release(0.70.0): sealed car — REPLACES #4372 by noahgift · Pull Request #4429 · paiml/aprender

noahgift · 2026-09-25T14:22:55Z

REPLACES #4372. Operator ruling 16:14/16:15: the 0.70.0 sealed train car has fixed contents and no main-merges. It is cut fresh from main, and the scope is frozen.

Base: main a5efa14, so this PR is one commit ahead of main and not behind.
Contents: batch/0.70.0@596f06db2 (315 non-merge commits), reconciled with main. Main wins on 0.69.3-line code; the 0.70 additions are kept; roadmap.yaml is regenerated (make roadmap-aggregate-check ok).
Excluded per operator 16:20: the 2 commits whose patch-id is also in batch(0.69.5): EPIC #4249 patch train + #4313 f16 prefill — 0.69.5 RC #4318ff7f13e. They are d20f3a3 (ARB-AUD-10b: python3 out of the release path — 12 scripts / 43 sites + release_wall_time.py, jq with a parity table per script #4352) and 5f7dd19 (apr-cli --features wgpu does not compile: finetune.rs reaches entrenar's wgpu pipeline, which is gated behind entrenar's 'gpu' feature that apr-cli never enables #4056), reverse-applied here. After the exclusion, the car has 313 patch-ids, batch(0.69.5): EPIC #4249 patch train + #4313 f16 prefill — 0.69.5 RC #4318 has 62, and they share 0.
Local checks: fmt ok, explicit-test ordinals ok, cargo check rc 0 on apr-cli and aprender-train (--lib --tests), and on aprender-serve and apr-cli with --features cuda.

Tickets: #312 #315 #317 #328 #356 #385 #390 #420 #597 #1042 #1059 #1087 #2307 #2582 #2586 #2696 #2699 #2706 #2870 #3090 #3093 #3140 #3204 #3269 #3298 #3425 #3488 #3522 #3544 #3563 #3567 #3569 #3571 #3594 #3596 #3602 #3610 #3611 #3622 #3624 #3633 #3636 #3658 #3668 #3669 #3676 #3679 #3690 #3707 #3711 #3714 #3715 #3718 #3730 #3731 #3732 #3739 #3740 #3741 #3745 #3748 #3749 #3750 #3756 #3761 #3765 #3772 #3773 #3774 #3777 #3782 #3790 #3795 #3799 #3804 #3805 #3807 #3809 #3810 #3821 #3839 #3841 #3844 #3850 #3953 #3956 #3964 #3972 #3975 #3981 #4020 #4023 #4024 #4028 #4033 #4045 #4046 #4047 #4048 #4056 #4057 #4059 #4060 #4062 #4076 #4083 #4085 #4096 #4097 #4101 #4107 #4110 #4113 #4114 #4127 #4129 #4130 #4138 #4144 #4149 #4152 #4162 #4168 #4173 #4175 #4179 #4183 #4192 #4200 #4207 #4219 #4223 #4224 #4228 #4233 #4244 #4252 #4256 #4267 #4275 #4282 #4285 #4286 #4287 #4290 #4292 #4312 #4314 #4315 #4326 #4327 #4328 #4329 #4336 #4339 #4340 #4342 #4344 #4345 #4351 #4352 #4368 #4369 #4373

🤖 Generated with Claude Code

keep-open: the Tickets: line is a traceability index of every issue cited by the car's 313 commits, not a closing list. Each stays with its own owner/PR (APR-EPIC-001: only the named closers close); this PR closes only #4047 and #3972 (aprender-ea verified ONT-4c5 on the car, 11/11).
ont-delta: resolves ONT-4c5 (#4047 #3972, verified on car 93aa211)

Closes #4047
Closes #3972

github-actions · 2026-09-25T14:33:58Z

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=4429 head=0d7e7e49e4f469210c5f306d5a37ec5011c44409 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

