ARCADIA-Studio/harness/experiments/analysis-mixed-emission-20260925/RUNTIME-REVIEW.md

Independent emission runtime readback

Final source: 8d52dab97c92cc31baae9f20d8c324b55818aa5d. The completed focused run passes11 new tests; full oa-core library execution completes253pass/16 inherited failures, listing269 names. Strict oa-core all-target Clippy exits0. Format exits1 with unchanged baseline bytes. This is not a green full-suite/format claim or accepted-output authority.

Reviewed and executed only the prepared read-only inspect_runtime.py, after parent confirmed all five gates complete, using bundled Python -I -B with final source, 8d52dab, 11. No Cargo, product binary, test executable, debugger or repeated-control launcher was run by this reviewer. The checker was unchanged from its prepared SHA-256 aec366f529b9c8466316a304bcc7ca6688863be2b8da88f03d94cb6a8114019b.

Final evidence reconciliation

• EMISSION-RUNTIME-READBACK.json:5930 bytes, SHA-256 c9d4abb060a34e862769c4a6428da560cb5bb4f6daf82b57d1ad5ff828b9b55e. It records each exact command, source/receipt/log/manifest/launcher pin and the11 added test names.
• All555 declared inputs match immutable Git blobs and current raw disk bytes; final HEAD matches and tracked tree is clean. All five manifests are identical at SHA-256 42b4e07ac4d92c0e4f04d98db862e57f95ad7bf9b5fb3b485d7e0cc971918b50. All receipts report unchanged inputs and no verification failures.
• Focused/full/list identify the same actual test executable:42,402,816 bytes, SHA-256 29d8a4761fd8f09a3c0eafeb11cca07f0d45c711bf1b969d97c8b29d2b3ab446. Current bytes, actual logged Running paths and receipt identities agree.
• Every final launcher matches before/after SHA-256 7819914383abe8a578676aed661704c7ecd0ce9894491152e1859035316e8401. Independently compared its complete source with the previously reviewed carried launcher: only source pin/message and output directory changed. Actual command status propagates; changed source/helper checks remain intact.
• Recorded environment matches the prior carried environment exactly: one compiler job, incremental0, isolated target/toolchain directories, absent RUSTFLAGS/RUSTC/RUSTUP_TOOLCHAIN and absent RUST_MIN_STACK/ARCADIA_CODEC_LIFECYCLE_PHASE. This is declared-source and recorded-environment reconciliation, not universal toolchain/RSS provenance.
• All258 prior outcomes match by exact name/status. Focused names equal exactly the11 added successful names; full and listing name sets match with no duplicates. The16 failure blocks are byte-identical after replacing only numeric thread PID; this checker does not normalize line endings, trim whitespace or rewrite locations/messages. Format bytes match exactly: c1c6b12861c8117ce6ffc6d6742ebf92618aafb10160ec6f99d370ca72ebab2d.
• Baseline full/list/fmt raw logs and receipts match members of published carried evidence ZIP pinned to SHA-256 2eca143b706458c1c0217af715fa6f17c01a0b9f85a5313d6a231dabe1aba7c9. No editable baseline summary substitutes for those bytes.

Initial failure remains failed

Initial source 19a4082117b88caf332198e4c685bde91005cb52 focused run passed11, raw SHA-256 96158d78a50b54963dcd39f1a2e642671cf78bdfa97c995979c71541e7cabdb5. Its full run completed only135 outcome rows, then aborted at the existing ordinary-index small-stack control with3221225725/0xc00000fd. Raw SHA-256 7a23c1151cfba02f6527bef6c8ffd53ffa5eac0e0598e37728733bc85230e538. It is not a normal completed16-failure result. Both original manifests pin 7f025170115bdf4884b00af20019565c9d214713d5a8f8fc9d78d3c819db4d00; recorded input checks are clean.

The preserved initial binary is42,402,816 bytes with SHA-256 777457cfefb8a3046e3f0809cabd936cfb742abe792befad2c50e05dfac1bf6b, verified directly from artifacts/19a4082-oa_core.exe. Equal length does not imply identity with the repaired binary.

Debugger raw SHA-256 fca0fc9a3c59b95beb84d1690df6f0b289dcc46bd949337de555f7102b2cce59 matches its receipt and shows0xc00000fd with Exact::parse_decimal/parse_literal, Cursor::exact_literal and syntax::binding active. Debugger process exit0 is not a test pass. The initial source's worker/parse_impl/parse_with_profile/parse/accepted/indexed prologue reservations match the preserved prior-fa7 records exactly; attributing this recurrence to growth in those frames is unsupported.

Repair and repeated-control scope

The final code delta from19a4082 changes syntax.rs and the repair note only; emission source/tests remain unchanged. Independent source-review ownership for the parser extraction remains separate from this runtime audit.

Static prologue observations: initial binding reserves0x6a88=27,272 bytes. Final binding reserves0x9d8=2,520; prefix reserves0x2dc8=11,720, so their jointly active prefix/numeric branch reserves14,240, a13,032-byte reduction in those recorded frames. Continuation reserves0x3468=13,416; binding plus continuation totals15,936. Prefix and continuation must not be added together as simultaneously active frames. Other recorded numeric, unit and parse_typed prologues are unchanged. These are static reservations, not measured total runtime stack use or a universal native-stack proof.

Read repeat_stack_control.py before inspecting its results; did not run it. All three recorded fresh subprocess invocations target exactly the unchanged language::analysis_source::mixed::project::ordinary_index::tests::small_stack_success_and_late_header_failure_leave_no_partial_owner with --exact --test-threads=1. Each raw log contains one pass, zero failures/ignores and268filtered; exit0. Each log hash and helper hash matches its receipt, and all three name the final focused artifact/source. The recipe checks clean source and absent stack override, stops on any failed attempt rather than retrying until pass, and verifies the artifact hash before/after the loop. These are three observed successful controls in this environment, not guarantees across arbitrary layouts/platforms.

The final completed full run also preserves every prior successful control by exact outcome matching. No test expectation was relaxed to turn the aborted original full run into acceptance. Existing16 failures, format failure and initial crash/debugger records remain part of the deliverable.

The reviewed result supports the private unverified-emission prerequisite and the separately reviewed parser frame repair. It does not establish actual-output parse/resolve/equality verification, before-allocation completeness, public mixed editing, persistence/history adoption, a merged follow-up branch or a new release.