fix(artifacts): move 276 statuses onto rivet's lifecycle vocabulary

[avrabe]

Closes #371.

Stacked on #368 (base is fix/360-rivet-source-dedup); GitHub will retarget to main when that merges. The two touch artifacts/requirements.yaml and so does #373 — whichever lands last needs a rebase.

What this is

spar carried a private status vocabulary on 276 artifacts. None of the four words is in the schema's allowed set, so every one was a status-allowed-values error and rivet validate could not run clean.

from	to	n	lossy?
passing	implemented	177	no
planned	proposed	40	no
not-implemented	proposed	51	no
partial	approved	8	yes

passing is the group worth naming. It is a verification verdict sitting in a lifecycle field, and a hand-edited YAML file cannot honestly assert that a test passes — only that the measure exists. That is what implemented says.

The lossy eight

partial sits strictly between approved and implemented, and the vocabulary has no such rung. approved is the conservative floor — the strongest claim certainly true — rather than implemented, which would overstate.

Two of the eight (RENDER-REQ-003, COVERAGE-GAPS) already itemized their residue in prose and were left alone. The other five now say the residue is not itemized. That wording is deliberate and I want it reviewed as a choice, not skimmed as boilerplate: lower.rs has 54 #[test] functions, but STPA-REQ-036 requires one test per lowering rule — a mapping question a count cannot answer. Writing "54 tests exist" would be the necessary-but-not-sufficient trap. An explicit "not measured" is falsifiable; an invented figure is not, and inventing one is the exact failure #294 was reopened for.

No new guard ships here — on purpose

rivet already enforces status-allowed-values at ERROR severity and rivet validate is already a required check, so the regression is mechanically blocked the moment this lands. A bespoke checker would add a weaker second oracle beside a working one.

This is the opposite of #370 / #373, which needed tools/check_release_plane.py precisely because rivet was blind to that defect.

Verification — three planes that agree

1. The migration proves its own edit. Per file it asserts the artifact-id set is unchanged, every record is byte-identical outside status, and the text-rewrite count equals the semantic-move count. That last assertion is the load-bearing one: "N lines rewritten" is equally consistent with having rewritten N prose lines inside description: blocks while leaving N real statuses alone. Block scalars are tracked and skipped rather than sed'd through.
2. The oracle flips. rivet validate → exit 0, 0 error-severity diagnostics under common@0.3.0 (embedded), down from 276. The diff is 276+/276- on the status lines — a strict 1:1 swap, no structural change.
3. The gate is live, not vacuous. Re-introducing one off-vocabulary status reproduces exactly one status-allowed-values error; restoring it returns exit 0 with a byte-identical file. Distinct inputs → distinct verdicts. This is the property Certify AADL↔WIT↔canonical-ABI byte-layout equivalence with ordeal (close the #319 layout gap) #327's withdrawn certificate did not have, so I no longer report an exit-0 without it.

Reconciliation: why 303 raw hits but 276 errors

	count	why
raw status: text matches, 9 committed files	303
− safety/requirements.yaml	−21	rivet never loads it — it is under no path in rivet.yaml's sources:
− safety/stpa/architecture.yaml	−6	nested in the fields: extension bag, where rivet cannot see them
= rivet's error count	276	✅ exact

Both remainders are left in place and named in REQ-GUARD-STATUS-VOCAB-001 rather than silently absorbed into the total. The second one is the #370 defect one field over — status: instead of release: — and tools/check_release_plane.py from #373 would catch it by changing a single module-level constant (FIELD = "release"). That generalization is a follow-up, not this PR: doing it here would stack this branch three-deep and break the "vocabulary only" scope I posted to #371.

Deliberately NOT in scope

Re-typing the 198 TEST-*/VAL-* measures to unit-verification and re-pointing satisfies → verifies. As measured on #371, those are ASPICE types requiring sw-detail-design/sw-req, of which spar has zero across 883 artifacts — adopting them would drag in SWE.1/SWE.2/SWE.3 and take feature-stage coverage from 194 requirements to 22. Separately scoped; the root friction is filed as pulseengine/rivet#748.

🤖 Generated with Claude Code

[avrabe]

Correction to the PR body: the base is now main, not fix/360-rivet-source-dedup.

I opened this stacked on #368 for a clean diff. That was wrong, and wrong in a way worth recording because it is the exact defect class #372 is about.

ci.yml triggers on pull_request: branches: [main]. Based on another PR's branch, no workflow triggered at all — total_count: 0, zero runs. The PR presented as clean; tot=0 pend=0 bad{none} is indistinguishable from "all green" in every surface I check. I nearly reported this as awaiting-green when it was awaiting-nothing.

Retargeting to main did not fix it either: GitHub's default pull_request types are opened, synchronize, reopened, and changing a base is edited. It took a close/reopen to fire the event. Now at 7 check-runs across CI, Lean Proofs and Verification Gate.

Consequence for review: until #368 merges, this PR shows 3 commits — #368's two plus this one — because that is genuinely the diff against main. Once #368 lands, GitHub recomputes and only a3512eb remains. The rivet validate exit-0 result holds either way; it was measured on top of #368's tree, which is what this branch contains.

Written up in #372 as an argument that the proposed standing check should also assert every open PR has a non-zero required-context count — no amount of auditing ci.yml catches a PR that never triggered it.

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!