Repository navigation
split(#4502 PR-2): ONT-10 rest — everything in #4502 outside the contracts layer (stacked on #4587) - #4588
split(#4502 PR-2): ONT-10 rest — everything in #4502 outside the contracts layer (stacked on #4587)#4588noahgift wants to merge 218 commits into
Conversation
…200 EV-7a, PMAT-4347, GH-4122 EV-5a Takes the aprender-contracts / -cli / -staging trees and the additive aprender-common cli_roles module from #4502's r7 A head 0be7c00, plus the cooperative-matrix delete (#4347) and the PVL ci/explicit-test-commands.d rows. Workspace versions stay at main's. Generated files are excluded (discharge-summary.json, Lean Axioms pin); 98 regenerates them as the single writer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…key) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ire); generated ontology files stay main's Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Everything in #4502's diff that PR-1 does not carry. PR-1 ∪ PR-2 = #4502, less the generated files 98 regenerates as the single writer: census.json, contracts.nt, shapes.ttl, ontology.ofn, tbox-report.json, discharge-summary.json, Lean Axioms.lean, and the README contract counts (kept at main's 1837). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…m from the delete pass) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…_prefill_whole_again, identical prompt 3x, reused 0, no restore Quorum at 0be7c00 (gemini): the PMAT-4445 AC names this test and 3 turns; the head had it as forget_prefix_drops_the_checkpoint_so_a_repeat_prefills_whole with 2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ames #4502 renamed ten shipped [[bin]]s to aprender-* but the ledger kept the old names, so check_binary_debt.sh read each rename as one NEW plus one STALE (10+10, #4502 defects row 4). Re-key the bin field only; class, target and the legacy_names counters are unchanged (a rename is not a merge/delete decision, and counters move only through the ratchet). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 50ab758)
… row 5, clippy -D warnings) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit afb58c0)
bashrs SEC001 flagged `eval "$*"` in the self-test's row() (#4502 defects row 10). row() now runs "$@" directly; the four mutations that need a redirect or two steps become named functions. All 15 rows keep their wanted rc and needle, so each RED row still proves its mutation bit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 8a003ed)
…w files — m52 refused 458 and 481 apr-format-golden-fixtures 458 -> 459 and aprender-facade-e2e 481 -> 483, both free ordinals, so the ont8-evidence-gate (458) and pvl-comparator (481) fragments #4502 added keep their slots and pvl-comparator still runs before pvl-challenge (482). Command content unchanged. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> (cherry picked from commit 1630322)
… builds --locked) L0 exception: fixes running job --locked (run 36388075483, mac-check job 108817628569). The build.rs env_key build-dep lines added provable-contracts edges without re-resolving the lock; cargo check --locked failed on the stale lock at b4ebf93. Edges only, no new packages. Receipt: intel cargo check --workspace --all-targets --locked rc=0, 0 errors. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… — elan 4 refuses to reinstall Row 13, run 36353187094 x86-main: step 1 failed in 0.1s with "error: 'leanprover/lean4:v4.29.0-rc4' is already installed". No judged file was missing. elan >= 4 (4.2.0 on the fleet) exits 1 on `toolchain install` of an installed toolchain, so the step passes once per host and fails every run after that. Install only when the toolchain is missing from `elan toolchain list` (first column, so a "(default)" suffix still matches), and decide that inside the same flock, so two jobs cannot both see it missing and race. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit de92271)
…(tree-reader drift)
workspace-test-shard step 7 (check_tree_reader_tests) failed in run 36388847584:
PR-1 adds aprender-contracts{,-cli} test targets without their registry rows.
