TS — TFW-56: Remove the Review Mode Axis¶

Date: 2026-08-13 Author: Coordinator (Claude Code) Status: 🟡 TS — approved by the owner 2026-08-13 Parent HL: HL-TFW-56 — 🔒 FROZEN, re-frozen 2026-08-13 after A1–A5, A7 Contract baseline: git log -E --grep="^\[[^]]*/TFW-56/freeze/" — two commits: initial freeze and re-freeze Research: iteration 1 — 637 rows / 203 reviews / 3 installs. H3 refuted; the promotion set below is the corrected one

1. Objective¶

Delete the code / docs / spec selection from review — the config key, the 🛑 WAIT gate, three byte-identical mode files and four template fields — and promote the checks it was gating into the universal Judge checklist, so every review asks them instead of one genre in three. The selection is what a 637-row measurement found to be ceremony (0 verdict flips in 203 reviews); the checks are not (~8% firing, matching the universal rows they join). This is a single phase: 22 files touched, almost entirely deletion, with one recorded grep as its acceptance test.

2. Scope¶

In Scope¶

Deletion of .tfw/workflows/review/ and its three mode files
review.md: mode step, WAIT gate and mode-file load removed; steps renumbered contiguously
Universal Judge checklist corrected to 10 rows — U2 sharpened, S1/S2/S4 promoted, Content quality dropped
The three orphaned docs/spec verify actions given an unconditional home in verify.md, or declined in writing
Mode: / Review Mode fields removed from all four templates
tfw.review.default_mode removed from both config files; min_verify_ratio untouched
config.md propagation table corrected
conventions.md review entry + one anti-pattern; glossary.md name disambiguation
Six adapter and entry-point copies re-synced
Version bump, CHANGELOG entry naming the removed key, TD-106 closed

Out of Scope¶

The general update.md removed-key rule — A5 was narrowed by the owner to the CHANGELOG clause only. The framework-wide rule is deferred (HL §8)
The rigour axis — 8 of 13 field qualifiers encode verification depth, not genre. Touching it means touching min_verify_ratio, which DoF-4 protects. Sibling task
Review consolidator / subagents — TFW-45 addendum, downstream of TFW-53 Phase C
Goal defence in review — TFW-53 Phase C owns it. This task must not pre-empt its enforcement site
Rewriting history — existing REVIEW files keep their mode headers; past CHANGELOG entries stay as written
min_verify_ratio behaviour or value — it survives the removal of its sibling key unchanged

3. Principles Check¶

#	Principle (from HL §7)	Enforced by	Gate
P1	Delete, don't relabel	AC-1, AC-6	Folder absent; no `Mode:` field, no `default_mode`, no `none` placeholder value anywhere — AC-12 grep
P2	No coverage loss without a recorded home	AC-3, AC-4	RF carries a row for each of the 8 checklist rows and each of the 3 verify actions: promoted / already covered / declined with reason
P3	A check that cannot fail is not a check	AC-2	Each promoted row ships with its measured firing rate cited; a row without one violates DoF-2 as sharpened by A6
P4	What to check is declared once	AC-1, AC-7	No second declaration survives: no mode step, no config key, no propagation row
P5	Explicit N/A over silent skip	AC-2	Skipping a promoted row must be visibly marked; a silent ✅ fails the AC
P6	Single Source of Truth, behavioural adapter parity	AC-9	Six copies carry no mode reference and behave as `.tfw/` does
P7	History is evidence, not debt	AC-12	Existing REVIEW headers and past CHANGELOG entries unchanged in the diff
P8	Structural enforcement over promise	AC-12	The acceptance test is a recorded command with its output in the EV file, not a claim

