RF — TFW-60 / Phase AC: Update Without Guesswork¶

Date: 2026-08-30 Author: Claude Code (Executor), on_behalf_of: saubakirov, via: claude-code Status: 🟢 RF — Complete Parent HL: HL-TFW-60 at 56c3d70, amendment A7 approved 2026-08-30 Phase HL: HL Phase AC TS: TS Phase AC — revision 2 (after onboarding) ONB: ONB Phase AC — two blocking questions, both ruled; nine recommendations, all ruled Commits: d2e6bae (ONB) · 4fe3b1d (scripts and their tests) · d047286 (workflow, canon, templates, adapters, release surface, integration tests) · this RF Measured against: b9baec2 (TS approval). The working tree also carried other sessions' edits (editions/, ASSISTED15, TFW-55 research); none entered any commit of this phase.

1. What Was Done¶

Every place where the update path guessed or decided for the owner now either asks, derives, or refuses:

BEFORE (v2.0.0-dirty.4)                           AFTER (this phase)

Step 0   source_head = HEAD                        Step −1  read the TARGET's update.md, follow it
         tag must equal HEAD → never on a          Step 0   operator names target_ref → source_head derived
         live source                                        from it → VERSION read from that commit = tag name
Step 3   "choose containers deliberately",         Step 3   🛑 three questions before the first durable write:
         handle from git user.name                          who acts · where tasks go · build.* — AG: one message
Step 5   cp -r payload → config overwritten        Step 5   copy loop with declared exclusions, prints "skipped:"
Step 6   7 rows, "managed CLAUDE.md content"       Step 6   Kind column: 5 copies (cmp) · 2 blocks (markers, §9 rule)
         allowlist: migration/changelog only                allowlist: text whose purpose is to retire the term
Step 7   {resolved-source}@{tag} → D:/… paths      Step 7   {upstream}@{verified-tag}; --check project reports a path
—                                                  Step 8a  briefing: Added·Changed·Fixed·Removed → four blocks
migrate  first [A-Z_]+ token → "DONE", no file     migrate  whole-or-UNDECLARED; own manifest heading; phases named
--check  "4 tasks validate" over 4 stateless       --check  failure on a live task; one informational line per
         phase dirs                                         terminal/stateless task
plan     ABBR = "UPD"                              plan     title first, then its initials, approved together

New Files¶

File	Description
`.tfw/templates/briefing.md`	The update's last message: four blocks bound to `Added` / `Changed` / `Fixed` / `Removed`, absent section → nothing in this release, no free text (A7)

Modified Files — 26 counted, 6 copies (S32), measured `git diff b9baec2 d047286`: 943+ / 172− over the counted set¶

