6.2 KiB
Independent S2c executable evidence review
Reviewed frozen source f1d79df4979121335ed7b4ce4da84b2fca8a9ef4 and the actual five closed gate triplets. No actionable discrepancy found. The full library gate remains FAILED: 136 passed, 16 failed, 0 ignored, Cargo exit 101. This is private staged typed-writer evidence; it does not establish public source8, S3 indexing/update, S4 activation, native workflows or release acceptance.
Independently verified results
| Gate | Actual result | UTC closure |
|---|---|---|
| list | Exit 0; 152 unique names; admitted | 2026-09-13T12:08:00.913025+00:00 |
| typed | Exit 0; 44 passed, 0 failed/ignored, 108 filtered; admitted | 2026-09-13T12:09:47.043155+00:00 |
| raw tokens | Exit 0; 1 passed, 0 failed/ignored, 151 filtered; admitted | 2026-09-13T12:13:55.998167+00:00 |
| full | Exit 101; 136 passed, 16 failed, 0 ignored | 2026-09-13T12:15:46.279177+00:00 |
| strict | Exit 0; oa-core all-targets Clippy with -D warnings | 2026-09-13T12:24:00.805354+00:00 |
Raw listing order, unique membership, result-record membership, each focused/full outcome and exact summary agree. All 142 previous a2 names and statuses remain, plus exactly the ten test functions independently extracted from committed analysis_source/write/tests.rs. Every writer case passed in both typed and full runs. There are exactly three pass admissions: listing, typed and raw tokens. The full comparison is a comparison receipt, not a full-pass admission.
All five input manifests are byte-identical (SHA-256 6a4c7e3c8d1e6edb7ad306c19abd05009ee410312eb34b68ec30d6408387967b). Their exact 534-path scope, sizes and SHA-256 values were independently compared with committed f1 Git blobs. Git batch type/size/terminal framing and complete consumption were checked. Every result has matching source/head, ordered closed timestamps, no changed inputs, the appropriate exact command, and the same recorded environment. The runner/admitter/closure/comparison and baseline record hashes match their bindings.
The four test commands record the same emitted executable: 37,146,112 bytes, SHA-256 786c5e6deec4fd6f00cc9954ca8190fa47c3c0bdfb0b44dccf032795155438af. The raw Running line matches the recorded Windows path. Strict records no test executable and its raw log ends with the completed unoptimized dev profile, with no warning/error lines. These are historical executable receipt bindings; this review did not read a mutable target executable or run it. Commands use locked oa-core library tests, and all-targets Clippy; they do not contain --offline and do not claim full all-targets runtime or documentation tests.
Failure causes, not merely matching names
The complete 16 panic sections were inspected and compared with a2 raw full evidence at immutable publication commit 355f6d4d244e6b57fd89a8afd80a2dacdea69ba1. The panic source locations and complete diagnostic lines are equal after ignoring only thread IDs and the optional one-time backtrace hint. Every failure reports Unsupported schema version: expected arcadia 3, 4, 5, 6 or 7. at source line 1, column 9, span 8..9. Six names are nesting tests and ten are source-file transaction tests. The machine audit records every exact name, both panic-block hashes and the complete diagnostic line.
Read-only Git equality confirms the six relevant source files remain byte-identical between a2 and f1: model.rs, language.rs, language_nesting_tests.rs, source_project.rs, source_project/source_migration.rs, and source_files.rs. The prior source-derived cause therefore still applies: the current model/writer emits version 8 while the public parser accepts 3–7; bundle migration also reparses through that unchanged public gate. Transaction cases fail in their fixture compilation at source_files.rs:1560, before their later transaction/recovery assertions. These failures do not verify those later behaviors and cannot be waived into a pass.
The prior S2b archive separately preserves the original 472 focused failure and subsequent test-only correction. This checkpoint references that immutable evidence rather than copying it or relabeling historical outcomes. The preserved S2c proposal and source/admission reviews likewise retain their original pre-execution language; the new raw receipts supply later execution evidence.
Portable assembly and limits
The archive copies original raw gate triplets, three admissions, full comparison source/receipt, closure, runner, original/r2 admitters, proposal and relevant handoffs/reviews byte-for-byte. Its manifest pins every local payload (including this report, README, audit and verifier) and the required prior records by exact immutable commit/path/hash. Local original paths are provenance labels, not replay dependencies. Binaries, toolchains, caches and source bodies are omitted; source verification requires a Git repository containing f1, a2 and the prior evidence commit.
verify.py uses explicit require failures, rejects optimized Python before reading evidence, verifies exact archive membership and payload hashes, checks pinned prior Git evidence, independently rereads all 534 source blobs and recomputes the audit. It does not invoke Cargo, tests, recorded helpers, artifacts, GUI or network. The original comparison script is preserved as historical evidence and not rerun: its absolute-path/exclusive-output behavior is replaced for portability by the independent read-only verifier, not by changing its bytes.
The raw recorder relies on assertions without an optimization guard. Its preserved preflight proposal records unset PYTHONOPTIMIZE and the closure reports unoptimized Python; the r2 admitter explicitly refuses optimization. This review independently checks all recoverable raw/source invariants and does not claim forensic proof of the historical process environment or protection against wholesale fabricated raw evidence. Runtime success of the public parser, mixed ordinary/typed updates, save/history installation and native editing remains outside this private milestone.
No compiler/test, GUI, source-input mutation or Gitea/push was performed by this review. Source analysis remains separate from the production root. The evidence is suitable for parent review/publication with its failure scope intact.