docs: record Gate 3 conditional evidence

This commit is contained in:
Joseph Magly
2026-08-16 09:34:24 -04:00
parent fa233fd8c9
commit 42b30f7e5b
4 changed files with 152 additions and 44 deletions
+17 -11
View File
@@ -2,9 +2,11 @@
Date: 2026-08-16
Exact base: `9683e0be4d892e6a6a134b45a065e591e139da3c`
Exact implementation head: `acc6b3b254e6d358ddabe495b1ca79295dd36d70`
Status: PASS locally after the portability correction; hosted exact-head rerun
and canonical merge remain pending
Exact core implementation head: `acc6b3b254e6d358ddabe495b1ca79295dd36d70`
Exact canonical head: `b961623513a0e137b959f608ce31511d59e85888`
Canonical run: https://github.com/elder-plinius/OBLITERATUS/actions/runs/31947837466
Status: PASS; PR #109 was exact-head audited, signature-preserving merged, and
canonically reconciled
## Results
@@ -27,9 +29,11 @@ and canonical merge remain pending
| Package | Wheel and sdist built at version 0.1.2; both pass Twine metadata validation |
| Policy | Ruff, immutable quality policy, conditional policy, risk map, mutation-target guard, global/mature coverage, and changed-line gates passed |
The item-7 candidate reaches every quantitative Gate 3 repository and mature
CPU exit criterion. Ordinary feature work remains held until this increment is
merged and the conditional-evidence and final-reporting items are reconciled.
The canonical item-7 head reaches every quantitative Gate 3 repository and
mature CPU exit criterion. Its post-merge run passed all eight required jobs;
the quality-depth lane completed the repeat and 2,057-mutant campaign in
28m26s. Ordinary feature work remains held until the conditional-evidence and
final-reporting items are reconciled.
## Contracts added
@@ -105,6 +109,8 @@ policy at 94.04% statements and 84.61% branches.
(wheel) and
`f487391c98e4931896238b38b5232fb2f7c08147ec90767328734bf04c5c5c69`
(sdist)
- Canonical reconciliation comment:
https://github.com/elder-plinius/OBLITERATUS/pull/109#issuecomment-5307612606
The first isolated sync selected unsupported Python 3.13 and was not used as
evidence. A separate reused Python 3.12 environment correctly failed two
@@ -126,8 +132,8 @@ canonical commits.
This increment uses a synthetic random-initialized offline model and isolated
temporary filesystems. It makes no network, external-model, accelerator,
credential, or publication claim. The exact-head PR must still pass all hosted
CPU, repeat, mutation, package, Windows, and supply-chain gates before a
signature-preserving merge. Conditional evidence reconciliation and final Gate
3 reporting remain open after merge, so ordinary feature work does not resume
yet.
credential, or publication claim. The exact-head and canonical runs passed all
hosted CPU, repeat, mutation, package, Windows, and supply-chain gates, and the
signed commits were preserved on `main`. Conditional evidence reconciliation
and final Gate 3 reporting remain open, so ordinary feature work does not
resume yet.
+99
View File
@@ -0,0 +1,99 @@
# Gate 3 increment 8 conditional-evidence and waiver report
Date: 2026-08-16
Exact base: `b961623513a0e137b959f608ce31511d59e85888`
Exact policy implementation head: `fa233fd8c97d2a1463141535f2a288b264abe93b`
Status: PASS locally, on exact-head software infrastructure, and on Titan/Mutsu
operator hardware; hosted PR verification and canonical merge remain pending
## Results
| Gate | Result |
|---|---|
| Focused policy contracts | 17 conditional-policy, summary, CUDA-overlay, and evidence-freshness tests passed |
| Policy regression set | 110 policy and CI-contract tests passed |
| Mandatory CPU lane | Python 3.12: 2,122 passed and 6 environment-dependent tests skipped in 81.44s; repository coverage 83.51% statements |
| Policy validation | Committed conditional policy passes direct CLI validation on 2026-08-16 |
| Software conditional gates | Exact-head run 31949881557 passed pinned-model runtime, external evaluation, loopback network, operator UI, policy, and final summary |
| CUDA and bitsandbytes | Titan RTX 4090: Torch 2.13.0+cu130 / CUDA 13.0; both gate invocations passed 2/2 at exact head; GitHub runner registration remains waived through 2026-09-15 under issue #110 |
| Apple MPS | Mutsu Apple M4 / 16 GB: native Torch 2.13.0 reports MPS built and available; gate passed 1/1 at exact head; GitHub runner registration remains waived through 2026-09-15 |
| Apple MLX | Mutsu Apple M4 / 16 GB: MLX 0.32.0 and MLX-LM 0.31.3; gate passed 1/1 at exact head; GitHub runner registration remains waived through 2026-09-15 |
| Remote execution | Required repository variables and secrets are unset; waived through 2026-09-15 under issue #110; no support, compatibility, correctness, or performance claim |
| Runner inventory | Titan exposes Gitea GPU/host runners only; Mutsu exposes a Colima builder but no runner process; the GitHub token also receives HTTP 403 from the repository-runner inventory API |
| Signing | Policy commit is signed and verified with publish-key fingerprint `62297562B1C7053088F405DB0117DAAA677A5BF2` |
## Contracts added
1. Every unavailable conditional lane has one
versioned waiver naming its gate, canonical issue, reason, opening date,
expiry date, and the exact claim that is blocked.