4. Affected Files¶

File	Action	Description
`.tfw/workflows/review/code.md`	DELETE	16 lines
`.tfw/workflows/review/docs.md`	DELETE	13 lines
`.tfw/workflows/review/spec.md`	DELETE	13 lines — folder becomes empty and goes with them
`.tfw/workflows/review.md`	MODIFY	Remove Step 1 (mode selection + 🛑 WAIT) and the mode-file load in Verify; renumber remaining steps contiguously; fix internal step references
`.tfw/templates/review/judge.md`	MODIFY	Remove `Mode:` and the Mode-Specific section; sharpen U2; add S1, S2, S4 with structural explicit-N/A grammar
`.tfw/templates/review/verify.md`	MODIFY	Remove `Mode:`; add the three `docs`/`spec` verify actions as unconditional, or none if declined
`.tfw/templates/review/map.md`	MODIFY	Remove `Mode:`
`.tfw/templates/REVIEW.md`	MODIFY	Remove `Review Mode` header field and the mode placeholder comment; realign §3 Judge table row-for-row with judge.md
`.tfw/project_config.yaml`	MODIFY	Remove `tfw.review.default_mode`; leave `min_verify_ratio`
`.tfw/templates/project_config.yaml`	MODIFY	Same
`.tfw/workflows/config.md`	MODIFY	Drop the `review.default_mode` row; correct the `review.min_verify_ratio` row's step pointer
`.tfw/conventions.md`	MODIFY	Review subfolder entry cleared of mode vocabulary; §14 anti-pattern added
`.tfw/glossary.md`	MODIFY	Disambiguate `Reviewer (AI — coordinator in review mode)` so "review mode" has one meaning
`.tfw/VERSION`	MODIFY	Bump
`.tfw/CHANGELOG.md`	MODIFY	New entry; `### Removed` names the config key, not only files
`.claude/commands/tfw-review.md`	MODIFY	Re-sync from `.tfw/workflows/review.md`
`.claude/commands/tfw-config.md`	MODIFY	Re-sync from `.tfw/workflows/config.md`
`.agent/workflows/tfw-review.md`	MODIFY	Re-sync
`.agent/workflows/tfw-config.md`	MODIFY	Re-sync
`.tfw/adapters/codex/skills/tfw-review/SKILL.md`	MODIFY	Remove the "review-mode WAIT gate" instruction
`.agents/skills/tfw-review/SKILL.md`	MODIFY	Same, installed copy
`TECH_DEBT.md`	MODIFY	Close TD-106 with the reason
`evidence/EV__TFW-56__review_mode_removal.md`	CREATE	Mandatory evidence artifact

Budget: 0 new framework files (1 new evidence artifact), 19 modifications, 3 deletions — 22 framework files touched. Project limits: 30 files, 15 new, 3000 LOC, 30 modified. Within budget on every axis; net LOC is negative.

5. Acceptance Criteria¶

AC-1: The selection is absent from the framework¶

No path remains by which a reviewer is asked to choose a review mode. - [ ] .tfw/workflows/review/ does not exist - [ ] review.md contains no mode step, no mode WAIT gate, no mode-file load - [ ] review.md steps are contiguous with Step 0 = Session Naming — the TFW-standard opening, which this file has never had - [ ] Every internal reference in review.md to a renumbered step resolves to the intended step Gate: ls .tfw/workflows/review/ fails; read review.md end to end and confirm step numbering and internal references Evidence: record the ls failure and the full renumbered step list in the EV file

AC-2: The universal Judge checklist is the corrected ten rows¶

The checklist carries every check that survived the measurement, and no row without an evidenced rate. - [ ] judge.md has no Mode: field and no Mode-Specific section - [ ] U2 Philosophy aligned is sharpened to cover design soundness — whether the design itself serves the stated principles, not whether it is named well - [ ] Three rows added: S1 evidence sufficiency, S2 backward compatibility, S4 safety - [ ] S1 is worded so it cannot be read as U7: U7 asks whether evidence exists, S1 whether it establishes the claim - [ ] Content quality is not present — it was the one true duplicate - [ ] Each promoted row carries its measured firing rate as justification for being there - [ ] Skipping a promoted row requires a visible mark; a silent ✅ is not a permitted way to skip Gate: judge.md has exactly 10 universal rows; read S1 and U7 side by side and state in the RF why they cannot collapse Evidence: fill the corrected checklist against one archived RF/REVIEW pair from this repository and record whether S1 and U7 produced different answers. If they produced the same answer, the naming failed and AC-2 is not met

AC-3: Every removed check has a recorded home [depends: AC-2]¶

Nothing disappears silently — the accounting is in the RF, not in the reviewer's memory. - [ ] All eight mode checklist rows appear in the RF, each marked promoted (naming its destination), already covered (naming the universal row that holds it), or declined (with a reason) - [ ] The accounting matches what the files actually do Gate: cross-read the RF table against judge.md; any row whose stated destination is not in the file fails the AC Evidence: N/A — this is a documentation-completeness check, verifiable by reading