…ratchet was cut guard-cargo's include_fmt_ratchet REFUSEd 17 include!d files (+2 test submodules rustfmt follows) that main gained since B2 and that are not in the baseline this PR introduces. Formatted rather than baselined: the baseline may only shrink. Refs #4427 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…th import, simulate needless borrow) Same env_key rewrite as aprender-rag. Local `cargo clippy --all-targets --keep-going -- -D warnings -A unused-variables` (ci / lint's command) now rc 0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…SIFY-BOOK-CLI-PARITY-001, 113/114) The batch folded the aprender-ptx-debug binary into `apr ptx-debug` without a chapter. Adds book/src/cli/ptx-debug.md + its SUMMARY entry; example-block and linkcheck gates pass locally (115/115 chapters). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ed: same-family, agy 429) Round 1 on aec0b67: sonnet PASS, gemini PASS, haiku FAIL (deprecation warnings removed from 4 main.rs files). Rebutted: those files are byte-identical to main and to the quorumed batch; the warnings belong to out-of-scope #4058. Round 2 on dd6afad: sonnet, haiku and a 2nd sonnet all PASS. The agy lane returned a 429 error and is not counted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…asure it — extract worst_normalised check_complexity_ratchet RED NEW: iq4_nl_device_ab_tests.rs::real_model_iq_ab cyclomatic 17 cognitive 28 after 71d0b2a reformatted the file. The normalised-error closure moves to a helper; pmat now reads 9/10 and 9/18. Behaviour unchanged (same loop, same NaN->inf and zero-denominator rules). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…y ratchet to 445 (#4430) contract_data_integrity reads falsifier ids in file order; 004 before 003 counted as a gap. tested: 4 now matches the four falsifiers. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…4056) apr-cli's `wgpu` feature enabled only `inference`, but the finetune wgpu path uses entrenar items gated behind entrenar's own `gpu` feature. `wgpu` now enables `entrenar?/gpu` (weak: only when `training` brings entrenar in, which is exactly when commands::finetune compiles). Compiling that surface for the first time exposed 10 clippy findings in aprender-train's gpu code, which no gate had ever linted: - 3 unnecessary parens, 1 redundant field name, 1 unused var, 1 needless mut - await_holding_refcell_ref in transformer/wgpu_block: the PipelineCache is now moved out of the RefCell for the await and put back before `?` - dead code (keep-alive GPU buffers, an unread eps, an uncalled dispatch_lora_addmm): #[allow(dead_code)] with a reason, like vram_guard Measured on lambda, base batch/0.70.0@5a70d9195: cargo clippy -p apr-cli --lib --features wgpu -- -D warnings rc=0 (was 5 E0432/E0433/E0425, then 10 lints) cargo clippy -p apr-cli --lib -- -D warnings rc=0 cargo test -p aprender-train --lib --features gpu --no-run rc=0 cargo fmt --all -- --check rc=0 Not run: the wgpu_block GPU tests (gpu-q queue 6 deep, timed out waiting). Follow-up for the fold: scripts/check_clippy_feature_matrix.sh (fold/4310-b4) excludes `wgpu=#4056`; once both land, drop the exclusion and add a wgpu axis. Refs #4056 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 5f7dd19) (cherry picked from commit b155a6d)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…f; regenerated from the kernel headers Fragment 490 wires sprint9_filtering_tests for the first time, and on the gx10 aarch64 runner three of them failed: `echo` never showed brk. It did call it. ld.so's first syscall, brk(NULL), printed as `recvmsg(0x0, 0x0, 0x0) = <heap address>`, because socket..mremap sat two numbers high (brk 216, not 214). 34 aarch64 rows were wrong, with sched_yield and getrandom each listed twice. The x86_64 table had setgid..setsid one high and pkey_mprotect at 302 (prlimit64). Both tables are now regenerated from asm-generic/unistd.h and x86_64 asm/unistd_64.h with the same name set, and they compile on every host. The new abi_numbers tests pin every corrected row against the kernel numbers. They require strictly ascending numbers and unique names, because binary_search depends on both. Run against the old tables, all three fail. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ling gate (stable 1.98.1) is clean chunks_exact(N) with a constant N becomes as_chunks::<N>(): fixed-size array chunks, so byte decodes read *b instead of [b[0], b[1], …]. float16_dot takes its remainders from as_chunks' second half. as_chunks is stable since 1.88; the declared MSRV is 1.91. make lint-current: RED (12) -> GREEN. aprender-serve quantize/float16/qwen35 unit tests: 2486/2486 before and after. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…46 + PMAT-4421) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…enerated by make roadmap-aggregate (union, check ok) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…ase 0005847; merge delta quorum 3/3 PASS) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

