{"node_id":"TA_kwDOTKah0NoAKDg5YWRmZWRkNjZlYTRiZGMxNzdhZTVmYWIxZjI3NWY2NjliOGVmNzU","sha":"89adfedd66ea4bdc177ae5fab1f275f669b8ef75","url":"https:\/\/api.github.com\/repos\/b7n0de\/proofbundle\/git\/tags\/89adfedd66ea4bdc177ae5fab1f275f669b8ef75","tagger":{"name":"b7n0de","email":"45888075+b7n0de@users.noreply.github.com","date":"2026-11-08T15:39:31Z"},"object":{"sha":"ec95b24fdc7baae89dfb4a0d7c50c0110561a5e9","type":"commit","url":"https:\/\/api.github.com\/repos\/b7n0de\/proofbundle\/git\/commits\/ec95b24fdc7baae89dfb4a0d7c50c0110561a5e9"},"tag":"v6.5.0","message":"proofbundle 6.5.0\n\nFeature release 6.5.0. See CHANGELOG.md and the release notes for release information. RESTRISIKO_650.md contains the existing risk record for this version; the four additional findings from the deep gate are listed below and remain open in this release.\n\nTHE DEEP GATE FOR THIS RELEASE DID NOT PASS. This tag records the owner's decision to release with the findings listed below. The candidate is 65bd4b7a34b84724298961254eaba5d67bb3fc3b, the head of main after pull request 491. The deep gate ran at 8048103c34c31f22fdffcc194acb29da3f9afea7, the head of main after pull request 490; pull request 491 changes only the date line in CHANGELOG.md, the release-notes source release_notes\/release-source.json and the rendered release notes, and nothing under src\/ or tests\/. The full test suite referenced by the pre-tag receipt, a 300-second short soak and the Python\u2013Rust differential were run at the candidate commit; the full mutation run used the candidate plus one change to a test helper, as described below. The tagged commit is the evidence commit on the candidate; it changes only the pre-tag receipt, the short soak record and the differential record, and a build at it gives the same sdist and wheel bytes as at the candidate.\n\nAt the tagged commit, with the candidate sdist and wheel in place, the audit candidate matrix measures 33 of 34 checks: 29 PASS and 4 FAIL. C6.2, C6.3 and C8.2 FAIL because the signed short soak and the signed differential carry the deep gate verdict REFUTED_IN_ITERATION_1, which is not a pass. C10.2 FAILS because the 6.5.0 readiness slot is not filled; it is an informative cell. EXT.1, the independent external audit, is not measured, and audit_candidate_ready is false. Do not read this tag as a gate that was passed.\n\nHead: ec95b24fdc7baae89dfb4a0d7c50c0110561a5e9, the evidence commit for candidate 65bd4b7a34b84724298961254eaba5d67bb3fc3b. The main-head measurement at 2026-11-08T15:03:46Z records merge commit ac4129f7430688f71ad45827c2b316cdbcf94147 with the candidate and the evidence commit as its parents, in that order; its tree is identical to the evidence commit's tree.\n\nPre-tag audit receipt: audit_artifacts\/650\/pre_tag_receipt_v6.5.0.json\n  subject_tree_digest 93af86e13b7a542885816d4b14293f3d938488796a43874ad50006b4edf47403\n  audit_output_digest 61f2dff31bd0eefa6b236bdd1e2dadfa81f80f44ab56b5131a6b95ac72e5e522\n    (sha256 of the full-suite log on 65bd4b7a34b84724298961254eaba5d67bb3fc3b: 26223 passed, 156 skipped, 25526 subtests, exit 0)\n  produced 2026-11-08T10:52:50Z, runner farmer\n  signed with the pinned key iJipntJA8N\/\/h+ln9CgLzeC9n\/M5OdCZNeBhbfagom8=; the pre-tag check at the tagged commit reports this receipt as verified\n\nMutation run:\n  117 operators, 115 KILLED verdicts and 2 SURVIVED verdicts recorded as expected, 0 reported gaps;\n  run with scripts\/mutation_check.py from local commit eb55f2211313011d31da59e497c045d6b6744dad,\n  whose only committed change from the candidate is to the test helper in tests\/test_plugin_gate_log.py;\n  after fetching and pruning origin, no local remote-tracking ref contained that commit;\n  the baseline time limit was set to 10800 s for this run only, while the committed value was 5400 s;\n  shards 1 to 4, 7 and 8 of 8 ran as shards, and the 30 operators of shards 5 and 6 ran in 6 groups of 5;\n  for those groups only, scripts\/mutation_baseline_allowlist.json had an uncommitted change from its\n  empty committed entry list, allowing call:AssertionError for these two timing tests under load:\n    `tests.test_budget_kostenkurve.TestObergrenzeAmGroesstenZugelassenenWert::test_kosten_am_limit_unter_der_obergrenze[renewal_work]`\n    `tests.test_core_layer_gate_sees_every_spelling::test_CONTROL_a_chain_of_literal_attribute_reads_is_read_in_time[1500 one-argument __getattribute__ calls]`\n  the renewal_work test failed in the baselines of groups 5 and 6;\n  verdict over all 12 logs: 117 verdict lines for 117 operators, none missing, none twice\n\nShort soak:\n  300 s at