title: "Map — revision 3 — "What was done?"" source: "tasks/TFW-60__conflict_resistant_shared_workspace/phase-aa/review/rev3/map.md"

Map — revision 3 — "What was done?"¶

Mindset: Experienced newcomer. Understand before you judge. Scope: the R4 corrective pass — TS revision 4, AC-15, twelve items. Earlier passes stand: ../map.md · ../rev2/map.md TS: TS Phase AA at revision 4 RF: RF Phase AA — still at revision 2. It does not describe this pass. Commits: 22de861 · d62fd26 · 2d269e4 · f7c9dfe · 94f02a6 · fd85b7c · 5e9b0a1 · b75bef1 Owner's stated emphasis: quality, architecture, value, goal. Counts are not the question.

Understanding¶

Two things happened, and only one of them is a correction.

The awaited evidence arrived. A second real external project, innoforce-ai-first, ran the full 1.3.0 → 2.0.0-dirty.2 update including the board migration, and wrote its own retro. The coordinator filed it verbatim at task root. It says the phase achieved what it was for: the first consumer "spent the rest of the session reconstructing what to do and in what order"; this one spent nothing — Step 3 routed it to the guide and the guide had the order with the consequence of skipping each step. Zero unrecognized directories, zero unaccounted rows, and the identity gate refused a bad value instead of swallowing it.

And it named a design error. Writing an obvious actor: claude-code was refused; the operator read the validator's source to invent claude-20260828a; two external projects ended up minting a profile per agent session, and one later deleted them — leaving its build gate red permanently, because events are immutable and profiles are not. The owner ruled: actor carried two unrelated jobs — say who wrote this and make the filename unique — and those two contradict each other. Remove it until TFW-54.

So the pass removes a field from the event carrier, replaces the filename's third component with a four-hex opaque token whose only job is uniqueness, defines bindings.yaml (instructed by seven workflows, defined by none), admits per-phase journals to the model, adds the missing Claude Code row to Step 6, records installed_from, and rewrites §10.4 as a rule rather than patching its example.

TS ↔ RF Alignment¶

This is the finding of the Map stage, and it is not a small one. The RF header reads "revision 2, after REVIEW 440d6fd", cites the TS at revision 3, and its §3 acceptance table has fourteen rows and no AC-15. Neither 5e9b0a1 nor b75bef1 touched it.

TS revision 4	RF says	Aligned?
AC-15, twelve items, one of them architectural	nothing — the AC does not appear in the RF	❌
The `actor` removal, its rationale, its tolerance rule	nothing — no §2 decision row	❌
`260 passed 1 skipped`, three gates, mkdocs	in `evidence/r4_gates.txt`, not indexed in the EV	⚠️
`update.md` at 1380 words against a ceiling of 1200	`r4_gates.txt` says "OVER, see the RF" — and there is no RF to see	❌
AC-13 half two	RF still reports UNMET; EV E63 records the arriving artifact and routes the ruling to the reviewer	⚠️ correct routing, stale RF

The evidence artifacts themselves are good — census_r4.md measured before the first edit, r4_gates.txt, ac15_actor_tolerated.txt. They are simply not gathered into the two documents a review reads.

Deviations from TS¶

The TS contradicts itself, and the executor is not the cause. §1 still says "The model itself does not change. No carrier, schema, vocabulary, lifecycle value or identifier rule is touched"; §7 DoF still ends "❌ The model changed: any edit to a carrier schema, the event grammar…"; §8's last risk row says "A finding about the model is filed, not fixed here"; §9 says the phase "changes no carrier they will extend." AC-15 mandates exactly what those four forbid. Coordinator's, not executor's.
PROVIDER_FAMILIES was deleted rather than documented. AC-15 item 2 asked for it to be named in payload prose. With actor gone its only reader — the gate refusing a provider name in actor — has no subject, so the set went instead of the prose. Better than specified, and it needs a line in an RF that does not exist.
census_r4.md raises the budget under the two methods (22 by ruling S32, 38 by distinct paths) and proceeds under the declared one, before acting. That is the return-to-coordinator rule working as designed.

Checkpoint¶

Self-check: - [x] Read RF §1–§5 completely? — yes, and established that it predates this pass - [x] Read TS DoD and matched each item to RF §3? — AC-15's twelve items match nothing in the RF - [x] Read HL §7 Principles — can I state the design philosophy? — principle 3 (one normal writer) and the frozen §3.1 question "who/what acted" are what the identity model answers, and the R4 change is measured against them in verify - [x] Read ONB — were blocking questions resolved? — the eight stand; R4 raised none

Stage complete: YES