diff --git a/artifacts/findings.yaml b/artifacts/findings.yaml index ecc6c5a..e4ec923 100644 --- a/artifacts/findings.yaml +++ b/artifacts/findings.yaml @@ -241,7 +241,7 @@ artifacts: title: rivet 0.16.0 canonical status enum is breaking - no migration, drops 'accepted', misfits ai-found-defect status: open description: > - Adopting rivet 0.16.0 flips the jess store from PASS (4 warnings) on 0.15.0 to FAIL (21 errors) with no artifact changes - all 21 are the new canonical status enum (PR #419 / #352): status values accepted (x12), open (x7), resolved (x2) are no longer allowed; the enum is [draft, proposed, approved, implemented, verified, released, deprecated, rejected]. Three friction points: (1) breaking with no migration path - a previously-valid store needs 21 manual edits; (2) 'accepted' was the documented terminal lifecycle state (draft->...->verified->accepted in rivet's own release/agent guidance) and is now gone; (3) one global document-lifecycle enum doesn't fit ai-found-defect, whose natural vocabulary is open/triaged/resolved/wontfix. Suggested: ship a status remap (rivet migrate / validate --fix) + documented mapping, reconcile 'accepted', and allow per-schema status enums. jess stays on rivet 0.15.0 for now to keep the store valid and defect semantics correct. Filed https://github.com/pulseengine/rivet/issues/522 - supervising. UPDATE 2026-06-10 (iter#4): maintainer triaged and confirmed all three points as independently actionable; at their request jess added an Acceptance-Criteria block to the issue for slice 2 (re-add 'accepted' to the canonical enum - clears 12 of 21 errors, one-line schema change), with slices 1 (rivet migrate / validate --fix) and 3 (per-schema defect status enum) scoped out for separate PRs. jess stays on 0.15.0 until at least slice 2 ships. UPDATE 2026-06-10 (iter#6): slice-2 draft PR rivet#525 verified PRE-MERGE against the jess store - built pr525 head 117e9c7 and ran validate: accepted errors 13 -> 0, remaining 13 are exactly the out-of-scope slice-3 vocabulary (open x7, resolved x6 on ai-found-defect). Posted the verification table on rivet#522 (issuecomment-4673352099); PR is parked as draft only on the pulseengine.eu blog 503 process gate. jess upgrades off 0.15.0 once slice 2 ships in a release. UPDATE 2026-06-14 (iter#19): rivet#525 MERGED to main and rivet#522 CLOSED - but the merge (13:13) postdates the latest release v0.16.1 (08:20), so 'accepted' is fixed on main but NOT YET in a release. jess holds at 0.15.0 and adopts the first rivet release that carries #525; will re-run validate on that release to confirm accepted 13->0 before upgrading. UPDATE 2026-06-20 (iter#41): rivet v0.17.0 RELEASED carrying #525 (slice 2). Re-verified against the jess store on the v0.17.0 darwin release: accepted errors 0 (slice 2 confirmed shipped), remaining = EXACTLY 27 status-enum errors, ALL on ai-found-defect: 14 open + 13 resolved (slice 3, the per-schema defect status enum, NOT yet shipped); no non-status errors. Bump gated solely on slice 3. Rather than migrate 52 defect artifacts to an ill-fitting canonical mapping, jess filed the focused slice-3 ask https://github.com/pulseengine/rivet/issues/550 (per-schema status enum for ai-found-defect, or a documented defect mapping + rivet migrate). jess stays on 0.15.0 and bumps to v0.17.0 the moment defect statuses validate; offered pre-merge branch testing as for #525. + Adopting rivet 0.16.0 flips the jess store from PASS (4 warnings) on 0.15.0 to FAIL (21 errors) with no artifact changes - all 21 are the new canonical status enum (PR #419 / #352): status values accepted (x12), open (x7), resolved (x2) are no longer allowed; the enum is [draft, proposed, approved, implemented, verified, released, deprecated, rejected]. Three friction points: (1) breaking with no migration path - a previously-valid store needs 21 manual edits; (2) 'accepted' was the documented terminal lifecycle state (draft->...->verified->accepted in rivet's own release/agent guidance) and is now gone; (3) one global document-lifecycle enum doesn't fit ai-found-defect, whose natural vocabulary is open/triaged/resolved/wontfix. Suggested: ship a status remap (rivet migrate / validate --fix) + documented mapping, reconcile 'accepted', and allow per-schema status enums. jess stays on rivet 0.15.0 for now to keep the store valid and defect semantics correct. Filed https://github.com/pulseengine/rivet/issues/522 - supervising. UPDATE 2026-06-10 (iter#4): maintainer triaged and confirmed all three points as independently actionable; at their request jess added an Acceptance-Criteria block to the issue for slice 2 (re-add 'accepted' to the canonical enum - clears 12 of 21 errors, one-line schema change), with slices 1 (rivet migrate / validate --fix) and 3 (per-schema defect status enum) scoped out for separate PRs. jess stays on 0.15.0 until at least slice 2 ships. UPDATE 2026-06-10 (iter#6): slice-2 draft PR rivet#525 verified PRE-MERGE against the jess store - built pr525 head 117e9c7 and ran validate: accepted errors 13 -> 0, remaining 13 are exactly the out-of-scope slice-3 