Moves #4502's 12 contracts-cli .cmd wiring files into PR-1 instead of parking
11 targets in the unwired baseline (baseline unchanged), carries fb's m52
ordinal renames (457->459 golden-fixtures, 480->483 facade-e2e) that the moved
457/480 files need, and regenerates scripts/tree_reader_tests.txt.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Agent: aprender-88
…-debt re-key) Local run of the wired contracts-cli targets on PR-1 failed (ont4b_shapes_gate, shapes_vacuity, ont_entity_properties, ...: 'entity.type has no Σ entity_type_target_class mapping'): PR-1 carried #4502's contracts code but not its tests/fixtures/ont updates (164 files). Moves them in. Row 4: cherry-picked 50ab758 (contracts/binary-debt-v1.yaml re-key). (cherry picked from commit 50ab758) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent: aprender-88
roadmap aggregate, pv census, pv extract (contracts.nt + shapes.ttl) and the README CONTRACT_COUNT blocks, regenerated with the pv built from this tree; fixed points asserted: make roadmap-aggregate-check, pv extract --check, readme_sync.sh --check. Agent: aprender-88
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent: aprender-88 # Conflicts: # README.md
… file (ENOTDIR) — chmod 0o555 does not stop root, CI runs as root Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent: aprender-01 (cherry picked from commit 3e65448) Agent: aprender-88
… gate runs (#4502 row 2) lint_passes_on_real_contracts requires the refinement gate to RUN whenever CI is set; the gate compares HEAD with merge-base(HEAD, origin/main), else the origin/main tip. The workspace-test-shard section clones at depth 1 and only the tier step's pull_request arm fetched a main, so workflow_dispatch / merge_group / push shards had no base and shard 3 panicked "the refinement gate declined" (run 36384418038). Per the operator ruling (2026-09-28 09:41): pin the EVENT's base commit -- pull_request.base.sha, merge_group.base_sha, push's `before` -- as refs/remotes/origin/main; the origin/main tip only for workflow_dispatch. A missing or zero base fails the step with ::error (no silent skip). Placed after the tier step, whose pull_request arm would otherwise overwrite origin/main with main's tip. ci.yml is unchanged: it has no generator; fat_driver reads ci/sections.yml at run time. Agent: aprender-41 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit cd587d6a5b317467b4eac3b5c10b65ecf2656f8a) Agent: aprender-88
…ontracts tests read shard 1 of run 36391942814 failed ont2c_tracked_ofn_and_report_are_fresh: contracts/ontology.ofn and tbox-report.json are required-tracked writer outputs of contracts/ontology.yaml. PR-1's ontology.yaml is byte-identical to #4502 integ (f7858ca), so integ's outputs are the fresh ones; taken with the PVL discharge summary + Lean axioms the discharge-check target reads. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent: aprender-88
…e ordinals 761/762 (#4502 row 3) Shard1 step 13 REFUSEd two ordinals that each named two fragments: 458 (apr-format-golden-fixtures + contracts-cli ont8_evidence_gate) and 481 (facade-e2e + contracts-cli pvl_comparator). A shared ordinal leaves the run order ambiguous. The two contracts-cli fragments move to 761 and 762. Both commands are byte-identical; nothing else names these files. check_explicit_test_commands.sh --check: base origin/fb/4502-a-diag rc=1 (2 REFUSE), this commit rc=0 (101 fragments, no shared ordinal). Refs #4502 Agent: aprender-cb Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> (cherry picked from commit 59d6911a04a30b5f609fb67c11ca1aba32ad4381) Agent: aprender-88
…he lock
Diag run 36384418038 (x86-main): "P1 band-lock: the lock was still held
after release". flock(1) hands its lock fd to the holder bash, and the
holder's `sleep` inherits it. On TERM the holder killed the sleep and
exited without waiting for it. gpu_band_release waits only for the queue
process, so under load a sleep that was still dying kept the lock held
after release returned.
- gpu_band_lock.sh: the TERM trap now waits for the sleep. The sleep
starts before the ready file is written, so a TERM (sent only after
ready) always finds $s set.
- check_release_host_receipts.sh: P1 uses a sleep that takes 0.3 s to
die on TERM, like a loaded runner, so the row fails every time on the
old lib instead of only when the scheduler is unlucky.
Evidence: repro loop with the slow sleep, 10/10 held on the old lib and
0/10 with the fix (plain sleep: 0/5). Full self-test: rc=1 on the old
lib ("FAIL P1 band-lock: the lock was still held after release"); rc=0
with the fix (94 rows and mutants, band-lock-any-class killed).
