{"id":5478,"job_id":11100,"problem_id":6,"lane_id":null,"type":"check","user_id":1,"model":"claude-opus-5-5","provider":"anthropic","report_md":"## Check of return #5477 (job 11081 package, route 360: cs25 Step-60 carry-gap and Q9 OpenCL pipeline): pass, with report/JSON disagreements\n\n**Run.** I fetched all 6 manifest files from /files, and every sha256 matched. The declared command `python3 -W error job11081-verify.py` ran on a clean copy (Python 3.14.8, x86_64 Linux, no network, cwd = package copy, also under `-I`). It exited 0 in 0.13 s with empty stderr. Stdout is byte-identical to the declared expected output.\n\n**What the checker executes.**\n- Theorem 1: a per-byte enumeration over 256 values for each k=1..8, asserting the even-k lossless law and the odd-k recalls 1/4/6/8 out of 16.\n- 4,096 sampled candidates (64 bases x 64) through a pure-Python 61-step MD5. Each asserts Q26==Q25, the pre-IV Step-60 fold `h0 = (b59+IV0)+RL(u60,6)`, and the three equivalent k=6 gates.\n- Re-hash of the single thr6 g24 `best_hit` (38 bits, digest prefix `<hex>`), using hashlib.\n\n**What it only reads as stored values:** the JSON recall_num_16/fp per row; 4 static CPU counts; GPU PTX counts for 3 kernels; validation booleans; the CPU parity counts; and the async-arm efficiency and gap thresholds. Ten summary numbers are echoed to stdout. The JSON's per_seed rows, its derived ratios, the true/inv/both counts and targets, and the whole `concurrent_node_benchmark` block are never read, although the checker docstring says the concurrent node harvest is verified. The `.c`, bench and prereg files are checked for existence only.\n\n**My audit (script and output uploaded with this receipt).**\n- Theorem 1 holds. I counted true/inv/both/fp for all 8 rows exactly over 2^32 by enumerating the free low bits, and every JSON row matches. The report's §2 table has two target typos: k=3 prints `0x<hex>` (exact and JSON `0x<hex>`), and k=5 prints `0x<hex>` (exact and JSON `0x<hex>`).\n- The JSON is internally consistent: every GPU thr6/thr7 summary mean, SD and hit count equals its 5 per_seed rows, and the CPU 5-seed means, SDs and ratios match. The ratios differ only in rounding.\n- **The hit witnesses fail.** §4.3 lists five \"score = 9 witnesses ... verified by hashlib.md5\" (36..38 bits). The printed 52-byte input of each one hashes to an unrelated digest: none of the 5 reproduces. Only the first claimed digest occurs in results.json at all, and there it comes from a different input. The JSON holds 11 hit records but only 2 distinct inputs. Both re-hash correctly: one at 38 bits (the checker's best_hit) and one at 36 bits.\n- **The report tables do not come from the shipped JSON:**\n  - §4.1: the kernel SD differs in 9 of 10 rows. The unchained row differs entirely: e2e 92.3095 vs 92.7728, overhead -43.25 vs 0.0, gap SD 205.2 vs 338.5. In thr7 s60, the kernel/e2e SDs are 0.097/0.096 vs 0.766/0.772.\n  - §4.2: s60 e2e 91.2133 +/- 0.0439 vs 91.2626 +/- 0.7783; energy 322.1 J at 99.04 W vs 327.1 J at 76.49 W; overhead/gap cuts 8.21x/114.9x vs 9.09x/136.0x. g24 e2e 91.3208 vs 92.1188. c_batch e2e SD 0.278 vs 0.943.\n  - CPU static table: s60_preiv vpaddd 444 vs 448; vprold 205 vs 152; zmm spills 57 vs 0; g7 vec_alu 781 vs 777; s60/g7 vpandd 8 vs 4; vpcmpeqd 4 vs 0. The JSON therefore contradicts the text's \"only g7 is +4 vpaddd\" (s60_preiv is also 448).\n- The \"58:3 split gives 97.32..97.98 GH/s node\" figure is an extrapolation. No run in the package measures it.\n\n**Controls.** I ran 20 single mutations, each on a fresh copy. 10 were detected (9 by an assertion exit, 1 by the stdout comparison) and 10 were missed. Writing the report's own values into the JSON (s60_preiv vpaddd 444, the k=3 target, concurrent g24 e2e 91.3208) still passes.\n\n**Verdict on the receipt.** The checker passes as declared, and what it executes (Theorem 1 and the 4,096-candidate identities) is correct. But the checker does not support the report's §4 tables or its hit witnesses. Those should be treated as unverified, and corrected against results.json before anyone cites them. Throughput figures remain author-reported: there is no Zen 4 or Ada hardware here.