the candidate, seed 0, 2063928 iterations over 72 parsers, 0 untriaged crashes, 0 false accepts;\n  5 parsers skipped because they need a fixture (agt_receipt twice, sigstore_bundle, trace_record, enclave);\n  signed with the pinned key and bound to the candidate commit, its tree digest, the sdist\n  74cb56560447cf5b03c2778a677b407dc74f666aab518647bfd05cade9161ad0 and the wheel\n  c7e1b1d61c9fbc4e59d0d132de05b33d81057a837a8705002557b15908d6b8cf;\n  its gate line carries the deep gate verdict REFUTED_IN_ITERATION_1, so the matrix cells C6.2 and C6.3, which read it, are FAIL;\n  record audit_artifacts\/360\/fuzz_soak_latest.json, sha256\n  c95188b45da2787c7a2d6ba1477953d8a1aeee1b6f14cfcb415bc4d459a59dd9\n\nPython and Rust differential:\n  45 relation vectors at the candidate; the differential record reports agree_python_rust=true for every row. In the two valid-baseline rows, Python records lineage as null and Rust records it as NOT_EVALUATED. This is agreement under the differential's comparison, not identical output or a correctness proof of either implementation;\n  measured with a debug build of pb_verify_rs (rustc 1.95.0 (59807616e 2026-04-14)), Python 3.10.12;\n  signed with the pinned key and bound to the same commit, tree digest, sdist and wheel;\n  its gate line carries the deep gate verdict REFUTED_IN_ITERATION_1, so the matrix cell C8.2 is FAIL;\n  record audit_artifacts\/360\/rust_differential_matrix.json, sha256\n  f46d3081d4c2df941b2826ff880956126253f9a1e3ea7d13fa6d5148937b2825\n\nFindings register: audit_artifacts\/findings_register_361.json\n\nKnown limitations of this release.\n\nDeep gate for this release: REFUTED_IN_ITERATION_1. A limited adversarial review used one read-only lens at 8048103c34c31f22fdffcc194acb29da3f9afea7; the second model family and a second iteration did not run. The changes from that commit to this release leave the verifier code unchanged. The review confirmed four refutations of pre-registered targets. This release has no WITHSTANDS verdict. The four findings remain open in this release; the first pull request after the tag is planned to address the trace record finding across all of its path walks.\n\nverify_sigstore_bundle (proofbundle.adapters.sigstore_bundle) reads the subject of the leaf certificate outside its error handling. A leaf certificate whose subject UTF8String carries invalid UTF-8 makes verify_sigstore_bundle raise a ValueError before any chain or signature check. This reader is new in 6.5.0. Treat any exception from this reader as a failed verification, never as acceptance.\n\nThe trace record reader (proofbundle.adapters.trace_record) still builds one path string per member that carries all of its ancestors, as in 6.4.0. In the measurement for this release, 3 percent more input gave 108 times the traced memory, and under a 600 MiB address-space limit verify_trace_record raised an uncaught MemoryError on a 649,293-byte input before any signature was checked. An input-size limit alone is not an established mitigation. Until this is fixed, do not process untrusted records in a shared or critical process. If such processing is necessary, isolate it in a worker with enforced memory and time limits, and treat an exception or worker termination as a failed verification, never as acceptance.\n\nAn eval-result v0.1 statement whose claims list holds an empty object ({}) verifies, although docs\/IN_TOTO_PROFILE.md asks for metric, comparator and threshold in each claim entry. Do not treat a successful verification as evidence that each claim entry carries these fields; check them where they matter.\n\nissue_sd_jwt accepts a vct of 1,000,001 characters and emits a compact serialization that check_binds_bundle then refuses. The same behavior was observed in 6.4.0. Treat a refusal by check_binds_bundle as a failed check, and keep vct values short.\n\nThe deep gate verdict is REFUTED_IN_ITERATION_1, not WITHSTANDS. This leaves the audit-candidate matrix cells C6.2, C6.3 and C8.2 red, because the short soak and the differential records carry the same gate line. This is not a passing matrix result, and it does not establish that the readiness artifacts passed the provenance or freshness checks of the matrix. At the tagged commit, the recorded pre-upload comparison exited with code 0 and reported: \"OK: the tag run built the signed candidate pair\". This result is separate from the failed matrix checks and is not evidence that the matrix's provenance, freshness or gate-line requirements passed.\n","verification":{"verified":false,"reason":"unsigned","signature":null,"payload":null,"verified_at":null}}