "North Star"` reports it as drift forever. Not fixed: rewrapping the line is a formatting change to Phase C's section and buys nothing but a cleaner grep

| 8 | .claude/commands/tfw-handoff.md | Step 11, §5 guidance | environment | ⟳ Re-homed from §8 S4 by AC-10. This session was invoked from the stale copy it was sent to repair: /tfw-handoff loaded the adapter file, which still instructed an inline RF §5 evidence table and permitted skipping evidence collection when no AC carries an Evidence: field. The canonical .tfw/workflows/handoff.md requires a structured EV file and allows no skip. I followed the canonical file. It is TD-157's failure mode observed live on a second workflow and by a second role — and it is why adapter drift is not a tidiness problem: it silently changes what an agent is instructed to produce. Closed by this phase's sync | | 9 | .tfw/workflows/config.md | § Drift check | naming | ⟳ Re-homed from §8 S2 by AC-10, operative half only. A terminology gate written as a raw substring grep cannot distinguish a concept from its denial: the sweep for committed baseline matched inside uncommitted baseline, which names the absence of the thing. A mechanical pass would have substituted it into nonsense. The fix is not a smarter regex but a ruled exclusion recorded in the spec, which is what AC-2's amended table now carries. Worth keeping because it is why the AC-2 gate had to be re-read at ONB before it could be run | | 10 | .tfw/workflows/config.md | § Drift check | environment | ⟳ Re-homed from §7 FC5 by AC-10. The drift check is a bash snippet and must be run in a POSIX shell. This project's primary shell on Windows is PowerShell, where the snippet is a parse error; it was run in Git Bash. Any executor on a Windows workstation hits this before the check produces its first line | | 11 | repository-wide | — | naming | AC-11's sweep for the retired PROJECT_CONFIG.yaml filename, outside CLAUDE.md: nothing live. Every remaining occurrence is either history that must not be rewritten — .tfw/CHANGELOG.md recording the rename itself, KNOWLEDGE.md D16/D20/D22/D48 and its Legacy row, knowledge/philosophy.md F23, two README.md task titles — or the rule stating the retirement (conventions.md:468, "not PROJECT_CONFIG.yaml"). 136 files under tasks/ carry it as sealed trace. .tfw/workflows/init.md is clean, although TD-102's row still quotes it as if it were not. Reported, not fixed, exactly as AC-11 directs |

On AC-2's third-synonym-pair clause: none found. Checked while sweeping — goal check 0, goal defence 0, purpose check 12 (one form), amendment log 6 (one form), freeze commit 2 (one form). The three hyphenated north-star uses (conventions.md:106, :113, REVIEW.md:54) are adjectival — "a north-star clause" — which is ordinary English, not a second defined term.

7. Fact Candidates¶

fact-candidates: processed 2026-08-18 (/tfw-knowledge, TFW-53 A–E)

⟳ Corrected 2026-08-18 (AC-10). The first pass carried five. Four failed the template's Human-Only Test — "would this fact be unknown without the human saying it?" — and the reviewer was right to fail them. FC1, FC2 and FC4 are dropped: each restates a coordinator ruling that the ONB and the TS already carry, and re-recording a decision away from where it was made is duplication, not knowledge. FC5 is not dropped but re-homed to §6 observation 10 — it is an environment fact an agent discovers by running the command, which is a §6 item, not a fact candidate.

#	Category	Candidate	Source	Confidence
1	stakeholder	The owner ruled on 2026-08-13 that all fourteen drifted adapter copies be re-synced, not only the six this task caused — because a binary check that keeps printing eight failures after the phase whose deliverable is adapter parity stops being read. The reasoning generalises past adapters: it is a rule about when a passing-with-known-failures signal loses its value	Owner decision 2026-08-13, recorded in TS AC-3; reasoning extends D53	High

(Numbered FC3 in the first pass; retained on the reviewer's own recommendation. It is the one entry no agent could have derived from the repository — the owner's reasoning existed only in the instruction.)

8. Strategic Insights (Execution)¶

No strategic insights.

