verify(trace): promote 105 verification artifacts from EXECUTED gate evidence (TRACE-P02)

[avrabe]

The right side of the V was almost entirely unpromoted — 22 of 247 requirements verified, with the evidence already existing and simply unbound. This binds the part a gate run can actually justify.

Method: executed evidence, not a verifies link

A full sweep ran 323 steps, 0 FAIL. An artifact qualifies only if it had ≥1 step that genuinely executed and none that failed. Skipped steps don't count — bench-only, enforced-by-kani-gate and skip-no-steps are all exclusions, because "the gate did not run it" is not evidence that it passes.

artifacts in the sweep	162
≥1 executed step, zero failures	130 → promoted
only skipped steps	32 → left unpromoted, deliberately
any FAIL	0

Net 105 promoted here (25 were already verified). Verification artifacts now 130 verified / 123 approved / 34 implemented.

A third category TRACE-P02 didn't anticipate — found by doing it

The mechanical sweep initially promoted six artifacts whose own text says they must not be verified:

artifact	why it must stay implemented
FV-FALCON-OCI-002	"stays implemented until the fuse is run"
FV-FALCON-OCI-003	wasm.directory criteria (b) and (c) not met
FV-FALCON-OCI-004	kill-criterion owned by gale/meld, unrun
FV-FALCON-OCI-005	ditto
FV-FALCON-PERF-002	criterion (d), the PX4 head-to-head, not done (#362)
FV-FALCON-REL-001	failure injection not run

All six have passing executed steps. The gate result is real. It is also not sufficient, because each names a criterion the gate cannot check — an external kill-criterion, a live-page fetch, a comparison nobody has run.

TRACE-P02 anticipated two populations: gate-runnable (promote) and bench-only (waive). This is a third: gate-passing and self-blocked. Promoting them would have asserted exactly the outcomes those criteria exist to test — the error v1.133 avoided, and which jess's meld run then vindicated by rejecting.

All six reverted. The rule this establishes: an artifact naming an undischarged criterion is excluded from mechanical promotion no matter how green its steps are. Gate evidence is necessary, never sufficient.

Deliberately not promoted

• 32 artifacts whose steps only skipped (bench-only / kani-enforced)
• the 6 self-blocked above
• 45 artifacts with no steps at all — cannot be evidence as they stand; they need steps written or an honest demotion (this is where the Verus finding Verus is the unenforced proof track: 19 verus_test targets defined, ZERO run in CI — and 16 artifacts cite target names that don't exist #364 came from)
• requirements themselves — this promotes verifiers; requirement promotion follows once their verifiers are all verified

Code-free, per the two-commit rule. rivet validate PASS, exit 0.

🤖 Generated with Claude Code

[temper-pulseengine]

Automated review for PR #365

pulseengine/relay:verify/trace-p02-promote-executed → pulseengine/relay:main

Verdict: 💬 Comment

Summary: The code snippet provided appears to be a list of verification artifacts for a software project, specifically related to the FALCON (Flight Automation and Control Library) HAL (Hardware Abstraction Layer). Each artifact has an ID, type, title, status, and description. The status of each artifact is updated from 'approved' to 'verified'. This suggests that the verification process has been complete

Findings: 0 mechanical (rivet) · 1 from local AI model.

Findings (1):

1. artifacts/verification/FV-FALCON-NID-002.yaml:14

   status: approved
   

   The status of each artifact is updated from 'approved' to 'verified'. This suggests that the verification process has been complete.

---

Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location.

Reviewed at 295652a