… CI signer)

…car #4429 — 5/5 leak on rc.1, 5/5 clean on 0.70.0 Refs #4344 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit e3bcd6b)

…are on no PR (#4083) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ms; rc_fleet_stage label no longer reads as a bare apr The car's kernels_generate_gemm_cuda.rs gained three indirect GDN arms (+13 lines above the fused arms), so contracts/kernel-fusion-v1.yaml still cited 371/374/377 and fusion_call_site_guard_3985 failed on workspace-test shard 3. They are 384/387/390, each a live call of its kernel. check_apr_bin_pinned.sh read the self-test label 'mutant (apr verify deleted)' at rc_fleet_stage.sh:302 as an unpinned apr invocation (guard-cargo RED). The label now says 'asset verify'; rc_fleet_stage --self-test PASS, pin guard OK (386 files). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

… CI signer)

… — setsid as PID 1 reaps an orphan and exits 1 #4427 wraps guard steps in `setsid --wait`. In the 4 steps that run it inside `docker run`, setsid is the container's PID 1: orphaned grandchildren (the nextest probe's cargo/rustc) reparent to it, its wait() returns one of THEIR pids instead of bash's, and util-linux errs "setsid: wait: Success", rc 1, before the guard prints a verdict. guard-cargo failed this way on gx10-build and yoga-build3 (runs 36182568927 a1/a2); main has no setsid, so it never did. Measured on gx10, image sha256:0569673d, same script + config as main: without --init 3/3 rc=1 'setsid: wait: Success' with --init 3/3 rc=0 'PASS: every key is understood.' without setsid rc=0 PASS --init makes tini PID 1, so orphans are reaped there, not by setsid. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ee m53) token_usage_report.py derives repo roots from SRC; provable-ladder ELAN_HOME defaults to $HOME/.elan via GITHUB_ENV; dogfood 0.68.2 evidence uses ~/models. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…achine path Path-only (operator go via cop 77): vars.APR_QWEN35_MODEL still wins; the fallback is $HOME/models/… and is exported for the cargo test. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

- include-fmt: rustfmt the 3 include!d files - bashrs: suppress 4 findings in ci_guard_steps.sh, each with a reason - census: restore main's train snapshot (#3569); README frontmatter contract_count = census n_files - baselines: classify bin_names_pending_fold.txt and bin_renames.txt (#4430) - pipe-grep-q: 6 new sites become here-strings (66 -> 60) - tracked-ignored: un-ignore kernel A/B evidence logs (321 -> 320) - guard-steps self-test: run the no-manifest fixture outside CI mode (it died rc 2 under GITHUB_ACTIONS) - mutants: car/* of this repo defers to mutants-nightly (operator 2026-09-27, verbatim "exempt the car"); rule 0 in mutants_diff_gate.sh with 7 case rows and 5 mutants Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…in (619 -> 617) scripts/ci/fat_driver.py::cmd_self_test (C, 252 lines, #4507) is now seven _st_* section functions; its 56 rows are unchanged (56/56 before and after). scripts/nightly_manifest.py::main (B-) is now _parser() plus one _cmd_<name> per subcommand behind a COMMANDS table; its --self-test and check_guards_nightly_manifest.sh --self-test pass. Both arrived through main merges after the 617 re-baseline (13b0e89). The baseline stays 617 (#4101). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… on the car (the_tracked_repo_graph_is_fresh) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…us, not the walked tree (guard-cargo#m29) extract:readme grades README frontmatter contract_count as 'resolves: census' (== contracts/census.json n_files). readme_sync wrote the walked count there. census.json is the release train's snapshot (#3569) and lags on any branch that adds a contract, so the two gates demanded different numbers (1844 vs 1837) and one of them was always RED. Body CONTRACT_COUNT blocks stay walked (FALSIFY-README-002); the frontmatter now reads census.json, failing closed if it is unreadable. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…3/3 non-Claude); CI signer stamps patch-id Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

