# Private S3 Windows evidence

Source `1b1c3e3abb6fc3d0934c9dbfbd21fddc8ae4f594`, following `90fae14818f684bf778640be2e613f34e3c7e999`. This checkpoint verifies the private typed index/rename slice and the rejected-depth syntax cleanup. **It does not establish current full-core runtime, public source8, integration, native UI or release acceptance.** Parent owns review/publication and integration.

| Recorded scope | Result |
| --- | --- |
| Full library listing | 167 test names; original 166 retained plus one cleanup test |
| Typed source runtime | 59 passed, including all 9 index and 5 update tests |
| Existing raw-token control | 1 passed |
| Three focused 128 KiB controls | Original whole-flow, eight-parent cleanup and synthetic 256-frame Path tests passed |
| Strict core, all targets | Clippy with `-D warnings`, exit0 |
| Original90fae typed gate | Stack overflow retained; aborted run is not a passing full gate |
| Temporary diagnostics/probes | Two dirty overflow reproductions; four dirty direct-parser malformed-input probes passed |

Actual platform was Windows x86_64-pc-windows-gnullvm, Rust/Cargo 1.98.1, one compiler job. Every clean gate binds 538 committed inputs before/after, exact commands and source. Raw source manifests, logs, receipts, helper scripts and diagnostic diffs remain byte-for-byte unchanged in [evidence.zip](evidence.zip). Its SHA256 is `4486f4c42b140ae4b9b3caf0aacdf151a2acb1fd85eda09a57764fd69e980574`; 658251 bytes, 61 members. No executables, caches or profiles are included.

The failure occurred when `Cursor::node` recursively dropped a rejected depth129 syntax tree below parser frames. Complete depth128 parsing/indexing/renaming/output/model equality/admission and earlier cleanup had finished. The source fix uses iterative rejected-kind disposal and preserves the existing stack request, depth limit and assertions. The 14 original S3 tests and the added mixed-shape cleanup test now have actual runtime evidence. Ordinary AST destruction and all possible parser-error paths are not claimed iterative or universally stack-safe.

## Reproduce evidence verification

With Python 3 standard library and Git, from any working directory:

```text
python -B PATH/verify.py --repo PATH/TO/GIT/REPOSITORY
python -B PATH/negative_controls.py --repo PATH/TO/GIT/REPOSITORY
```

The repository needs immutable commits `1b1c3e3`, `90fae148` and `1f828323` (the publication commit contains that ancestry). Verification reads commit blobs; it neither depends on nor changes the current checkout. `--evidence-dir PATH` optionally selects another complete extracted payload directory. No archived helper, compiler or native artifact is executed. Optimized Python is refused.

The verifier checks both manifests/ZIP byte integrity, actual listing and exact selected runtime membership, 538 committed inputs, recorded commands/times/environment, strict completion, historical source-artifact manifests and the original aborted gate. It reconstructs the dirty source patches in memory and verifies the corresponding 538 test-input hashes. Broader tracked-file hashes also remain in the original captures; the verifier does not claim to audit unrelated tracked files outside that test-input scope. Binary identities are checked against raw executable-path receipts, not rehashed executable bytes, because binaries are deliberately absent. Initial focused tests and final restored typed/token/listing tests were separate builds of identical source and have their own binary receipts.

[checks.json](checks.json) records the positive working evidence audit and 13 discriminating negative controls. Those controls corrupt evidence only in memory; raw archives stay intact. Semantic cases reach their specific checks instead of failing only at the outer ZIP checksum. Package verification also runs against the saved commit and a fresh extraction; exact publication-commit readback is reported in the handoff to the parent, since embedding a commit's own hash in its contents is not possible.

## Review and historical evidence navigation

Extract the ZIP only if desired; the verifier itself inspects it in memory. Current authoritative raw closure is `closure.json` plus `raw/`. Relevant retained notes inside the ZIP:

- `notes/analysis-source8-s3-index-independent-review-20260913.md`: original S3 source review and required corrections.
- `notes/analysis-source8-s3-correction-independent-review-20260922.md`: corrected Path/index source review before runtime.
- `notes/analysis-source3-stack-cleanup-review-20260922.md`: independent review of iterative rejected-kind cleanup.
- `notes/analysis-source3-stack-cleanup-handoff-20260922.md`: exact failure isolation, fix and completed Windows gates.
- Original index/correction source manifests bind their respective immutable Git versions; their historical `unrun` labels are preserved and superseded only by the later actual receipts.

The ZIP's initial `README.md` contains the earlier portable-verifier proposal. It is retained as historical package text; this outer README and actual `verify.py` describe the completed verifier. No peer should execute the archived local orchestration helpers to verify evidence.

Source-bound original-span export, mixed ordinary/typed composition, complete output recompile/admission/intended-Project equality and public dispatch/migration/save integration remain separate work. This evidence does not close those requirements.