Skip to content

audit: 2026-09-17 05:30Z — second T-2 stop of the 0.68 train, three root causes, autopilot on #3411 #3414

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
noahgift merged 1 commit into from
Sep 17, 2026
Merged

audit: 2026-09-17 05:30Z — second T-2 stop of the 0.68 train, three root causes, autopilot on #3411 #3414

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions docs/audits/autonomous-run-2026-09-15.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -632,3 +632,18 @@ scheduling chance, and it costs ~3× (gx10 10.6 min vs intel 34.5 for step 1).
### Still open

- T-2 verdict → T-3 (tag, release, clean-room dispatched on `v0.68.0`, assets) → T-4 (cascade, install, host + installer receipts, close). Ledger record at close → docs PR.

## Interval — 05:30Z (2026-09-17) — second T-2 stop, three root causes

### What moved

- **T-2 NO-GO #2 on release commit `27f070324`** (03:15Z): three RED rows, none of them a PR-level required check —
1. `pv-contracts`: `contracts/external-corpora.yaml` (ONT-1 census declaration, #3281) is not a KernelContract; `pv validate` had no kind for it → **#3410** `ArtifactKind::ExternalCorpora`, one struct shared with `pv census`, 1841 contracts 1→0 failing; merged 05:13Z.
2. `pmat-comply` CB-200: 604 below B vs baseline 602 — measured real debt under the same pmat 3.40.2 (baseline commit reads 602); roster diff: `forward_qwen35.rs::from_model_and_layers` (C+, #3114) → A+ by extraction, plus two C++ `main`s of the vendored llama.cpp reference harness under `evidence/` → `evidence/**` excluded from `[tdg]` as reference harness, baseline **lowered** 602 → 601 with its paired `scripts/cb200_baseline.txt` → **#3411** (queued).
3. `model-parity` C14: `apr parity` is architecture-blind (dense loop) and after #3325 reaches MoE/qwen35 files — "EMPTY data buffer qtype=0" is the tool defect #3367 named, not a model defect → **#3412** pre-load refusal (exit 12, raw arch tag, never the normalizer's fold), C14 reports `UNMEASURED-TOOL (#3367 / #3090)`, crashes still FAIL; measured on lambda with a CUDA build: 3 PASS, 3 UNMEASURED-TOOL, 0 FAIL (queued).
- Autopilot relaunched `3411 wait close` (pid 2009080) — #3411 enters the queue last, so its merge commit is the release commit.
- Memory: release-phase-only gates (perf041 age, `pv validate` over all contracts, CB-200 pair, C14) must run on `origin/main` BEFORE the bump — 0.67 ×3 and 0.68 ×2 NO-GOs after the cut are the same shape.

### Still open

- #3412 → #3411 → T-1 → T-2 → T-3 (clean-room on the tag) → T-4 → close → ledger record.
Loading