Source 18073a6f2e, pushed and ref-verified on codex/analysis-mixed-rename-plan-956atf-20260924. The accepted contract is implemented by a private consumed-source plan borrowing one whole desired Project. It validates every non-name field, simultaneous global/scoped names, the complete original-coordinate edit set, current bundle identity and combined output sizes. It does not emit or admit rewritten source, activate public source8 or integrate into the published Windows preview.

Windows gnullvm, rustc/Cargo 1.98.1: 17 new tests pass, and strict core all-target Clippy passes. The actual listing contains 246 names. The full library remains FAILED: 230 passed, 16 failed; all 229 prior outcomes and all 16 failure bodies are unchanged (only numeric thread IDs normalized). Formatting remains FAILED, byte-identical to the eight inherited blocks in five files. Linux/WSL checks were not run.

Independent runtime review verifies all 551 declared inputs against immutable Git and working bytes, exact named outcomes, launchers and the 42,081,792-byte test executable, SHA256 2c8bd7b9a6adfda3f74b74bae00453b1b1348cbed47f04c8e8c1de9152899416. Source review applies to unchanged production rename bytes; the module-wiring file was subsequently normalized to LF, SHA256 bd0cea93721a23cb2aaad1f1cb1893b45a5ba38dc6067a0c99d6a05c9bab42dd. Test review and the independent oracle cover complete OA/SA and Unicode/CRLF maps, scoped swaps, all 36 ordinary-category edits, historical exclusion, same-spelling wrong-callee corruption, unsupported semantic changes and signed-zero equivalence. No expectations were generated from planner output.

Actual inherited entry caps B+37/B+38 clear all predecessor phases and separate the final reservation refusal from success retaining R+M=38 entries. The included-file variant preserves its precise diagnostic. With a 930-byte input cap, ordinary alias shrink offsets typed growth to a valid complete 924 bytes; the 933-byte variant refuses, counting the untouched file. These are planning sizes, not emitted-output acceptance.

The stack investigation preserves three initial focused aborts, two marker runs, the original artifact's explicitly filtered 16-test pass, two prologue observations and three debugger runs. LLDB located an overflow in parsing/exact-number conversion before the planner. One debugger run passed the same artifact, so no precise cause for environmental variation is asserted. The final evaluator prepares source/desired prerequisites on the normal test thread, then moves the complete source owner into a 128 KiB scoped worker for planning and destruction. All six deep/wide success/collision/ non-name cases pass. This meets the planner-specific small-stack contract; the prior full-pipeline 128 KiB setup failure remains unresolved.

evidence.zip contains 76 members, 584,235 bytes, SHA256 1c041d181540b474e14e80677cfe287d305951eaee48e492de7f59dcdea2391c. manifest.json pins every member. Raw initial/final gates, baseline records, exact launchers, diagnostic scripts/logs, source differences, oracles and reviews are preserved. The nine readable review/checker artifacts mirrored from the archive retain exact bytes; the critique navigation page links to its earlier authoritative home. Toolchain version text is the previously captured, unchanged local toolchain record, not a newly executed version probe.

Reproduce subject gates from the exact source with the project Windows toolchain, one compiler job, CARGO_INCREMENTAL=0, and no lifecycle or RUST_MIN_STACK override:

cargo test --locked --offline -p oa-core --lib -- --list
cargo test --locked --offline -p oa-core --lib language::analysis_source::mixed::project::ordinary_index::rename::tests:: -- --test-threads=1
cargo test --locked --offline -p oa-core --lib -- --test-threads=1
cargo clippy --locked --offline -p oa-core --all-targets -- -D warnings
cargo fmt --all -- --check

Raw replay helpers retain the producing machine's directory layout and must be read before reuse. Receipts bind declared inputs and recorded environment fields; they are not exhaustive environmental capture or authenticated build provenance. The readback checker does not invoke Cargo or the executable. Do not treat a debugger's process exit zero as a passing ordinary test.

Actual emission/readmission with carried output storage, scalar/structural editing, public source8, persistence/history and graphical/API parity remain required work under issue7. The selected successor direction and rejected envelope approach are recorded in DECISIONS.md.