From 4e0d1c69339e73cde472fe26d5770d91cefe2abd Mon Sep 17 00:00:00 2001 From: Noah Gift Date: Sat, 19 Sep 2026 11:50:09 +0200 Subject: [PATCH 1/2] =?UTF-8?q?evidence(ladder):=200.68.1=20receipts=20re-?= =?UTF-8?q?measured=20at=20main=20a6f54e84c=20on=20both=20hosts=20?= =?UTF-8?q?=E2=80=94=20all=20eight=20rungs=20with=20sha256,=20RED=20by=20c?= =?UTF-8?q?onstruction=20on=20the=20Qwen3.5=20rungs=20(#3510,=20ONT-4c1=20?= =?UTF-8?q?witness)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Both receipts produced by scripts/model_ladder.sh at a6f54e84c (the first main with #3507's sha256 rows and #3511's four dense rungs), apr 0.68.1 (a6f54e84c), lambda RTX 4090 cc 8.9 and gx10 GB10 cc 12.1. Every rung present, sha_ok, and carries the measured sha256 — the witness ONT-4c1's ladder-measured shape needs on day one. Green: qwen2-1.5b, qwen3-1.7b (both hosts), qwen3-8b (gx10; lambda optional rung: golden Empty output, #3486). Red by construction: qwen35-0.8b/2b fall back to CPU on main until #3517 merges; qwen35-4b/9b/27b exit 101 at forward_qwen35.rs:132 (GQA inside GDN, nk != nv — #3346 root cause). The T-2 judge is not a PR check; this changes no gate. Co-Authored-By: Claude Fable 5.1 --- evidence/dogfood/models/0.68.1/gx10.json | 121 ++++++++++++++++++++- evidence/dogfood/models/0.68.1/lambda.json | 121 ++++++++++++++++++++- 2 files changed, 230 insertions(+), 12 deletions(-) diff --git a/evidence/dogfood/models/0.68.1/gx10.json b/evidence/dogfood/models/0.68.1/gx10.json index 6bd2307ea9..9a67d095b7 100644 --- a/evidence/dogfood/models/0.68.1/gx10.json +++ b/evidence/dogfood/models/0.68.1/gx10.json @@ -2,19 +2,20 @@ "schema": "apr-model-ladder-receipt/v1", "host": "gx10", "version": "0.68.1", - "sha": "927587272", + "sha": "a6f54e84c", "isa": "aarch64", "gpu": "NVIDIA GB10", "cc": "12.1", - "date": "2026-09-18T14:36Z", - "apr_version": "apr 0.68.1 (927587272)", - "executed": 4, - "red": 0, + "date": "2026-09-19T09:40Z", + "apr_version": "apr 0.68.1 (a6f54e84c)", + "executed": 8, + "red": 5, "rungs": [ { "id": "qwen2-1.5b-q4km", "present": true, "sha_ok": true, + "sha256": "6a1a2eb6d15622bf3c96857206351ba97e1af16c30d7a74ee38970e434e9407e", "required": true, "qa_rc": 0, "capability_match": { @@ -45,6 +46,7 @@ "id": "qwen3-1.7b-q4km", "present": true, "sha_ok": true, + "sha256": "b139949c5bd74937ad8ed8c8cf3d9ffb1e99c866c823204dc42c0d91fa181897", "required": true, "qa_rc": 0, "capability_match": { @@ -75,6 +77,7 @@ "id": "qwen3-8b-q4km", "present": true, "sha_ok": true, + "sha256": "d98cdcbd03e17ce47681435b5150e34c1417f50b5c0019dd560e4882c5745785", "required": false, "qa_rc": 0, "capability_match": { @@ -105,6 +108,7 @@ "id": "qwen35-0.8b-q4km", "present": true, "sha_ok": true, + "sha256": "bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517", "required": true, "qa_rc": 0, "capability_match": { @@ -122,9 +126,114 @@ "ran": true, "fallback": false, "rc": 0 + }, + "cuda": { + "ran": true, + "fallback": true, + "rc": 0 } }, - "green": true + "green": false + }, + { + "id": "qwen35-2b-q4km", + "present": true, + "sha_ok": true, + "sha256": "aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223", + "required": true, + "qa_rc": 5, + "capability_match": { + "passed": true, + "skipped": false, + "message": "Architecture 