File	Changes
`.tfw/workflows/update.md`	Rewritten at 1174 words (was 840; ceiling 1200): Step −1; Step 0 pin from `target_ref`; Step 3 🛑 gate with the three questions and the AG rule; Step 5 copy loop with exclusions and `skipped:`; Step 6 `Kind` column, §9 pointer, allowlist wording; Step 7 `{upstream}@{verified-tag}`; Step 8a briefing
`.tfw/scripts/migrate_board.py`	`classify_status()` whole-or-refuse: one declared token, no second declared token, no `So` symbol after it; `U+FE0F`/`U+200D` skipped; result gains `signals`. `render_manifest()`: Rows carrying more than one lifecycle signal, Phase directories (via `iter_phase_dirs`), Task state written note rewritten
`.tfw/scripts/gen_index.py`	`check_tasks()`: stateless phase directories — failure on a live task, informational line per terminal / stateless / malformed task, summary count, nothing written. `check_project()`: `installed_from` machine-local report, never rewritten
`.tfw/scripts/test_migrate_board.py`	+11 tests: `AILAB-2` exact shape, second emoji alone, second token alone, `→ + = <` non-signals, bare variation selector, refused row receives state, manifest headings, phase listing
`.tfw/scripts/test_gen_index.py`	+12 tests: stateless phase under live / terminal / stateless / malformed task, one line per task, nothing written, repository census (six tasks), `installed_from` reported ×3 and accepted ×4
`docs/scripts/test_integration.py`	Block sync helpers and tests (region equality, outside untouched, no-markers → None); root `CLAUDE.md` block checked with installed copies; no `{version}` in templates; project-owned payload files derived and required in Step 5; TD-198 row in `RETIRED_WORDINGS`; staging `update.md` path exempted
`.tfw/conventions.md`	§4 Identifier: acronym of the approved title, proposed with it, never derived silently said both ways; §4 Artifact file naming: current-grammar rows, no title appended; §9: the one marker rule (three cases)
`.tfw/glossary.md` · `.tfw/workflows/plan.md` · `.tfw/workflows/init.md` · `.tfw/templates/HL.md`	The same abbreviation rule in each carrier's words; worked examples title → initials; HL header Title then Abbreviation
`.tfw/adapters/claude-code/CLAUDE.md.template`	`<!-- TFW:CLAUDE:START/END -->` around the `## TFW` section; block brought current (`status.md` in context loading, `/tfw-knowledge`, `/tfw-config`, `Version: see .tfw/VERSION`)
`CLAUDE.md` (root)	Carries the block byte-identical to the template's; Conduct and Execution Modes follow it as project text
`.tfw/adapters/claude-code/README.md`	The managed block section; setup step says what is outside the block is the project's
`.tfw/adapters/codex/README.md`	Step 3: no markers → report and leave, pointer to §9 (was: append)
`.tfw/adapters/antigravity/tfw-rules.md.template` · `.agent/rules/tfw.md`	Template = rendered rule, byte for byte; `{version}` gone (TD-204)
`.tfw/adapters/cursor/tfw.mdc.template`	`{version}` ×3 gone; a whole copy (R2)
`.tfw/CHANGELOG.md`	`.3` entry: dated addendum quoting the retired sentence verbatim and what a project that went thin does (TD-198); `.2` entry: dated closing line on the `[TD-11](../../../reference/tech-debt.md)` paragraph (TD-191); `2.0.0-dirty` entry: `> Superseded by .tfw/migrations/2.0.0.md` above the migration fence (TD-190). No entry rewritten in substance; `[Unreleased]` untouched
`RELEASE.md`	§5: three checklist items — reach every earlier tag and open with Step −1; quote reversed wording verbatim; superseded-by line, entries append-only. §6 step 3 names them
`.tfw/templates/journal/event.md`	`via` free-form non-empty provider/tool text at `:49`, `:70`; "provider family is not a writer" reworded (TD-200)
`.tfw/templates/team/profile.md`	One file per person; role and context in `team/README.md`; `since` defined; `type: agent` admitted and unusable (TD-203)
`.tfw/templates/status.md`	The phase sentence, when to use which, who writes a phase file (by hand, never migration)
`.tfw/templates/project_config.yaml`	`installed_from` comment: the one form, never a machine path, `self` for the framework; marker stays `← PROJECT`
`.tfw/migrations/2.0.0.md`	Step 1 opens with the target's `update.md`; every command from the root (`cd .tfw` rewritten); `--working-tree` sentence; one manifest location `tasks/MIGRATION-2.0.0.md`; the two manifest sections to read; step 3a phase state by hand; multi-signal rows under `UNDECLARED`; three grammars unchanged
`.claude/commands/tfw-{update,plan,init}.md` · `.agent/workflows/tfw-{update,plan,init}.md`	Re-synced byte-identical in the same commit as their source (copies, not counted)

What a project on `2.0.0-dirty.2`, `.3` or `.4` must know (AC-11, for the `.5` entry and the briefing)¶

/tfw-release writes the entry; this is its updating section's required content, verified as a .2 reader's path in skipped_tag_read_through.txt:

Read the target's .tfw/workflows/update.md from the pinned payload and follow it (Step −1); pin from the tag you name, never from the source's HEAD.
On .2: also perform the .3 section — re-sync .claude/commands/tfw-*.md, add tfw.installed_from, delete per-session agent profiles; read the .3 entry's addendum on the retired sentence "Commands never duplicate workflow content — they reference it" and grep your own rule files for it.
On .2 or .3: also perform the .4 section — remove tfw.id_max_retries and review.default_mode; rename nothing.
Answer the three questions the update asks before its first write: who acts, where tasks go, build.*.
installed_from is {upstream}@{verified-tag}; --check project reports a machine path — record the upstream reference. All three local consumers carry a D:/ path today and will be told so.
CLAUDE.md: insert the  block once, from the template; until then Step 6 reports the file and leaves it.
--check tasks names a phase directory without status.md; under a live task that is a failure — author the file from the template's phase paragraph.
The abbreviation of a new task is the initials of its approved title, proposed and approved with it.
Read the briefing the update delivers last — four blocks, in your content_language.

The briefing for this repository's own .3 → .5 delta is rendered in briefing_dirty3_to_dirty5.md; its .5 bullets are re-derived from the cut entry.

2. Key Decisions¶

A second status signal is a character of Unicode category So, not any category S — the ONB's blocking question, ruled by the coordinator on the measurement: over 114 rows on four pinned corpora, category S entire refused three single-signal rows on a plus sign or an arrow ([TFW-52](../../TFW-52__tfw_light_v1/HL-TFW-52__tfw_light_v1.md), HD-30, live HD-31); So refuses exactly the eight rows carrying a second emoji or a second declared token. U+FE0F and U+200D are skipped explicitly. A declared token in outcome prose is a signal (HD-19 ✅ DONE (KNW deferred …)): the conservative direction, ruled and kept.
The old classifier stripped every symbol from the cell before matching; the new one keeps prose punctuation in the outcome. Three already-migrated rows would read +1 UX and → [TD-126](../../../reference/tech-debt.md) instead of 1 UX and [TD-126](../../../reference/tech-debt.md) if re-migrated; existing state files are not rewritten (immutable by rule), so nothing changes for them. Recorded so the reviewer does not read the three text-only differences as classification drift.
classify_status() returns a signals list — empty for a clean cell — and the manifest prints it. A refusal that names what it saw is what lets a person resolve the row with one transition event instead of re-reading the board.
AC-7: the exclusion list carries both project-owned files; the payload keeps carrying them (ONB rec. 1, accepted). A root .gitattributes export-ignore would govern git archive alone and add a root file outside the census; the loop in Step 5 covers every materialization and the payload test derives the excluded set from the payload itself (every root .yaml with a template counterpart), so a new project-owned file fails the test until it is excluded. The root cause — the payload carries this repository's own project_config.yaml and knowledge_state.yaml, which conventions §10.3 says are never sourced from upstream — is §6 O11 for the payload boundary.
Step 5's copy is an executable find | while read loop in POSIX sh, not rsync or cp --parents: it runs in Git Bash on Windows, where three consumers were updated, and it prints skipped: for each exclusion. A copy printing nothing skipped on a project that has both files is declared a failed step.
The TFW:CLAUDE block bounds the ## TFW section — context loading, commands, key references — and was brought current before being bounded (coordinator's widening of rec. 2): status.md in the context-loading order, /tfw-knowledge and /tfw-config in the table, Version: see .tfw/VERSION. Project identity, mandatory rules, execution mode and code standards stay outside. Root CLAUDE.md was reordered so the block is contiguous; its Conduct and Execution Modes text is unchanged and follows the block.
One marker rule, in conventions §9, for every block row: markers present → replace between them; file absent → create from the template; no markers → report and leave. The Codex README's append the complete block became the pointer. Appending to a file that already carries an unmarked hand-written TFW section would produce two sections — the fourth report's CLAUDE.md.
Cursor admitted to the census (Q2, ruled A): {version} ×3 removed, the template is a whole copy, the Step 6 table has no footnote.
--check tasks: a stateless phase directory is a failure only when the task's own status.md is live; terminal, absent or malformed task state makes it an informational line that says which — one line per task, a summary count, exit code unaffected (R2). On this repository: six lines, 17 directories, exit 0.
--check project on installed_from: a drive letter, a leading / or a backslash is machine-local and a problem (exit 1); self, unrecorded and a URL pass; nothing is rewritten. The three local consumers will be told at their next update — the intended effect.
The briefing reads four sections (A7) and nothing else: Why this release exists, Verification, Known open, Canon explain the release to its authors. An absent section yields nothing in this release; a bullet with no CHANGELOG bullet behind it is forbidden by the template. The positive framing the owner asked for is the block order, not adjectives.
TD-198 is now a registry row: "Commands never duplicate workflow content" in RETIRED_WORDINGS, scanned over the installed adapter surface — where the fourth report found the stale principle. The CHANGELOG quotes it (excluded from the scan by rule) and .tfw/adapters/claude-code/README.md keeps its blockquote (outside both scans).
Word ceiling met by removing duplication first: Step 3's team/ paragraph folded into the gate; Step 5's two-line recheck became one line against the derived commit; Step 8's list stopped repeating Step 6; Step −1 kept to three lines. 1174 of 1200.
Evidence at pins, never HEAD: the fabricated tag and the dry-run payload live in a scratch clone; the four corpora were read with git show {pin}:tasks/BOARD-SNAPSHOT.md; the consumers were read with grep, git show and sha256sum only.
[Unreleased] is untouched. The .5 entry is /tfw-release's act after review (AC-11); its required updating section is §1 above. Additions to .2, .3 and the 2.0.0-dirty entry are appended and dated.
templates/project_config.yaml keeps ← PROJECT on installed_from (ONB inc. 6, accepted): the key is framework-shipped, the value is project-written by each update; the marker names who writes the value.