… CI signer)

…gner push fires no pull_request_target) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…ard schema (unsigned; CI signer re-stamps) A4 rejected crux_coverage=partial. Local run of main's check_pr_review_receipt.sh then found two more old-schema fields; all three are now fixed, and it passes (throwaway key): - crux_coverage partial -> none: the ptx-debug surface has no crux contract; finding PRREV-4429-CRUX-PTX-DEBUG-COVERAGE-GAP records that - duplication_coverage not_run -> none (verdict is BLOCK, not PASS) - the ptx-debug cited finding gains its excerpt + excerpt_sha256 from contracts/apr-dogfood-coverage-v1.yaml:109 at fbf106b (total 4, not 3) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

… CI signer)

noahgift · 2026-09-27T19:05:39Z

Closed as superseded by #4502, which now carries this release car's full content (merge 5c54a8e of car/0.70.0 into batch/ont-10). The only commits not carried are 55f86db and 0d7e7e4, which are this PR's own receipt evidence and irrelevant once it is closed. Operator ruling (single release PR): one CI, one review, one merge, no duplicate changes across two open PRs. The branch is kept.

OPERATOR ruling via 77: head vs base, not a stored number. Main aca6f2d has 76 theorem-bearing unrefined modules; this head had 131 (+55, all entered at cf038d3). infra-83: none of the 55 is named by an ONT-10 row. Of the 55, 27 hold a contract-bound root (a lean_theorem names one of their theorems): taking them out of the cone makes `pv discharge check` RED with ORPHANED-ROOT (measured: 85 roots). They stay, pending the cop. The other 28 bind nothing; their root imports move here to the 0.70.1 branch (a move, not a copy; the .lean files are on main already). Each had two import lines (the #4429 merge union), so 56 lines go. Measured: cone 163 -> 135, theorem-bearing 131 -> 103; `pv discharge check --no-lake`: ROOTS 249 pinned, 0 ORPHANED-ROOT, ok. Modules: ProvableContracts/Theorems/AdamW/Analytic.lean ProvableContracts/Theorems/AprCode/HarnessIrRoundtrip.lean ProvableContracts/Theorems/BLAS/Trmm.lean ProvableContracts/Theorems/BLAS/Trsm.lean ProvableContracts/Theorems/DPO.lean ProvableContracts/Theorems/FFT/Bluestein.lean ProvableContracts/Theorems/FFT/Fft2d.lean ProvableContracts/Theorems/FFT/Fft3d.lean ProvableContracts/Theorems/FFT/FftBatched.lean ProvableContracts/Theorems/GPU/DimensionIndependence.lean ProvableContracts/Theorems/Image/Canny.lean ProvableContracts/Theorems/Image/ConnectedComponents.lean ProvableContracts/Theorems/Image/Conv2d.lean ProvableContracts/Theorems/Image/Histogram.lean ProvableContracts/Theorems/Image/HsvRoundtrip.lean ProvableContracts/Theorems/Image/Morphology.lean ProvableContracts/Theorems/Image/Resize.lean ProvableContracts/Theorems/Image/RgbToGray.lean ProvableContracts/Theorems/Image/Sobel.lean ProvableContracts/Theorems/MatMul/MatVecLinearity.lean ProvableContracts/Theorems/Rand/Philox.lean ProvableContracts/Theorems/Rand/Threefry.lean ProvableContracts/Theorems/Sparse/BsrSpmv.lean ProvableContracts/Theorems/Sparse/SellSpmv.lean ProvableContracts/Theorems/Sparse/Spgemm.lean ProvableContracts/Theorems/Sparse/Spmm.lean ProvableContracts/Theorems/Tensor/Einsum.lean ProvableContracts/Theorems/Tokenizer.lean Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…us (1837), per readme_sync #4429 The fold wrote the walked count (1889) into the frontmatter too. extract:readme grades it against contracts/census.json (the train snapshot, #3569), so lint_passes_on_real_contracts failed once census.json was restored. Body blocks stay 1889 (FALSIFY-README-002). check_readme_claims: 0 FAIL. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…aiml#4468) * ci: yoga's CUDA jobs run test binaries under the one host GPU lock yoga's GPU is shared by CI and by agents, and nothing serialised them: an agent test holding the GPU with no lock hung paiml#4429's cuda-unit. Operator ruling 2026-09-26: the GPU is exclusive for CI jobs, via one host lock that every GPU user takes. - cuda-unit (ci/sections.yml, the ephemeral docker runner) and cuda-nightly ada-yoga (the host runner) set CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER to `flock -E 175 -w 600 /run/lock/fleet-gpu/gpu.lock`. cargo runs each TEST BINARY under it, so compiling holds no lock; 175 = not free after 10 min. - Each job first asserts the lock is the HOST's. flock creates a missing file, so without the bind mount it would take a private lock that serialises nothing: cuda-unit asks for the mountpoint, ada-yoga for a root-owned file. The host side (tmpfiles lock, read-only bind mount into the job container, the agents' cargo wrapper and a watcher that ANDONs on an unlocked GPU pid) is paiml/infra#1178, applied through forjar. Refs paiml/infra#1178 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * PMAT-4469: ticket for the yoga CUDA GPU-lock change Refs paiml#4469. Host side: paiml/infra#1182. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * PMAT-4469: quorum receipt — 3/3 PASS on 69c8bad Tool round (quorum-review.sh, lint ok, agreed): gemini-3.1-pro-high PASS + claude-haiku-4-5 PASS. Third lane: a separate claude-sonnet-5 conversation, PASS. Refs paiml#4469 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…p move to PR-2b Cop ruling 19:35Z (split). Gate-graph walk from rc.1 (rc-cut.yml: ci/gate + workspace-test; gate.needs closure) and FINAL (paiml-ontology @60179ae2:714, ONT-10 depends_on): neither row is needed, directly or transitively. - x86-main: runs-on back to main's [self-hosted, Linux, X64, clean-room]. The gate needs the job, not the runner label. - #3668: the `changes` job, guard-cargo's needs/if, the gate's GATE-DOCS-ONLY-RULE, ci_docs_only.sh, check_ci_gate_docs_only_rule.sh, fat_driver's outputs pass-through and the PMAT-3668 roadmap entry are removed. The three guard steps return to guard-cargo. The gate again reads guard-cargo and determinism-compare strictly, as main does, so no gate is loosened versus main. Receipts: fat_driver self-test 59/59; check_ci_gate_mutants_rule, check_guards_are_wired, check_ci_fat_secrets_plumbed, check_ci_gate_mutants_table_rule rc 0; actionlint 2 findings = HEAD's 2; make roadmap-aggregate-check ok. Pmat-Ticket: PMAT-4502 Agent: aprender-07 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