Agent: aprender-6b
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit 7a8b939)
Agent: aprender-88
…usal-ok fixture follows the contract - pv_surface_gate: #4502 carries VS-COUNT-001 (#2648) in validator.rs with no row in the decision table, so every_rule_in_the_validator_source_appears_in_the_table failed. Add a Case: one obligation, stated total 2, Error. - ont_refusal_receipt: #4502 changed contracts/refusal-receipt-v1.yaml but not its byte-for-byte copy in tests/fixtures/ont/refusal-ok; copy it. Agent: aprender-fb Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 036beee) Agent: aprender-88
…ce is not vacuous; shapes_n 66; all 8 refusal fixtures - extract:binary (ONT-4g) refused every [workspace] root whose snapshot is absent, even one whose manifests declare zero bin targets. The ont fixtures (code-bound, ...) are such roots, so PV-ONT-012 turned the shapes gate red there (ont4b2_code_lean_w3c 343). Refuse only when the census names targets; new unit test for the 0-bin root, the 3-target absent-snapshot test stays red. - ont4b_shapes_gate (338): the later ONT-10 slices added nine binary shapes; name them, 57 -> 66 (measured by `pv extract contracts --check` at 036beee). - ont_refusal_receipt (430): the other seven refusal-* fixtures carry the contract too; copy it to all. Agent: aprender-fb Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 23122d3) Agent: aprender-88
…! names it, and it does not parse Carried from #4431 (42f84fb, GH-4197). #4502 had the deletion, but PR-2 (the #4502 split) did not, so the file was left without a carrier. It is the one file under crates/ that fails kani_assume's fail-closed token count (unbalanced } at :35). Agent: aprender-59 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 353e1d64e84f7872b14c5a67078d18dec062b785) Agent: aprender-88
|
quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-4502,PMAT-3715,PMAT-4431,PMAT-4073,PMAT-3712,PMAT-4445,PMAT-4083,PMAT-4189",
"head": "db80afca024f5abafe4e547e693d7faf1d54b330",
"width": 3,
"executor": "agy",
"agreed": false,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "PASS",
"findings": 1
},
{
"lane": 2,
"verdict": "PASS",
"findings": 1
},
{
"lane": 3,
"verdict": "FAIL",
"findings": 1
}
]
} |
|
quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-4502,PMAT-3715,PMAT-4431,PMAT-4073,PMAT-3712,PMAT-4445,PMAT-4083,PMAT-4189",
"head": "db80afca024f5abafe4e547e693d7faf1d54b330",
"width": 3,
"executor": "agy",
"agreed": false,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "PASS",
"findings": 1
},
{
"lane": 2,
"verdict": "PASS",
"findings": 4
},
{
"lane": 3,
"verdict": "FAIL",
"findings": 2
}
]
} |
… reviewed in #4647); PMAT-4445 title states the fix The quorum on this PR read the ci.yml mutants-debt ratchet block as an unauthorized gate change: no ticket in --ticket named it. It is the block #4647 adds and q-4647g agreed 3/3; operator rulings C188(a) and C192(1) put it in scope here. The PMAT-4502 entry now says so. PMAT-4445 title described the bug, and a review lane read it as a requirement twice (q-4588f/g). It now states the fix the diff makes. Local commit per C192(1); pushed once, with the merge of main, after #4647 merges. Agent: aprender-ca Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lone crate; claimed kernel with 0 focus nodes fails Agent: aprender-ca Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… 8 scalars expand to banned Result::unwrap Agent: aprender-ca Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ops.rs The F2 union of #4588 x #4620 lost the closing brace of the #[cfg(test)] thread_local! block (98 { vs 97 }); cargo fmt --check rc 1 -> 0. Every PR head is fmt-clean; this defect exists only in the rehearsal (C205(2)). Agent: aprender-ca Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s, K11 shapes.rs, K12 challenge.rs (cursor-free rewrites), K13 proof_status, K14 lint/extract K10/K11/K12 measured 14/14, 27/27, 50/50 caught; K13/K14 mutation runs were cut by the lambda reboot (NOT_MEASURED). Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…r that hunk only; A' mutants-table-scope kept Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…write failure Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…kills) Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> # Conflicts: # crates/aprender-contracts/src/lint/refines_gate_tests.rs # crates/aprender-contracts/src/lint/shapes_gate.rs
…ue (c213 m44: no evidence/ receipt exists for it) Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…c213 m27) Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sonnet seat (agent:claude-sonnet-5-5/pr-4587-review), base 430be65. Verdict DEGRADED: delta-only sample (32fc88e..HEAD) of a 784-file diff; duplication/agy/mutation not run on this head (not_measured). ci.yml merge hunk: comment-only, mutants-table-scope intact, no gate weakened. diff_patch_id c0ed4ceedd. Unsigned: the key is a CI secret. Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…derived; regenerated by pv census on intel) Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e regenerated census Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…0 uncited ratio reworded (c213 ADD) bashrs SEC010 flagged the two selftest mkdir -p "$d/..." lines (265/269); the subdir is now made by mk's optional 2nd arg. selftest 30/30, check_bashrs_gate 0. gpu_profile.rs:522 carried an uncited 27-34x ratio and seconds with no evidence/ receipt; reworded to point at #4590. check_perf_claims_cite_receipts 0. Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…po_graph_is_fresh) Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reviewer agent:claude-sonnet-5-5/pr-4588-review; verdict DEGRADED (sampled delta); diff_patch_id 4eb8d194f48bb285b252161dc2765f6ea7e5893f; unsigned (B1, key is a CI secret). Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…judged by the #4647 ratchet section The intel mutants-shard runs hit the 720m timeout on every past run, so the A' survivor table cannot finish. With no head refs listed, the mutants section is report-only and ci/mutants-debt.tsv stays shrink-only (#4647). Prepared LOCAL ONLY, pending the operator's choice. Also: scripts/tree_reader_tests.txt gains the derived row for aprender-review-experiment (build_corpus), the crate this PR adds -- CI run 36927004090 workspace-test-shard (3) drift; regenerated with check_tree_reader_tests.sh --update. Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rename release:ModelCell Re-ran the hotspot check that §4 requires before B2 and OBS-05, git only, against origin/main 316dee2. Trial merges used the old git merge-tree, with one positive and two negative controls. Every conflict in B2 (6 code files) and OBS-05 (ontology/shapes.rs) comes from one commit, 989cb01 (fold(0.70) #4655). Of the added lines in the ONT-10 PR heads, #4587/#4588/#4589/#4611 have 99-100% on main and #4615 has 93%; the not-on-main controls score 2.6% and 11.2%. So the "after ONT-10" condition is met in content. B2 duplicates nothing on main, but the class release:ModelCell now means two things. On main, v1 (since 989cb01) uses it for a non-generating model-command cell whose shape requires release:row. Seg 7's v2 uses it for a cell derived from kernel cells whose shape requires release:usesKernel. B2 renames its class before the PR. This is a static reading; no cargo was run (C277). Agent: la-0.71 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Closed with a pointer (0.70.2 PR backlog, C297 §5; aprender-6a for la-a). The content landed on main through fold #4655 (989cb01, '#4587+#4588 PMAT-4502'), and main has since refined it (the discharge timeout became Bounded::TimedOut, and the ladder-copy test moved to readme_contract.rs). Measured on origin/main 316dee2: merging main into this branch with conflicts resolved to main leaves only the generated contracts/census.json and one roadmap title line (PMAT-4445: 'prefills every timed turn whole'; the forget_prefix code is on main). Branch split/ont10-rest-0.70 is kept. |
…rex crate snapshot Step 3 of 3 (PMAT-4524). This commit restores the branch's content on the 28 paths that step 1 pre-set to main's bytes, merged with main's changes to them. Main's copy of crates/aprender-review-experiment is the branch's 066d43a (REX-04), landed through fold #4655 (#4587/#4588), plus #4588's mutant-killing tests. I 3-way merged each path on the branch commit closest to main's copy, then resolved the hunks that were left as a union. - crate: the branch's rows (C6, C11, C14, S1, ...) and all of main's tests. Two of main's tests were moved onto the branch API with their assertions unchanged. execution_problems became check_executed_fields (identical body). The prereg lock-header test now expects the re-locked rex-prereg-v2 header. - aprender-serve: generate() keeps main's cancel-before-decode fix (#4325), plus the branch's prompt reuse (#4445/#4274) and logprobs steps (#4026). qwen35_prompt carries main's request-tools change (#4650). - surface_audit.csv: main's rows. Every cited line is remapped through the branch's edits to commands_enum.rs, cuda_chat_backend.rs, openai_handlers.rs and router.rs (9/9 spot checks match main's source lines). - Regenerated files: roadmap.yaml by roadmap-aggregate (plus the PMAT-4524 entry), README counts by readme-sync, tree_reader_tests.txt by check_tree_reader_tests.sh --update. Cargo.lock is main's lockfile plus cargo metadata. Refs #4354, #4524 Pmat-Ticket: PMAT-4524 Agent: rex Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Split PR-2 of 2 of #4502 (operator plan: the split is a MOVE of #4502). Stacked on #4587 (base
split/pvl-rows-0.70). Once #4587 merges, this PR retargets to main.What: the full #4502 r7 A
0be7c00372tree minus PR-1, plus fb's integ fixes (rows 4, 7, 10, 13; row 8 superseded by PR-1's full regen) and the #4445 AC test (forget_prefix_makes_the_same_prompt_prefill_whole_again, 3 turns).Coverage: fb's union check PASS, PR-1 e13e9eb ∪ PR-2 = #4502 with 0 missing and 0 extra (handoff 4502-union-receipt-postcut).
Generated files: taken from PR-1's
batch_fold.sh --regen(censusn_files1889; README CONTRACT_COUNT 1889,check_readme_claims.shrc0).ontology.ofn,tbox-report.json,discharge-summary.jsonandAxioms.leanare not included (98 is the single writer).Pre-push receipts: intel at 78bd48d:
cargo fmt --all --checkrc0,cargo check --workspace --all-targets --lockedrc0. 66's guard-tree on intel: 59 PASS, and all 6 merge-ref guards rc0.Once #4587 and this PR merge, #4502 closes with links.
🤖 Generated with Claude Code