vocabulary (open x7, resolved x6 on ai-found-defect). Posted the verification table on rivet#522 (issuecomment-4673352099); PR is parked as draft only on the pulseengine.eu blog 503 process gate. jess upgrades off 0.15.0 once slice 2 ships in a release. UPDATE 2026-06-14 (iter#19): rivet#525 MERGED to main and rivet#522 CLOSED - but the merge (13:13) postdates the latest release v0.16.1 (08:20), so 'accepted' is fixed on main but NOT YET in a release. jess holds at 0.15.0 and adopts the first rivet release that carries #525; will re-run validate on that release to confirm accepted 13->0 before upgrading. UPDATE 2026-06-20 (iter#41): rivet v0.17.0 RELEASED carrying #525 (slice 2). Re-verified against the jess store on the v0.17.0 darwin release: accepted errors 0 (slice 2 confirmed shipped), remaining = EXACTLY 27 status-enum errors, ALL on ai-found-defect: 14 open + 13 resolved (slice 3, the per-schema defect status enum, NOT yet shipped); no non-status errors. Bump gated solely on slice 3. Rather than migrate 52 defect artifacts to an ill-fitting canonical mapping, jess filed the focused slice-3 ask https://github.com/pulseengine/rivet/issues/550 (per-schema status enum for ai-found-defect, or a documented defect mapping + rivet migrate). jess stays on 0.15.0 and bumps to v0.17.0 the moment defect statuses validate; offered pre-merge branch testing as for #525. UPDATE 2026-06-23 (cron loop): rivet#550 is now CLOSED (2026-06-23T08:21) - the slice-3 per-schema ai-found-defect status enum is resolved on rivet main. BUT the latest rivet RELEASE is still v0.17.0 (2026-06-19), which PREDATES the close, so no release carries the slice-3 fix yet; jess CI stays pinned RIVET_VERSION v0.15.0 and validate PASSES there (confirmed this loop with a freshly-bootstrapped rivet 0.15.0 darwin binary). CAVEAT discovered: a local rivet that had DRIFTED to 0.17.0 (build d648ce44, main 2026-06-23) still REJECTS open/resolved on ai-found-defect (28 errors, allowed = the generic [draft..accepted] enum) - i.e. even the same-day main build does not yet expose [open/triaged/resolved] for the defect schema, so the #550 resolution likely needs a schema opt-in OR my local build predates the merge. NET: bump off 0.15.0 still blocked until (a) a rivet RELEASE past the #550 fix ships AND (b) jess re-verifies open/resolved validate on it (offer pre-release verification, as for #525). OPERATIONAL: validate jess with the CI-PINNED rivet 0.15.0, NOT a drifted local 0.17.0 (the latter gives false FAILs on the pre-existing defect statuses). tags: [release-watch, rivet, schema, breaking] fields: detected-by: jess loop iter#3 (2026-06-10) - rivet 0.16.0 validate on the jess store @@ -796,6 +796,13 @@ artifacts: the RC-input decode series (SBUS v1.88 + CRSF v1.89, both -> bounded RcInput). Per-piece GREEN on synth v0.12.0, authoritative skips 3/18 all #369 unchanged, bulk-mem clean - no new unlowerable op. Evidence: results/jess-build-falcon-v1.89.0.xml. + UPDATE 2026-06-23 (cron, falcon-v1.92.0 = DroneCAN foundation, DC-P01, relay #226): + v1.90-v1.92 series; v1.92 = "DroneCAN foundation - spar model + CanBus seam + + verified transfer layer" (the relay-side start of the CAN path, REQ-PIX-006). Tested + on synth v0.12.0: per-piece GREEN, authoritative skips 3/18 all #369 unchanged, + bulk-mem clean - the CanBus seam + transfer layer add no new unlowerable op. Also + meld v0.34.0 (resource_graph #295 / LS-A-17 hardening) - jess forward-chain fuse + stayed green. Evidence: results/jess-build-falcon-v1.92.0.xml. tags: [release-watch, synth, fpu, miscompile, correctness, on-target, blocker] fields: detected-by: jess REQ-PIX-001 value-level oracle - silent miscompile resolved in v0.11.46 (GI-FPU-001, verified loud-skip exit 1); OPEN remainder is GI-FPU-002 hard-float diff --git a/artifacts/phase2-pixhawk.yaml b/artifacts/phase2-pixhawk.yaml index 38e4da7..e71282b 100644 --- a/artifacts/phase2-pixhawk.yaml +++ b/artifacts/phase2-pixhawk.yaml @@ -139,10 +139,15 @@ artifacts: type: requirement title: DroneCAN/CAN-FD peripheral + smart-actuator path on the 6X-RT CAN buses status: draft - fields: - category: functional - priority: could - release: v0.8.0 + description: > + jess binds the CAN path to the RT1176 silicon (the 3x CAN-FD on the 6X-RT, + 2 broken out). RELAY-SIDE FOUNDATION LANDED 2026-06-23: relay falcon-v1.92.0 + ("DroneCAN foundation - spar model + CanBus seam + verified transfer layer", + DC-P01 / relay #226) - the portable CanBus seam + verified transfer layer that + jess will bind, the CAN analogue of the relay-hal transport matrix (DD-018). + Per-piece GREEN on synth v0.12.0 (AFD-024). jess's binding (CAN controller + + transceiver over the RT1176 FlexCAN, via the embedded-hal-async-style seam) is + the on-silicon counterpart, sequenced after the core HAL binding (REQ-PIX-012). links: - type: traces-to target: FEATURE-002 diff --git a/results/jess-build-falcon-v1.92.0.xml b/results/jess-build-falcon-v1.92.0.xml new file mode 100644 index 0000000..697e74b --- /dev/null +++ b/results/jess-build-falcon-v1.92.0.xml @@ -0,0 +1,12 @@ + + + + + + + + + + + +