noahgift mentioned this pull request Sep 25, 2026

batch/0.70.0: 0.70.0 fold batch (#4352, #4020, #4062, …) #4372

Closed

noahgift and others added 4 commits September 25, 2026 18:49

This was referenced Sep 25, 2026

ONT-4c5 aprender extractor: capability-cells shape in pv lint (per paiml/infra#950) #4047

Closed

ONT-4c5 (infra#921): aprender capability-cell extractor + arming rule (required Unknown{NotRun} meets to RED) #3972

Open

noahgift and others added 2 commits September 25, 2026 20:09

This was referenced Sep 25, 2026

P0: nightly publishes 1 of 28 [[bin]]s (apr CPU only) and is RED today — build every bin at main head + publish a SHA manifest for lambda #4189

Closed

Every [[bin]] --version prints the build SHA (shared build-sha), then #4189 nightly requires it #4219

Closed

noahgift and others added 8 commits September 25, 2026 20:40

release(0.70.0): bump-version 0.70.0

798fb8c

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

merge origin/main into #4427 — roadmap resolved as the union (PMAT-43…

a94d31c

…46 + PMAT-4421) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

merge origin/main (0005847, #4421) into car/0.70.0 — roadmap.yaml reg…

8bf305b

…enerated by make roadmap-aggregate (union, check ok) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

pr-review receipt for car #4429 at merge 8bf305b (re-bound to merge-b…

a5ce356

…ase 0005847; merge delta quorum 3/3 PASS) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore(pr-review): sign this PR's receipt (PR-REVIEW-SKILL-002 v2 §4.3…