AC-4: The three orphaned verify actions have a home¶

They are Verify-stage actions, so a promoted Judge row does not rescue them. - [ ] spot-check 2-3 key claims/sources, check citations traceable to real artifacts, verify data claims against primary sources are each either present in verify.md as an unconditional action, or declined in the RF with a written reason - [ ] All four code-mode verify actions remain mandated — the build/test command by verify.md Checkpoint, the test-file check by the review.md Trust Protocol, the ratio and the DoD cross-check by verify.md - [ ] map.md and verify.md carry no Mode: field Gate: grep each action's substance in verify.md; for any declined action, confirm the RF carries the reason Evidence: N/A — file content, verifiable by reading

AC-5: The REVIEW template matches the new checklist [depends: AC-2]¶

The synthesis document cannot list checks the stage file does not have, or omit ones it does. - [ ] No Review Mode header field, no mode placeholder comment - [ ] §3 Judge table matches judge.md's ten rows one-for-one, in the same order Gate: diff the two row lists; any mismatch fails Evidence: N/A — template content

AC-6: The config key is gone and its sibling is untouched¶

[ ] tfw.review.default_mode absent from .tfw/project_config.yaml and .tfw/templates/project_config.yaml
[ ] min_verify_ratio present in both, value 0.42, with its comment intact and its meaning unchanged Gate: grep -n "review:" -A4 on both files Evidence: record both file excerpts in the EV file — this is the one place where collateral damage would be silent

AC-7: The propagation table is correct [depends: AC-1]¶

[ ] config.md no longer routes review.default_mode
[ ] Its review.min_verify_ratio row names the correct step number in the renumbered workflow — the pointer is stale today and must not stay stale Gate: read the row, open review.md at the named step, confirm it is the Verify step Evidence: N/A — cross-reference check

AC-8: Conventions and glossary carry no mode vocabulary¶

[ ] The Review subfolder entry in conventions.md describes stage files without reference to modes
[ ] conventions.md §14 carries a new anti-pattern: adding a review checklist row whose firing rate is not evidenced — so the axis cannot regrow under a new name
[ ] glossary.md defines no review-mode term, and the phrase "review mode" has exactly one meaning across .tfw/ Gate: grep -rn "review mode" .tfw/ — every remaining hit means the same thing Evidence: N/A — text content

AC-9: Adapter parity [depends: AC-1, AC-7]¶

[ ] All six adapter and entry-point copies carry no mode reference
[ ] Each behaves as its .tfw/ source does — parity is behavioural, not byte-level
[ ] The Codex skill no longer instructs the agent to follow a "review-mode WAIT gate" that does not exist Gate: grep the six files for mode vocabulary; diff each against its source for behavioural drift Evidence: record the grep across all six paths in the EV file

AC-10: Version and changelog [depends: AC-1]¶

[ ] VERSION bumped
[ ] CHANGELOG entry records the deletion, the corrected promotion set and the revocation of the mode-file decision
[ ] Its ### Removed block names tfw.review.default_mode explicitly — /tfw-update triages files, so a removed key is otherwise invisible to a project upgrading Gate: read the entry; confirm the key appears by name Evidence: N/A — changelog content

AC-11: TD-106 closed¶

[ ] TECH_DEBT.md marks TD-106 closed with the reason: the non-standard Step 0/Step 1 anomaly was deleted, not annotated Gate: read the row Evidence: N/A

AC-12: The sweep is complete and history is intact [depends: AC-1..AC-11]¶

The acceptance test is a command, not a claim. - [ ] grep -rn "code / docs / spec\|default_mode: code\|Review Mode\|review/{code" .tfw/ .claude/ .agent/ .agents/ --exclude=CHANGELOG.md returns zero matches - [ ] No existing task REVIEW file is modified — every Review Mode header in tasks/ survives unchanged - [ ] No past CHANGELOG entry is edited; only a new entry is added - [ ] git diff --stat shows 3 deletions and no framework file created Gate: the grep, plus git status/git diff --stat confirming nothing under tasks/ was touched except this task's own artifacts Evidence: record the grep command with its output, the diffstat, and the count of surviving Review Mode headers in tasks/

Evidence Artifacts¶

File	Description
`evidence/EV__TFW-56__review_mode_removal.md`	Environment header; per-AC table; the recorded grep gate with output; the deleted-folder check; both config excerpts; the six-adapter grep; the diffstat; and the S1-vs-U7 dry-run result (required)

6. Technical Guidance¶

Reference material, not instructions. Deviate with justification in the RF.