2. A waiver may last at most 30 days. Missing fields, non-object records,
duplicate gates, unknown gates, software-only gates, invalid issue URLs,
invalid or future dates, reversed dates, overlong lifetimes, and expired
records fail policy validation.
3. An unselected gate with an active waiver is emitted as
`waived_no_support_claim` with its tracker, expiry, and blocked claim.
Unselected gates without a waiver remain
`not_selected_no_fresh_evidence`.
4. Selecting and successfully executing a waived gate produces `success` and
therefore uses real evidence instead of the waiver. Selected failures remain
failures, including shared CUDA/bitsandbytes job failures.
5. An expired or malformed waiver also fails the final summary, preventing a
stale record from becoming a silent green result after the policy job.
6. Mandatory Linux CI remains CPU-only. The selected CUDA lane derives the
exact locked Torch base version, replaces only Torch with the same-version
official `cu130` build, asserts that the resulting build exposes CUDA, and
runs `uv pip check` before the hardware probe.
## Availability findings and claim boundary
`gh variable list` and `gh secret list` for the canonical repository returned
no configured entries on 2026-08-16. The runner inventory endpoint returned
HTTP 403 because the current maintainer token lacks repository-runner read
permission. Live read-only inspection then confirmed that Titan has an RTX
4090 and active Gitea runners, while Mutsu is a 16 GB Apple M4 builder. Neither
host exposes a GitHub Actions runner process, so the waivers now cover GitHub
runner registration and scheduled automation rather than hardware absence.
Equivalent operator runs used isolated temporary directories, checksum-verified
uv 0.12.4 binaries, a managed Python 3.12.13 runtime, the exact signed source
archive, and the committed dependency lock. Titan installed the official
same-version CUDA overlay and passed CUDA/bitsandbytes. Mutsu installed the
native macOS lock plus the MLX group and passed MPS/MLX. The evidence was copied
off-host and hashed before both remote temporary directories were removed.
The waivers are tracked by
https://github.com/elder-plinius/OBLITERATUS/issues/110. They expire on
2026-09-15, 30 days after opening. Before expiry, maintainers must either
configure and run the matching gate or review and replace the waiver under the
tracker. The fresh probes establish only the recorded operations on Titan and
Mutsu at the exact candidate. They do not create a broad support, compatibility,
correctness, or performance claim, and they do not replace the open work to
register maintainable GitHub runners.
## Verification and evidence
- Exact-head software conditional run:
https://github.com/elder-plinius/OBLITERATUS/actions/runs/31949881557
- Downloaded software JSON/JUnit evidence:
`/tmp/obliteratus-item8-software-final-GMGS0x`
- Exact-head hardware JSON/JUnit and SHA-256 evidence:
`/tmp/obliteratus-item8-hardware-evidence-GqqBSm`
- Focused local command:
`/tmp/obliteratus-item8-py312-env/bin/python -m pytest -q --no-cov
tests/test_conditional_gate_scripts.py
tests/test_conditional_evidence_freshness.py`
- Mandatory local environment: `/tmp/obliteratus-item8-py312-env`
- Local generated summary: `/tmp/obliteratus-item8-summary.json`
- Waiver tracker:
https://github.com/elder-plinius/OBLITERATUS/issues/110
- Hardware-evidence tracker comment:
https://github.com/elder-plinius/OBLITERATUS/issues/110#issuecomment-5307675226
## Scope and release decision
This increment changes conditional-test governance and CUDA environment setup.
It does not alter production behavior or weaken mandatory CPU, package,
coverage, repeat, mutation, Windows, or supply-chain gates. Gate 3 item 9 and
the final exact-canonical report remain required before the contributor merge
train resumes.
+30 -33
View File
@@ -21,35 +21,32 @@ Gate 3 increment 4 canonical commit:
Gate 3 increment 5 canonical commit:
`aa182cc44883890c019c393f17602e2e2702e7d7` (PR #105, rebase merged)
Gate 3 increment 6 canonical commit:
`9683e0be4d892e6a6a134b45a065e591e139da3c` (PR #108, signed fast-forward)
Gate 3 increment 7 canonical commit:
`b961623513a0e137b959f608ce31511d59e85888` (PR #109, signed fast-forward)
Current canonical AIWG and CI-contract baseline:
`e8ac3b65670d696ed5b68f06adf14706bc2ff865` (PR #107, rebase merged)
The latest canonical mandatory offline CPU lane passes 1,985 tests on Python
The latest canonical mandatory offline CPU lane passes 2,112 tests on Python
3.12 with 9 conditional tests deselected by policy, no failures, errors, skips,
or unexpected warnings, and 78.57% statement / 65.89% branch coverage. Mature
CPU-testable coverage is 93.09% / 82.89%. The canonical mutation campaign kills
1,807 of 2,020 mutants (89.46%), with 208 survivors and 5 timeouts, while 595
or unexpected warnings, and 83.50% statement / 71.08% branch coverage. Mature
CPU-testable coverage is 94.02% / 84.54%. The canonical mutation campaign kills
1,844 of 2,057 mutants (89.65%), with 208 survivors and 5 timeouts, while 702
tests pass in each of three file orders and hash seeds with no flakes. Package,
Windows checkpoint, supply-chain, and all other hosted jobs passed on attempt 1:
https://github.com/elder-plinius/OBLITERATUS/actions/runs/31937799938.
https://github.com/elder-plinius/OBLITERATUS/actions/runs/31947837466.