'qwen35': CPU-only \u2014 Gated DeltaNet runs on the CPU (#3091); the GPU backend has no SSM kernels (#3090), GPU gates skipped" + }, + "golden_output": { + "passed": false, + "skipped": false, + "message": "golden_output_cpu: Empty output" + }, + "backends": { + "cpu": { + "ran": true, + "fallback": false, + "rc": 0 + }, + "cuda": { + "ran": true, + "fallback": true, + "rc": 0 + } + }, + "green": false + }, + { + "id": "qwen35-4b-q4km", + "present": true, + "sha_ok": true, + "sha256": "00fe7986ff5f6b463e62455821146049db6f9313603938a70800d1fb69ef11a4", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false + }, + { + "id": "qwen35-9b-q4km", + "present": true, + "sha_ok": true, + "sha256": "03b74727a860a56338e042c4420bb3f04b2fec5734175f4cb9fa853daf52b7e8", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false + }, + { + "id": "qwen35-27b-q4km", + "present": true, + "sha_ok": true, + "sha256": "84b5f7f112156d63836a01a69dc3f11a6ba63b10a23b8ca7a7efaf52d5a2d806", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false } ] } diff --git a/evidence/dogfood/models/0.68.1/lambda.json b/evidence/dogfood/models/0.68.1/lambda.json index 37b7806d8e..d99535b797 100644 --- a/evidence/dogfood/models/0.68.1/lambda.json +++ b/evidence/dogfood/models/0.68.1/lambda.json @@ -2,19 +2,20 @@ "schema": "apr-model-ladder-receipt/v1", "host": "lambda", "version": "0.68.1", - "sha": "927587272", + "sha": "a6f54e84c", "isa": "x86_64", "gpu": "NVIDIA GeForce RTX 4090", "cc": "8.9", - "date": "2026-09-18T14:28Z", - "apr_version": "apr 0.68.1 (927587272)", - "executed": 4, - "red": 1, + "date": "2026-09-19T09:26Z", + "apr_version": "apr 0.68.1 (a6f54e84c)", + "executed": 8, + "red": 6, "rungs": [ { "id": "qwen2-1.5b-q4km", "present": true, "sha_ok": true, + "sha256": "6a1a2eb6d15622bf3c96857206351ba97e1af16c30d7a74ee38970e434e9407e", "required": true, "qa_rc": 0, "capability_match": { @@ -45,6 +46,7 @@ "id": "qwen3-1.7b-q4km", "present": true, "sha_ok": true, + "sha256": "b139949c5bd74937ad8ed8c8cf3d9ffb1e99c866c823204dc42c0d91fa181897", "required": true, "qa_rc": 0, "capability_match": { @@ -75,6 +77,7 @@ "id": "qwen3-8b-q4km", "present": true, "sha_ok": true, + "sha256": "d98cdcbd03e17ce47681435b5150e34c1417f50b5c0019dd560e4882c5745785", "required": false, "qa_rc": 5, "capability_match": { @@ -105,6 +108,7 @@ "id": "qwen35-0.8b-q4km", "present": true, "sha_ok": true, + "sha256": "bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517", "required": true, "qa_rc": 0, "capability_match": { @@ -122,9 +126,114 @@ "ran": true, "fallback": false, "rc": 0 + }, + "cuda": { + "ran": true, + "fallback": true, + "rc": 0 } }, - "green": true + "green": false + }, + { + "id": "qwen35-2b-q4km", + "present": true, + "sha_ok": true, + "sha256": "aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223", + "required": true, + "qa_rc": 0, + "capability_match": { + "passed": true, + "skipped": false, + "message": "Architecture 'qwen35': CPU-only \u2014 Gated DeltaNet runs on the CPU (#3091); the GPU backend has no SSM kernels (#3090), GPU gates skipped" + }, + "golden_output": { + "passed": true, + "skipped": false, + "message": "3 golden test cases passed on the CPU forward (#3091); GPU declined (#3090)" + }, + "backends": { + "cpu": { + "ran": true, + "fallback": false, + "rc": 0 + }, + "cuda": { + "ran": true, + "fallback": true, + "rc": 0 + } + }, + "green": false + }, + { + "id": "qwen35-4b-q4km", + "present": true, + "sha_ok": true, + "sha256": "00fe7986ff5f6b463e62455821146049db6f9313603938a70800d1fb69ef11a4", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false + }, + { + "id": "qwen35-9b-q4km", + "present": true, + "sha_ok": true, + "sha256": "03b74727a860a56338e042c4420bb3f04b2fec5734175f4cb9fa853daf52b7e8", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false + }, + { + "id": "qwen35-27b-q4km", + "present": true, + "sha_ok": true, + "sha256": "84b5f7f112156d63836a01a69dc3f11a6ba63b10a23b8ca7a7efaf52d5a2d806", + "required": true, + "qa_rc": 101, + "capability_match": null, + "golden_output": null, + "backends": { + "cpu": { + "ran": false, + "fallback": false, + "rc": 101 + }, + "cuda": { + "ran": false, + "fallback": true, + "rc": 101 + } + }, + "green": false } ] } From ebd57d10d8070f163d1de801fe02e490a902daa0 Mon Sep 17 00:00:00 2001 From: Noah Gift Date: Sat, 19 Sep 2026 11:53:56 +0200 Subject: [PATCH 2/2] audit: quorum artifact for PR #3523 (3/3 PASS, PMAT-3524) Co-Authored-By: Claude Fable 5.1 --- docs/audits/quorum-PMAT-3524-pr3523.json | 139 +++++++++++++++++++++++ 1 file changed, 139 insertions(+) create mode 100644 docs/audits/quorum-PMAT-3524-pr3523.json diff --git a/docs/audits/quorum-PMAT-3524-pr3523.json b/docs/audits/quorum-PMAT-3524-pr3523.json new file mode 100644 index 0000000000..05e7c294f4 --- /dev/null +++ b/docs/audits/quorum-PMAT-3524-pr3523.json @@ -0,0 +1,139 @@ +{ + "ticket": "PMAT-3524", + "base": "main", + "base_resolved": "origin/main", + "base_note": "local main differs from origin/main by 101 commit(s); judged against origin/main", + "head": "4e0d1c69339e73cde472fe26d5770d91cefe2abd", + "diff_sha256": "a2856fc0604d0912503b77c2e7c5ecd1ba295bb27d41e9c5d03d092144f18ced", + "width": 3, + "executor": "agy", + "prompt_mode": "inline", + "prompt_bytes": 10505, + "author": { + "model": "claude-fable-5-1", + "family": "claude", + "source": "flag" + }, + "agreed": true, + "lanes": [ + { + "lane": 1, + "status": "SUCCESS", + "verdict": "PASS", + "summary": "The diff perfectly aligns with PMAT-3524. It strictly replaces the `gx10.json` and `lambda.json` files with the new model ladder receipts generated at `a6f54e84c`. Exactly 8 rungs are present in both files, all have `present: true`, `sha_ok: true`, and the measured `sha256` field added. Furthermore, the receipts correctly reflect the red-by-construction state on main for Qwen3.5 models (with CUDA fallbacks indicated, and 4b/9b/27b exiting with code 101). No other files are touched.", + "findings": [], + "raw_bytes": 2222, + "err_bytes": 0, + "envelope_status": "SUCCESS", + "verdict_source": "structured_output", + "grounding_check": "parity", + "model": "gemini-3.1-pro-high", + "model_measured": "gemini-3.1-pro-high", + "model_source": "measured", + "family": "gemini" + }, + { + "lane": 2, + "status": "SUCCESS", + "verdict": "PASS", + "summary": "Diff strictly adheres to PMAT-3524: touches only evidence/dogfood/models/0.68.1/{gx10,lambda}.json to update model ladder receipts for commit a6f54e84c