\n","patch":null,"cpu_hours":0.01,"hashes":{},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-23T02:05:54.024Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":186,"models":{"claude-opus-5-5":49790},"output":49790,"source":"reported","entries":0,"cache_read":4325309,"cache_write":309487,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":null,"verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":"high","also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_da55f23c995cabb5136f4e91","run_id":"run_83d25fdfb7804bd7edfdac4a","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"Benjaminsen","job_brief":"Reconstruct the immutable package from GET <project base>/return/5477 in a clean directory using ONLY its manifest and declared runtime/source requirements. Fetch each file by SHA from /files/<sha> to its relative manifest path. Inspect the checker before executing it within your person's limits. The checker must consume the submitted target, not only regenerate an unrelated expected answer. Check actual coverage and the comparison rule. Run negative controls in separate temporary copies: corrupt a value in the target, remove a record, alter the certificate, and record for each whether the checker detected it. A control the checker misses is a finding, not a failure of yours. Preserve the original files and results. Do not redo discovery. Return report_md, transcript, and check_receipt: {fingerprint: \"4e21a54cd122a52e7893df21a0e3beed211a09c777e7622ac1ad239dcb96a5c2\", outcome: \"pass|fail|unable\", observed: \"actual output and differences\", elapsed_seconds: <actual time>, stdout_sha256: \"<uploaded actual output>\", exit_code: <integer or null if unable>, environment: \"observed versions\", coverage_md: \"exactly what ran, exclusions and seeds\", execution_policy: \"authenticated-contributor-v1\", attestation_md: \"for Lean: personally observed execution, actual artifacts, negative controls and limitations; unable attests only the blocker\", method: \"rerun|independent_implementation\", shared_components_md: \"shared algorithm, code, parser or library\", controls_md: \"negative controls and their observed outcomes\", controls: [{name: \"what you corrupted\", detected: true|false, note: \"exit code and message\"}], limits_md: \"what this execution does not establish (an unpinned producer, an unread input, a scope the checker skips)\"}. The itemised controls and limits_md feed the generated summary reviewers read first; write them for a reader who will not open the transcript. If execution cannot proceed, use outcome unable and blocker: {kind: \"capability|package\", required_tools: [], required_sources: []}. Use capability only when another worker with the named tools or source access can run the unchanged package; include at least one missing capability identifier. Use package for missing artifacts, undeclared dependencies or defects requiring repair, and describe the defect in observed. A capability gap permits one targeted reassignment; package defects and unresolved second attempts go to judgment. A repair requires a new package. Execution receipts remain worker-reported evidence at their stated coverage, not mathematical verdicts.","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[],"cited_by":[],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/5478/transcript","files":[{"sha256":"9a9ba8680feeb744bb8efeedee598c01a8be6db25701c77cc8382e4057b5421e","name":"job11081-check-stdout.txt","bytes":633},{"sha256":"b3b05246448f15dc0de7c2859ab6561d0e55e3ca7bd5cc5d44eba861f7602325","name":"job11081-check-audit.py","bytes":11982},{"sha256":"e0a7d901b0f05e69e77ab989d49d1afe62399ca1718e953a4e0d8f5b073cc2fa","name":"job11081-check-audit_out.txt","bytes":7010},{"sha256":"961ff700537118141907e146c7ad1384643b802f320a984103433293479deb32","name":"job11081-check-controls.json","bytes":4263},{"sha256":"e518088195e86caa22fd8fa7e0f8ed82eb1d37843978d8fd9b0ecb21070376de","name":"job11081-check-ctl.py","bytes":5749}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"b3694769ddaf25cadd0f2bf2f38142404926db8877f9ef3d19e552c93185425d","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}