Gate 3 increment 6 is implemented and locally verified at exact core evidence
head `77adf119f871ba820f1f02d55fe2760aa24c84b2`. Python 3.12 passes 2,057
mandatory tests with 2 expected source-shadow xfails and 9 conditional tests
deselected in 56.15s; coverage is 82.78% statements / 69.62% branches; mature
CPU coverage is 93.27% / 83.06%; 62/62 changed executable lines and all five
touched modules pass. All 9 new checkpoint-parser mutants are killed, the full
campaign expands to 2,029 mutants for hosted verification, and 669 tests pass
in each repeat order with no flakes. Exact-head PR audit, hosted matrix, and
canonical merge remain pending; details are in `gate3-increment-6-report.md`.
The latest conditional run is green for pinned-model download and evaluation,
loopback network, operator UI, and evidence-freshness policy at the earlier
Gate 2 commit:
https://github.com/elder-plinius/OBLITERATUS/actions/runs/31868394039. CUDA,
bitsandbytes, MPS, MLX, and remote execution were not selected and therefore
have no fresh support evidence or support claim. Gate 3 item 8 must reconcile
all conditional evidence at the eventual release candidate.
Gate 3 increment 8 is implemented at signed policy head
`fa233fd8c97d2a1463141535f2a288b264abe93b`. Fresh pinned-model download,
external-evaluation, loopback-network, and operator-UI evidence passes at that
exact head in run 31949881557. Titan CUDA/bitsandbytes and Mutsu MPS/MLX operator
probes also pass at that exact head. GitHub runner registration and remote
execution remain time-bounded no-support/no-correctness/no-performance waivers
through 2026-09-15 under issue #110. Details are in
`gate3-increment-8-report.md`.
The foundation, Wave A, Gate 1, and Gate 2 establish a sane,
evidence-producing baseline. The legacy contributor merge train proceeded one
@@ -58,9 +55,8 @@ property, mutation, determinism, and real-environment evidence without weakening
the mandatory CPU baseline.
The operator has now placed ordinary feature work back behind a testing-depth
gate. Gate 3 is adopted and items 15 are canonical; item 6 is locally complete
but not yet canonical; item 7 and the remaining quantitative exit criteria must
pass before unrelated feature work resumes.
gate. Gate 3 is adopted and items 17 are canonical; conditional reconciliation
and the final canonical report must pass before unrelated feature work resumes.
Correctness, security, data-loss, and test-infrastructure repairs remain
permitted when narrowly scoped and paired with reproducing tests.
@@ -399,14 +395,15 @@ single repository-wide coverage PR.
decision-seam contracts. Canonical run 31941334488 passed all eight required
checks; exact evidence is recorded in `gate3-increment-6-report.md` and the
canonical PR reconciliation comment.
7. **Implemented locally at `acc6b3b` against `9683e0b`.** Extend the installed
tiny-model vertical slice and quantized-storage semantics, including the
exact Float-to-integer restoration regression. The candidate reaches 94.02%
mature CPU statements and 84.54% branches and raises the immutable floors
to 94% / 84%; signed commit, exact-head audit, hosted checks, and canonical
merge remain pending.
8. Reconcile conditional software evidence, then run or explicitly waive each
unavailable hardware/remote environment under the claim policy.
7. **Completed at `b961623` (PR #109).** Extend the installed tiny-model
vertical slice and quantized-storage semantics, including the exact
Float-to-integer restoration regression. The canonical head reaches 94.02%
mature CPU statements and 84.54% branches, raises the immutable floors to
94% / 84%, and passes all eight hosted checks.
8. **Implemented at signed policy head `fa233fd`.** Reconcile fresh software
evidence and explicitly waive each unavailable hardware/remote environment
under the claim policy. Exact-head conditional evidence, PR audit, and
canonical merge remain pending.
9. Publish the Gate 3 execution, coverage, mutation, repeat, duration, and
conditional-evidence reports at one canonical `main` commit.
+6
View File
@@ -21,6 +21,12 @@ and expire on 2026-09-15; they are not evidence that any waived backend works.
They block support, compatibility, correctness, and performance claims for the
waived environment.
The current accelerator waivers cover missing GitHub runner registration, not
missing hardware. Titan has an RTX 4090 behind Gitea runners, and Mutsu is a 16 GB
Apple M4 builder. Exact-head operator probes may be attached to issue #110, but
they do not make the scheduled GitHub lanes runnable or establish a broad backend
support claim.
## Hosted gates
The model gate downloads only