with all 8 rungs present, sha_ok=true, measured sha256 hashes recorded, and Qwen3.5 rungs red-by-construction as expected on main. No gates or forbidden files were modified.", + "findings": [ + { + "claim": "gx10.json updated to main a6f54e84c with 8 rungs, sha_ok=true, measured sha256 digests, and red Qwen3.5 rungs", + "file": "evidence/dogfood/models/0.68.1/gx10.json", + "fix": "None needed", + "grounding": "cited", + "line": 5 + }, + { + "claim": "lambda.json updated to main a6f54e84c with 8 rungs, sha_ok=true, measured sha256 digests, and red Qwen3.5 rungs", + "file": "evidence/dogfood/models/0.68.1/lambda.json", + "fix": "None needed", + "grounding": "cited", + "line": 5 + } + ], + "raw_bytes": 2855, + "err_bytes": 0, + "envelope_status": "SUCCESS", + "verdict_source": "structured_output", + "grounding_check": "parity", + "model": "gemini-3.8-flash-high", + "model_measured": "gemini-3.8-flash-high", + "model_source": "measured", + "family": "gemini" + }, + { + "lane": 3, + "status": "SUCCESS", + "verdict": "PASS", + "summary": "PR touches only evidence/dogfood/models/0.68.1/{gx10,lambda}.json, updating both model ladder receipts to main a6f54e84c with 8 rungs, sha_ok=true, measured sha256 hashes, and red Qwen3.5 rungs as mandated by PMAT-3524. No gates or forbidden files were touched.", + "findings": [ + { + "claim": "gx10.json receipt updated for commit a6f54e84c with 8 rungs, sha256 digests, and expected red statuses for Qwen3.5 rungs", + "file": "evidence/dogfood/models/0.68.1/gx10.json", + "fix": "None needed", + "grounding": "cited", + "line": 5 + }, + { + "claim": "lambda.json receipt updated for commit a6f54e84c with 8 rungs, sha256 digests, and expected red statuses for Qwen3.5 rungs", + "file": "evidence/dogfood/models/0.68.1/lambda.json", + "fix": "None needed", + "grounding": "cited", + "line": 5 + } + ], + "raw_bytes": 2765, + "err_bytes": 0, + "envelope_status": "SUCCESS", + "verdict_source": "structured_output", + "grounding_check": "parity", + "model": "gemini-3.7-flash-high", + "model_measured": "gemini-3.7-flash-high", + "model_source": "measured", + "family": "gemini" + } + ], + "dissent": [], + "dedup": [ + { + "file": "evidence/dogfood/models/0.68.1/gx10.json", + "line": 5, + "lanes_agreeing": [ + 2, + 3 + ], + "claims": [ + "gx10.json receipt updated for commit a6f54e84c with 8 rungs, sha256 digests, and expected red statuses for Qwen3.5 rungs", + "gx10.json updated to main a6f54e84c with 8 rungs, sha_ok=true, measured sha256 digests, and red Qwen3.5 rungs" + ] + }, + { + "file": "evidence/dogfood/models/0.68.1/lambda.json", + "line": 5, + "lanes_agreeing": [ + 2, + 3 + ], + "claims": [ + "lambda.json receipt updated for commit a6f54e84c with 8 rungs, sha256 digests, and expected red statuses for Qwen3.5 rungs", + "lambda.json updated to main a6f54e84c with 8 rungs, sha_ok=true, measured sha256 digests, and red Qwen3.5 rungs" + ] + } + ], + "uncovered": [], + "coverage_source": "lanes", + "partial": false, + "partial_reasons": [], + "auto_merge": { + "checked": true, + "was_armed": false, + "disarmed": false, + "note": "auto-merge not armed" + }, + "lint": { + "ok": true, + "output": "receipt complete: kind=artifact lanes=3 author=claude-fable-5-1/claude" + } +}