3. Acceptance Criteria¶

AC-1 — the pin is derived from the tag¶

[x] target_ref named by the operator; source_head=$(git rev-parse --verify "$target_ref^{commit}"), never HEAD — update.md Step 0
[x] VERSION read from that commit and compared with the tag's name; a mismatch stops — Step 0; pin_on_live_source.txt Run 2
[x] Step 5 recheck against the derived commit: a source that moved elsewhere is not reported; a moved tag is — pin_on_live_source.txt last section
[x] status --porcelain -- .tfw/ rule as written
Gate: on a clone 29 commits ahead of v2.0.0-dirty.4, PASS; fabricated tag, STOP — both runs verbatim in the gates file

AC-2 — a receiver on any earlier tag finds its path¶

[x] Updating section names every intervening section; RELEASE.md §6 makes it a step — §5/§6, skipped_tag_read_through.txt
[x] RELEASE.md §5: reversed statement quoted verbatim; .3 entry gains the quote and what a project that went thin does (TD-198)
[x] .2 entry's [TD-11](../../../reference/tech-debt.md) paragraph closed, dated (TD-191)
[x] 2.0.0-dirty migration fence carries > **Superseded by**; RELEASE.md names the form (TD-190)
[x] update.md opens with Step −1; the guide's step 1 repeats it first; R2 the .5 updating section opens with it (§1 above)
[x] No entry rewritten in substance; additions appended, dated
Gate: the .2 reader's path recorded step by step

AC-3 — the gate can be literally green¶

[x] Allowlist admits text whose purpose is to retire the term, one sentence — Step 6
[x] The six known hits fall inside; a live use is not admitted — gates_and_word_count.txt: 0 outside here, 0 outside on innoforce-ai-first