⟳ Corrected 2026-08-18 (AC-10). The first pass carried four entries; all four were agent analysis, and the template is unambiguous here — "Only when the human provides domain knowledge… If no human interaction occurred — write No strategic insights." No human domain knowledge entered during execution: the owner's input reached this phase at TS approval, before it began, and every answer received mid-flight came from the coordinator, which is an agent.

Nothing was deleted — three of the four were re-homed and the RF says where: S3 → §2 D11 (it describes a change that was made), S4 → §6 observation 8, S2's operative half → §6 observation 9. S1 is dropped: it was commentary on this TS's measurement quality, already recorded in the ONB and in the coordinator's own commit.

The section's admission rule and §7's are not equally clear, and the difference matters: §8's test is unambiguous and I misapplied it, while §7's Scope line and its Human-Only Test contradict each other outright. That contradiction is filed as TD-174, with the measurement that shows it has been live for four consecutive approved reviews.

9. Diagrams¶

What Phase D changed, by kind of change.

                            .tfw/  (canonical)
                              │
  ┌───────────────────────────┼───────────────────────────────┐
  │ ADD                       │ SUBSTITUTE                    │ FIX (named debt)
  │                           │                               │
  glossary.md                 conventions.md      :94 :108    templates/HL.md      :10  [TD-164](../../../reference/tech-debt.md)
   ├─ ## Contract and         glossary.md         PV Index     compilable_contract  :81  [TD-165](../../../reference/tech-debt.md)
   │  Purpose Defence  [new]  templates/HL.md     :18 :20 :22  compilable_contract  :65  [TD-167](../../../reference/tech-debt.md)
   │   ├─ HL Contract         review/judge.md     :34 :35 :84  glossary.md    Knowledge
   │   ├─ Contract Baseline   workflows/review.md :28 :87        Gate step pointer   [TD-163](../../../reference/tech-debt.md)
   │   ├─ Frozen Section
   │   ├─ Amendment          frozen baseline    8 → 0
   │   ├─ Amendment Log      committed frozen   1 → 0
   │   ├─ Project North Star bare North Star    2 → 0
   │   ├─ Purpose Check
   │   ├─ not fit for purpose         plan.md   1195 → 1195  ── no hunk at all
   │   └─ Deferral confession         review.md 1176 → 1176  ── word-neutral
   └─ ## Knowledge Terms
       └─ Result Visualization  [new, between Value Flow and Findings Map]

The sync, and the check that made it safe.

  step 1  DIRECTION CHECK — before any copy, all 14 pairs
          ┌──────────────────────────────────────────────────┐
          │  for each pair: source-only lines vs copy-only    │
          │  read every copy-only line                        │
          │                                                   │
          │  result: every copy-only line is an OLDER form    │
          │          of a source-only line in the same hunk   │
          │  → no copy holds content its source lacks         │
          │  → AC-3 stop condition does NOT fire              │
          └──────────────────────────────────────────────────┘
                                │
  step 2  DRIFT CHECK — before                14 DRIFT lines, exit 0
                                │             ├─ ours (6): plan ×2, review ×2, research ×2
                                │             └─ earlier tasks (8): init ×2, handoff ×2,
                                │                                   update ×2, knowledge ×2
                                ▼
  step 3  cp  .tfw/workflows/{7}.md ──┬──► .claude/commands/tfw-{7}.md
                                      └──► .agent/workflows/tfw-{7}.md
                                │
  step 4  DRIFT CHECK — after         (silent), exit 0
                                │     diff per pair, 22 pairs → 0 lines each
                                ▼
          Codex routers: NOT copies — read, not cp
            tfw-review  → no false statement          → unchanged
            tfw-plan    → "revises APPROVED HL/TS"     → one word removed
                          (granted what §3 rule 3 forbids)

Release surface after v1.2.0.