50f6a98

… CI signer)

noahgift enabled auto-merge September 25, 2026 21:27

noahgift added a commit that referenced this pull request Sep 25, 2026

docs(findings): provable-ladder on car #4429 declines on inputs that …

802c5b3

…are on no PR (#4083) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

noahgift disabled auto-merge September 25, 2026 22:02

noahgift and others added 3 commits September 26, 2026 00:02

pr-review receipt for car #4429 at caa9ebe (fix delta quorum 3/3 PASS)

0ab685a

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore(pr-review): sign this PR's receipt (PR-REVIEW-SKILL-002 v2 §4.3…

bfd46bb

… CI signer)

noahgift enabled auto-merge September 25, 2026 22:08

noahgift mentioned this pull request Sep 25, 2026

perf(qwen35-cuda): DP4A requant staleness (#4258) + argmax first-of-ties (#4259) + allocation-free decode plumbing (#4215) — replaces #4322 #4332 #4439

Merged

noahgift and others added 4 commits September 27, 2026 16:04

fix(car): cuda-nightly #4251 step — model default from $HOME, not a m…

b686064

…achine path Path-only (operator go via cop 77): vars.APR_QWEN35_MODEL still wins; the fallback is $HOME/models/… and is exported for the cargo test. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/car-fix-fb' into e6/car-x86fix

76ee838

noahgift mentioned this pull request Sep 27, 2026

fix(bench): qwen35 bench forgets the session prefix before every turn (#4445) #4577

Merged

noahgift disabled auto-merge September 27, 2026 14:21

noahgift mentioned this pull request Sep 27, 2026

PVL-001 EV-9 — PVL-9: provable-ladder required on main #4083

Closed

noahgift enabled auto-merge September 27, 2026 14:28

noahgift added the owner:aprender-36 owning session (cop inbox claims) label Sep 27, 2026

fix(ont): regenerate contracts.nt — README contractCount 1844 -> 1837…

f2f6f8c

… on the car (the_tracked_repo_graph_is_fresh) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

noahgift disabled auto-merge September 27, 2026 15:24

noahgift enabled auto-merge September 27, 2026 15:26

noahgift disabled auto-merge September 27, 2026 17:20

noahgift enabled auto-merge September 27, 2026 17:31

noahgift disabled auto-merge September 27, 2026 17:50

noahgift and others added 2 commits September 27, 2026 20:02

evidence(pr-review): rebind #4429 receipt to fbf106b (deltas r32-r34 …

48b8708

…3/3 non-Claude); CI signer stamps patch-id Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore(pr-review): sign this PR's receipt (PR-REVIEW-SKILL-002 v2 §4.3…

59e35ff

… CI signer)

noahgift mentioned this pull request Sep 27, 2026

flow: signer-bot pushes leave CI action_required — approve held runs in-job #4579

Closed

noahgift and others added 3 commits September 27, 2026 20:37

ci(pr-review): re-trigger pr-review-quorum on the signed head (bot si…

7d0d403

…gner push fires no pull_request_target) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore(pr-review): sign this PR's receipt (PR-REVIEW-SKILL-002 v2 §4.3…

0d7e7e4

… CI signer)

noahgift closed this Sep 27, 2026

noahgift mentioned this pull request Sep 29, 2026

ci(#3668): docs-only PRs skip guard-cargo/determinism, fail-closed (PR-2b of #4588 split) #4639

Closed

Navigation Menu

Repository navigation

release(0.70.0): sealed car — REPLACES #4372 - #4429

release(0.70.0): sealed car — REPLACES #4372#4429
noahgift wants to merge 258 commits into
mainfrom
car/0.70.0

noahgift commented Sep 25, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Sep 25, 2026 •

edited

Loading

Uh oh!

noahgift commented Sep 27, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

noahgift commented Sep 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Sep 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

noahgift commented Sep 27, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

noahgift commented Sep 25, 2026 •

edited

Loading

github-actions Bot commented Sep 25, 2026 •

edited

Loading