AC-4 — every Step 6 row is executable the same way¶

[x] TFW:CLAUDE markers in the template, Codex pattern
[x] Step 6 table states copy / block per row; Claude rules and Codex routing are blocks, five rows copies
[x] R2 one marker rule in conventions §9, three cases; Step 6 and both READMEs point to it; Codex :104 is the pointer
[x] R2 the block brought current with the canon
[x] R2 root CLAUDE.md carries the block byte-identical; the installed-copies test checks it
[x] R2 Cursor template: {version} ×3 gone, a whole copy
[x] Antigravity template reads .tfw/VERSION as the rendered rule does; identical (TD-204)
[x] Claude adapter README describes the block and the first-run rule
Gate: fixture CLAUDE.md with text above and below — region equals the template's after sync, outside bytes unchanged (cmp both)

AC-5 — `installed_from` has one form¶

[x] {upstream}@{verified-tag} in Step 7 and the config template; local checkout named symbolically, path in the checklist
[x] --check project reports drive letter / leading / / backslash as machine-local; nothing rewritten
[x] self valid — this repository's --check project is consistent
Gate: D:/…@v2.0.0-dirty.4 → 1 problem; [email protected] → consistent

AC-6 — the owner is asked, not guessed¶

[x] Step 3 🛑 gate before the first durable write, exactly three questions; handle asked, never inferred; AG: one message, read-only steps continue, stop at the write
[x] Answers recorded in the checklist
[x] Step 8a briefing from templates/briefing.md; R2 absent section → nothing in this release; A7 Changed as a fourth block
[x] The briefing is the last message; the checklist records delivery
[x] update.md under 1200 words: 1174
Gate: AG dry run against a fixture consumer stops with the three questions, fingerprint identical before/after, git user.name present and unused; the .3 → .5 briefing rendered

AC-7 — the copy cannot overwrite project-owned files¶

[x] Step 5 names the exclusions; config merged key by key; state never touched
[x] The copy prints what it skipped; nothing-skipped on a project with both files is a failed step
[x] Executor decision recorded: exclusion list, payload keeps both (§2 item 4); R2 root cause in §6 O11
[x] Payload test covers the exclusion list: the project-owned set is derived from the payload and required in Step 5
Gate: fixture — both files byte-identical, both printed as skipped

AC-8 — the status cell parsed whole or refused; no phase left without a named state¶

[x] One declared token + free text with no further declared token and no further status symbol; R2 a symbol is category So; U+FE0F/U+200D skipped; fixture tests each signal alone, → and + as non-signals, a bare variation selector
[x] An UNDECLARED row is never terminal, never skipped by plan(); receives status.md at UNDECLARED
[x] Manifest heading Rows carrying more than one lifecycle signal; the Task state written note no longer implies skipped rows are terminal
[x] Manifest names every phase-* directory of every matched task and says phase state is not written by migration; author by hand
[x] --check tasks reports stateless phase directories: failure on a live task, informational otherwise; R2 one line per task, malformed task state → informational with the reason; on this repository six lines over 17 directories, exit 0; writes nothing
[x] templates/status.md carries the phase sentence and says when to use which
[x] The migration guide says all of it in the order met
[x] The AILAB-2 shape is a committed fixture: 22 of 28 new tests fail before, 28 pass after
Gate: fixture; --check tasks on a fixture and on this repository; four pinned corpora — 8 class changes, all multi-signal; 3 text-only differences (+/→ kept); 103 identical

AC-9 — the abbreviation is the initials of the approved title¶

[x] Conventions §4 Identifier: acronym of the approved full title, proposed together with the title, approved with it; never derived silently said both ways (never apart from a title, never without approval)
[x] plan.md 3.5 and 4.2, init.md Batch 1 and Mini-Setup 6, glossary: the same rule, worked examples title → initials (CRSW, ASSISTED15)
[x] HL template header: Title then Abbreviation, adjacent
[x] Artifact file naming: current-grammar rows with HL-TFW_20260829-172110_ABT.md, RES__TFW_20260829-172110_ABT.md; no title appended, the reason stated (TD-201)
Gate: five excerpts side by side in carriers.txt; gen_docs.py's task_id_source resolves both examples

