docs: nine statements that had stopped being true, and a closed issue four docs still cited - #592

… four docs still cited

The 2026-09-07 expert review accepted a documentation slice that never landed. This is
that slice, re-verified against the tree today rather than taken from the review, plus
three staleness defects the review predates.

Always-loaded, so every session inherited them:

CLAUDE.md said MM-2/3/4 were all blocked on externals. MM-2's build completed on
2026-09-12 and #574 shipped on 09-16; MM-3's design half shipped as #569. Only MM-3's
wiring and MM-4 are blocked.

CLAUDE.md's standing correction 1 said MIPS CQMs are "sunsetting ~PY2030". The fact
sheets propose EXTENDING them for PY2027+. The sunset that does exist is a different
policy — traditional MIPS as a reporting option, after PY2028 — and merging the two is
what the correction was supposed to prevent. Also relabelled: the FHIR transition is an
RFI seeking comment, which is weaker than a proposal, and the entry called itself
PROPOSED while its own body said "seeks comment on".

AI_GUARDRAILS attributed "the user must not proceed" to CDS Hooks' `critical`. The 2.0
spec defines it only as the top of an urgency ordering. The refusal is unchanged; it
rests on WorkWell being supplementary to WebChart, not on a spec sentence that does not
exist.

Stakeholder-facing:

WHAT_WORKWELL_IS said the Maui sandbox runs WorkWell's AUTHORED cms122/cms125. It has
routed six official measures since ADR-078 on 2026-09-08. The sentence has been flatly
false for twelve days in the document we hand to people.

Governing:

ROADMAP called the CMS137 final-rule confirmation "a formality to record". The sandbox
routes cms137 today, so the rule decides whether the PHI phase inherits it or the
routing comes out. The plan does not wait on it; the answer is not foregone.

ARCHITECTURE described the test-fixture gate as if it executed measures. `validateTests`
is four structural assertions — fixtures exist, each names a fixture, a subject and an
expected outcome. Studio's tick on that row means well-formed, not run-and-passed. The
Studio label itself still overstates and is a product change, not prose, so it is filed
rather than fixed here.

OSHA_1910_95_STS said the (g)(8)(ii) chapeau "switches off every follow-up action".
(g)(8)(i)'s 21-day written notification is a separate paragraph and is unconditional —
it survives a determination that the shift is not work related. A regulatory misstatement
in our own measure documentation.

STANDARDS_CONFORMANCE said occupational measures "cannot acquire" an answer key. A route
exists and is on the record — a steward publishing specifications and expected results,
or an independent implementation to disagree with. Neither exists nor is in progress,
which is the claim that holds: nothing here is graded by anyone but us.

Staleness the review predates:

guide ch.9 is the chapter that claims to own current state and still called #557 blocked
four days after it merged, and carried a 2026-08-08 suite count of 1,940 against today's
2,804.

And #533 closed on 2026-09-08 with its ingest half unshipped, while MEASURES, DECISIONS,
DEPLOY, DATA_MODEL_CONTRACTS and ROADMAP all cited it as the open tracker for that half.
A closed issue cannot carry work, so #591 does; every citation now points there. Journal
entries are left alone — they record what was true when written.

Review caught that the row I wrote reported 23 skips in one column while the next column
still said "The 15 skips" — I updated the count in one half and left the prose in the
other, which is the defect the row exists to prevent: a maintained state chapter giving
two contradictory numbers for the same run.

It also questioned the totals. Its arithmetic over journal deltas gave 2,801/2,777;
I had written 2,804/2,780 from a run taken before #588's review-fix commit added two
more tests. Rather than reconcile two pieces of reasoning, I re-ran the suite against
main's backend: 2,807 total, 2,783 pass, 1 fail, 23 skip. Neither prior figure.

Both columns now come from that one run, and the row says which run it was.