The renumbering trap is documented and this is its second occurrence. TD-106 exists because review.md was renumbered once before: it is the only TFW workflow whose Step 0 was not Session Naming. Removing the mode step finally makes it standard. Check config.md and its two adapter copies — they still point at "Step 0: Select Review Mode", which has been Step 1 since an earlier task. They are already wrong, and they are in scope.
The three mode files are byte-identical across three separate TFW installs. Read them before deleting so the coverage accounting in AC-3 is line-based rather than remembered.
The measured firing rates, for AC-2's justification: Analytical quality 25.0% · Test coverage 23.4% · Source attribution 22.2% · Source verification 12.5% — these four are S1, combined 16.1% across 174 rows. Breaking changes 8.5% → S2. Content quality 5.9% → dropped. Code quality 4.5%, six hard failures, all contract violations rather than style → S3, folded into U2. Security 4.0% → S4, retained on consequence rather than rate.
Adapters under .claude/ and .agent/ are full copies, not thin references — re-sync means overwrite from the .tfw/ source, then re-read to confirm nothing project-specific was clobbered.
min_verify_ratio is the collateral-damage risk. It lives in the same tfw.review block as the key being removed. AC-6 exists specifically to prove it survived.
Shared files with TFW-53 Phase C: review.md, judge.md, REVIEW.md, glossary.md, conventions.md §14. Phase C will replace U2's mapping-integrity check with a substantive purpose check — so sharpen U2, do not restructure it, and leave the Judge template's overall shape intact. If a change here would make Phase C's frozen deliverables unlandable, stop and report rather than solving it.
Line endings: this repository has CRLF working-copy files. Do not reformat whole files; keep diffs to the lines that change.

7. Definition of Failure¶

❌ A check or verify action available today disappears without a recorded home in the RF
❌ A promoted row ships without its measured firing rate, or the promoted set exceeds ten rows
❌ S1 and U7 are worded so that a reviewer answers them the same way — the naming is the deliverable, not decoration
❌ A stale step reference survives anywhere in .tfw/ or the six adapter copies
❌ min_verify_ratio changes value, comment or behaviour
❌ Any existing task REVIEW file or past CHANGELOG entry is edited
❌ Adapters left desynced, or the Codex skill still naming a WAIT gate that no longer exists
❌ The axis is renamed rather than removed — a "for information" mode field, a default_mode: none, or an empty review/ folder
❌ The grep gate is reported as passing without its output recorded in the EV file
❌ A change here forces an amendment against TFW-53's frozen sections

8. Phase Risks¶

Risk	Mitigation
Renumbering leaves a stale pointer — the exact trap TD-106 records	AC-1 reads the file end to end; AC-7 resolves the one known external pointer; AC-12's grep is the backstop
S1 collapses into U7 within a few reviews	AC-2's Evidence is a real dry-run against an archived review, and a same-answer outcome fails the AC rather than being noted as a caveat
Ten rows is at the edge of the readable band	U2 absorbs S3 instead of an eleventh row being added — the owner-approved modification. No further row may be added in this task
`min_verify_ratio` damaged as collateral	AC-6, with both file excerpts recorded as evidence
Collision with TFW-53 Phase C in five shared files	§6 constrains the U2 change to a sharpening; DoF forbids resolving any conflict by touching TFW-53's frozen sections. This task lands first by owner decision
Adapter copies drift because they are full copies	AC-9 checks behaviour, not bytes, and names the Codex skill line explicitly
The deletion looks like simplification while quietly losing the strongest check	AC-3 and AC-4 exist for precisely this; the whole reason the frozen contract was amended is that the first version of this task would have lost about two thirds of S1

TS — TFW-56: Remove the Review Mode Axis | 2026-08-13