AC-10 — carriers agree with the canon and each other¶

[x] event.md :49, :70: via free-form non-empty text (TD-200)
[x] profile.md: one file per person; role and context in team/README.md; no fifth key (TD-203)
[x] Guide: one manifest location (tasks/MIGRATION-2.0.0.md, beside the snapshot); the --working-tree sentence; cd .tfw && … rewritten from the root
[x] grep for the four retired wordings over the templates and the guide: nothing
Gate: the grep; the guide's retired-files command run from the root as written

AC-11 — the release describes what shipped¶

[x] executor: what a project on .2, .3, .4 must know — §1 above
[x] executor: every fifth-report §6 item and fourth-report defect 7 — fixed here or in §6:
manifest location → fixed (guide names tasks/MIGRATION-2.0.0.md, beside the snapshot)
role location → fixed (profile template: team/README.md)
since semantics → fixed (profile template: the date the participant joined the project — one line)
README route template → §6 O2 · stale index in a non-first container → §6 O3 · created seconds provenance → §6 O4 · --check project green before migration → §6 O5 · fourth-report defect 7 → §6 O6
[ ] /tfw-release, after review: version bump, CHANGELOG entry and tag as one act; the entry's updating section per §1 — not the executor's act
[ ] field: one consumer on the line updates to the new tag from Step −1 — after the tag exists
[x] 2.0.0 not claimed: VERSION reads 2.0.0-dirty.4

4. Verification¶

Lint (python -m pytest .tfw/scripts/ docs/scripts/ -q --collect-only): collects — every test file imports
Tests (python -m pytest .tfw/scripts/ docs/scripts/ -q): 315 passed, 1 skipped (baseline at b9baec2: 283 passed, 1 skipped; +32 tests, no test removed or weakened)
Verify (python .tfw/scripts/gen_index.py --check tasks): 54 tasks validate, 6 informational lines over 17 stateless phase directories, exit 0
--check project: consistent with the release it declares
--check index: stale — workspace/00-INDEX.md lags this phase's status.md and other sessions' task changes; a derived view with one writer, left for the coordinator to rebuild deliberately (§6 O12)
Adapter copies: update.md, plan.md, init.md byte-identical in .claude/commands/ and .agent/workflows/; .agent/rules/tfw.md identical to its template; root CLAUDE.md block identical to the template's
Word count: update.md 1174

5. Evidence¶

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

See EV file for evidence details — 37 rows, eight artifacts indexed.

Evidence verdict: 34/37 VERIFIED, 3 DEFERRED, 0 BLOCKED, 0 N/A — the deferred three are the .5 entry, the tag and the consumer run, assigned by the TS to /tfw-release and the field.