  VERSION            1.1.0 ──► 1.2.0   ┐
  project_config     1.1.0 ──► 1.2.0   ┘ lockstep, unbroken since 0.8.5

  CHANGELOG  ## [Unreleased]  ── "Nothing pending."
             ## [1.2.0] — 2026-08-14   one entry, Phases A–D
                ├─ why it exists ──► [TFW-54](../../TFW-54__agent_team_mode/)  (the pointer DoD-33 requires;
                │                             was 0 occurrences in .tfw/)
                ├─ Added    HL Contract · §12 Amendment Log · Purpose Check
                │           · PV priority 0 · ten glossary articles · §14 anti-patterns
                ├─ Changed  plan.md 6c inverted · research/base.md · Reviewer Identity
                │           · review.md:28 · REJECT branch (a) · Phase HL derivation-only
                │           · §3.1 mandatory · one name per concept
                │           · scope budgets  ⟵ folded from [Unreleased]
                │           · adapter sync    ⟵ folded, "12 … not yet repaired" → 14, repaired
                ├─ Removed  Judge mapping-integrity check · RES.md:32
                │           · inline recovery command · KNOWLEDGE.md §0 pointer
                └─ Fixed    [TD-165](../../../reference/tech-debt.md) · [TD-163](../../../reference/tech-debt.md) · [TD-157](../../../reference/tech-debt.md) (+ six further workflows)

⟳ The AC-7 ledger, reconciled (AC-9). The first pass reported twelve hunks, eight substitutions. Both numbers were wrong. Command, output and classification, so a reader can reproduce it:

  $ git diff -U0 ce30f3b^ ce30f3b -- <the six ledger files> | grep -c '^@@'
    16                          ← default context (-U3) merges adjacent hunks and gives 14

  file                        hunks   what each hunk is
  ─────────────────────────── ─────   ────────────────────────────────────────────────
  conventions.md                  2   @@ -94   substitution
                                      @@ -108  substitution
  glossary.md                     4   @@ -41,0 +42,29   ADDITION  ## Contract and Purpose Defence
                                      @@ -52,0 +82,3    ADDITION  ### Result Visualization
                                      @@ -213/+245      DEBT FIX  [TD-163](../../../reference/tech-debt.md)
                                      @@ -225/+257      substitution
  templates/HL.md                 4   @@ -10   DEBT FIX  [TD-164](../../../reference/tech-debt.md)
                                      @@ -18   substitution   ┐ field label
                                      @@ -20   substitution   ┘ + explainer
                                      @@ -22   substitution
  templates/review/judge.md       2   @@ -34,2 +34,2      substitution ×2 LINES in ONE hunk
                                      @@ -84             substitution
  workflows/review.md             2   @@ -28   substitution
                                      @@ -87   substitution
  compilable_contract.md          2   @@ -67   DEBT FIX  [TD-167](../../../reference/tech-debt.md)
                                      @@ -81   DEBT FIX  [TD-165](../../../reference/tech-debt.md)
  ─────────────────────────── ─────
  TOTAL                          16

  classification            hunks   lines
  ───────────────────────── ─────   ─────
  canonical substitution       10      11   ← judge.md @@ -34,2 carries two
  named debt fix                4       4
  addition site                 2       —
  ───────────────────────── ─────   ─────
                               16

  git diff --stat  (what AC-7 literally asked for)
    6 files changed, 47 insertions(+), 15 deletions(-)

Where 12 came from. The ledger table names six files — the five substitution and debt-fix targets plus glossary.md. The command actually run listed plan.md (0 hunks) and omitted glossary.md (4 hunks): 16 − 4 + 0 = 12. A number produced by one file set, reported against another.

  claimed set   conventions · judge · review · plan · HL · compilable_contract   → 12  ✗ reported
  ledger set    conventions · judge · review · glossary · HL · compilable_c.     → 16  ✓ correct
                                               ▲▲▲▲▲▲▲▲
                                               the four missing hunks

What did not change: every hunk still classifies as a substitution, a named debt fix or an addition, and the reviewer confirmed that independently at 100% file coverage. AC-7's claim holds; only its count was false — which in a phase about making rules findable is not a small thing. It is the same defect class as Phase C's truncated citation: a figure in an evidence file that a reader cannot reproduce.

RF — TFW-53 / Phase D: Glossary, Adapters and Version | 2026-08-14