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

#	File	Line(s)	Type	Description
O1	`tasks/TFW-55__canonization_program/`	—	ux	The board snapshot reads `DONE`; the working tree carries live research (`research/iter2/` uncommitted, `iterations.yaml` modified) and no `status.md` at task or phase level. Its two stateless phases are informational by the rule — correct, but the rule's premise (a terminal task is history) is not true here. Seen by the coordinator; the owner authors the state
O2	`README.md` route section (every consumer)	—	duplication	Fifth report §6: add a permanent route has no template, so each project writes its own paragraph and rewrites it in three places on a container change. A `templates/readme_route.md` with `{container}` is a new payload path — outside this census
O3	`.tfw/scripts/gen_index.py` `build()`	~1140	ux	Fifth report §6: after the first container changes, the old `{old}/00-INDEX.md` stays on disk as a stale copy and nothing names it. A line stale index at {old}: delete it when `00-INDEX.md` is seen in a non-first container. Phase B touches `gen_index`
O4	`.tfw/scripts/migrate_board.py` `build_status()`	~520	ux	Fifth report §6: `created` is taken from the directory's first commit (second-accurate) while the board carried a day; the provenance comment does not say where the seconds came from, and the status template promises a declared zero time for day-only sources. One sentence in the comment
O5	`.tfw/scripts/gen_index.py` `check_project()`	~1290	ux	Fifth report §6: with the board still in the README and no `status.md` yet, `--check project` answers consistent. One line — board still present at {path}: migration pending — when a `## Task Board` heading is found in the root README
O6	`.tfw/scripts/gen_index.py` `check_project()`	~1250	missing-test	Fourth report defect 7: `--check project` reads `team/` whole but not who references it; a deleted human profile leaves `on_behalf_of` dangling and only `--check tasks` goes red. Filed in the TS as an observation; `--check tasks` already catches it
O7	`team/README.md` (this repository)	3–4	style	Says humans and agents alike — that is why this container is not called `people/` — the contradiction TD-203 removed from the template stands in the project's own README, which the template now names as where a role goes. Project text, outside the census
O8	`docs/scripts/gen_docs.py`	766	style	Imports `mkdocs_gen_files` at module load, so the module cannot be imported outside a docs build (`mkdocs.yml` must exist) — an executor checking `resolve_references()` from the repository root gets a `ConfigurationError`. Guarding the import behind the build entry point would let the resolver be unit-checked
O9	consumers' migrated `status.md` (`HD-30`, `HD-31`, `[TFW-52](../../TFW-52__tfw_light_v1/HL-TFW-52__tfw_light_v1.md)`)	`outcome`	naming	The old classifier stripped every symbol before matching, so three already-written outcomes read `1 UX [TD-271](../../../reference/tech-debt.md)`, `fieldworker role directed assignment`, `blocked [TD-126](../../../reference/tech-debt.md)`. Immutable by rule; a person may correct with a `transition` event if the text matters
O10	`.tfw/adapters/codex/README.md`	96–110	style	Step 3 still says if root `AGENTS.md` is absent, create it with project guidance plus the block — consistent with §9 case 2; not changed. Noted because the Claude README now says the same in different words; a future pass may make the two paragraphs one pointer each
O11	payload boundary (`.tfw/project_config.yaml`, `.tfw/knowledge_state.yaml`)	—	security	Root cause behind AC-7, recorded on the coordinator's ruling: the payload carries this repository's own two state files, which conventions §10.3 says are never sourced from upstream. The exclusion list makes the copy obey the rule; the payload still ships them. A Phase AA-surface question for a later ruling — `.gitattributes export-ignore`, or a payload that is `.tfw/` minus two files
O12	`workspace/00-INDEX.md`	—	ux	Stale after this phase's transitions and other sessions' task work. Derived and non-blocking by design; the coordinator rebuilds it deliberately with `python .tfw/scripts/gen_index.py` — one writer, not the executor mid-phase

7. Fact Candidates¶

fact-candidates: processed 2026-08-30

Human-sourced only; the coordinator's rulings and the owner's words are the source.

#	Category	Candidate	Source	Confidence
1	stakeholder	The owner wants change presented positively and in this order — what is now possible, what changes in how you work, what was repaired, what you no longer have to do — "because people do not like change"; the structure is the framing, not adjectives	fifth field report, owner quote; HL §12 A7	High
2	philosophy	The framework is its own first consumer or the mechanism is untested where it lives: a rule this repository does not obey is advice. Applied: root `CLAUDE.md` carries the marker block and a test checks it like every installed copy	ONB §8, coordinator ruling on rec. 3	High
3	convention	Boards carry status markers beyond the declared glyph set — `🔄` for a live phase, `🟢` in outcome prose — so declared glyphs only is the wrong rule for a second signal; Unicode `So` is what a person reads as a marker, and `Sm` (`+ → =`) is prose	ONB §8, coordinator ruling on Q1, on the executor's four-corpus measurement	High
4	process	Two adapters with two first-run rules is the drift class this phase closes elsewhere; fixing one and filing the other "would be the scar tissue TFW-57 measures" — a rule is stated once and pointed to, never restated per adapter	ONB §8, coordinator ruling on inc. 1	Medium
5	process	The coordinator does not read around frozen text even when the reading is obviously right: the briefing's `Changed` input went through §12 as A7 and was approved the same day rather than being assumed	ONB §8, risk 5; HL §12 A7	High

8. Strategic Insights (Execution)¶

fact-candidates: processed 2026-08-30

#	Insight	Category	Source
S1	Mechanizing a copy of stale content makes the mechanism precise about the wrong text. The coordinator widened the marker deliverable with an obligation: bring the block current with the canon (context loading order, command table) before bounding it. Implication: every future "make this sync mechanical" deliverable carries a content audit of the thing being synced as its first step, or the sync will faithfully propagate a contradiction to every consumer	philosophy	Coordinator, ONB §8 ruling on rec. 2
S2	The least structured CHANGELOG section carries the most consequential news. Reversals (`Copies are the model`) and grammar changes live under `Changed`; a briefing built only from `Added` / `Fixed` / `Removed` would have printed nothing in this release over the two largest changes of the line. Implication for `RELEASE.md`: a reversal belongs under `Changed` with the retired wording quoted, and the briefing block what you now do differently is the reader's contract that the section will be read to them	process	Coordinator, A7 evidence column; owner approval 2026-08-30
S3	A refusal earns trust only when the operator agrees the refused thing was ambiguous. The owner-side test of the status rule was not "does it catch `AILAB-2`" but "does it refuse `A+B`": a gate that refuses plainly single-signal rows teaches operators to override refusals, and then the real refusal is overridden too. Implication: every whole-or-refuse rule this framework adds is measured on real corpora for false refusals before it is fixed in a fixture — the order this phase followed	philosophy	Coordinator, ONB §8 ruling on Q1

9. Diagrams¶

The status cell, whole or refused — one rule, applied after the identifier's:

cell ──► strip leading marker run (any S, U+FE0F, U+200D, spaces)
     ──► first [A-Z_]+ token
            │ not declared ──────────────────────────────► UNDECLARED, verbatim, signals=[]
            │ declared
            ▼
        trailing text (U+FE0F/U+200D removed)
            │ any [A-Z_]+ word ∈ declared ─┐
            │ any char with category So ───┴► UNDECLARED, verbatim, signals=[what was seen]
            │ neither                                       (manifest: "Rows carrying more than one lifecycle signal")
            ▼
        lifecycle = token · outcome = trailing, dashes trimmed
        Sm/Sc/Sk (+ → = < $) and prose pass through untouched

--check tasks, the phase-directory decision:

for each task dir ── for each phase-* without status.md
     task status.md ── absent ───────────► note (informational): "carries no status.md of its own"
                    ── malformed ────────► note (informational): "malformed, reported above"   (the malformed state is the failure)
                    ── lifecycle ∈ {DONE, REJECTED} ► note (informational): "the task is DONE"
                    ── live ─────────────► FAILURE: "{task}: N phase directories carry no status.md while the task is {lifecycle}: …"
one informational line per task · summary count · exit code from failures only · nothing written

The update, where it stops and what it derives:

operator ──target_ref──► Step 0: source_head = rev-parse target_ref^{commit}; VERSION(source_head) == tag name? ── no ──► STOP
                              │ yes: archive source_head → .tfw/.upstream (staging)
                              ▼
                         Step −1: follow .tfw/.upstream/.tfw/workflows/update.md
                              ▼
                         Steps 1–2 (read-only) ──► Step 3 🛑 who · where · build.*  ── unanswered ──► STOP before the first durable write
                              │ answered → checklist
                              ▼
                         Step 5 copy: every payload file except project_config.yaml, knowledge_state.yaml → "skipped: …" ×2
                         Step 6 copies (cmp) · blocks (region between markers; no markers → report, leave)
                         Step 7 installed_from = {upstream}@{verified-tag}
                         Step 8 --check project ──► Step 8a briefing (Added·Changed·Fixed·Removed) = last message

RF — TFW-60 / Phase AC: Update Without Guesswork | 2026-08-30