# Wordmark r02: independent technical recheck

September 13, 2026. **The r01 P2 is technically closed for the tested regression.** No remaining finding was reproduced within this recheck's scope. This is not source-aesthetic, moving-quality, or native Resolve approval.

## Executed results

The documented r02 Python suite executed **16 tests**, including nine actual invalid-manifest CLI cases, the valid/omitted CLI controls, and five retained ordinary tests. I also executed **29 additional invalid-input CLI cases**. The Node test suite was not run.

All **38 invalid CLI cases exited 1**. Every requested `--out` directory and `verification.json` remained absent, stdout was empty, and the last error identified `--critic` validation. In particular, the original zero-byte, `{}`, and `[]` bypasses now fail before preparation/output creation. The preparation check was not silently skipped and reported as successful.

| Control | Exit | Actual rows | Research execution |
| --- | ---: | ---: | --- |
| Original frozen manifest supplied | 0 | **906** | 151 existing-source comparison frames per identity |
| `--critic` omitted | 0 | **604** | Source and stage comparisons explicitly `not-run` |
| Explicit invalid manifest, 38 cases | 1 | None | No completed output |

The additional cases covered whitespace, numeric falsey values, empty string, truthy non-object JSON, invalid UTF-8, truncated/trailing JSON, boolean schema/frame counts, wrong recipe/source/interval, missing readiness, null/extra movie records, empty/missing paths, wrong hash syntax/binding, wrong byte counts, wrong dimensions/cadence, audio metadata, and a nonexistent manifest file. These are executed cases, not claims of exhaustive fuzz coverage.

The **604 ordinary rows were identical** between both successful r02 runs and the corresponding frozen r01 ordinary rows. Both original/adaptation clips retained 1920×1080, 151 frames, constant 60/1 timing, no audio, and the 39-frame raw hold. The valid control checked the existing 960×540, 151-frame, silent source clip. No source metric was used to change review10 ties or award acceptance.

`verify.py:138` now validates an explicitly supplied manifest before preparation; subsequent source branches and row requirements use option presence. The omission control still has a null critic binding and `not-run` research, whereas the valid control binds the unchanged frozen manifest SHA-256.

## Exact command forms

Private paths are represented by aliases. `verdict.json` contains the executed argument array, exit, fixture hash, and output-creation result for **all 40 CLI invocations** (38 negative, two positive), plus the suite command. Literal commands and stdout/stderr remain in the private worker.

```bash
"$PYTHON_WITH_PILLOW" -B "$R02_CODE/test_verify.py" -v

"$PYTHON_WITH_PILLOW" -B "$R02_CODE/verify.py" \
  --work "$R01_WORK" --code "$R01_CODE" \
  --out "$RECHECK/tests/valid-frozen/proof" \
  --browser "$WORDMARK_BROWSER" --encoder-lock "$SHARED_ENCODER_LOCK" \
  --critic "$R01_WORK/critic-inputs.json"

"$PYTHON_WITH_PILLOW" -B "$R02_CODE/verify.py" \
  --work "$R01_WORK" --code "$R01_CODE" \
  --out "$RECHECK/tests/omitted/proof" \
  --browser "$WORDMARK_BROWSER" --encoder-lock "$SHARED_ENCODER_LOCK"
```

Effective suite bindings: `WORDMARK_WORK` → `$R01_WORK`; `WORDMARK_CODE` → `$R01_CODE`; `WORDMARK_CRITIC` → the frozen r01 manifest; `WORDMARK_TEST_OUTPUT` → `$RECHECK/tests`; `TMPDIR` → `$RECHECK/tmp`. `WORDMARK_VERIFY_SCRIPT` was unset. Pinned Node v24.13.1 and the supplied dependencies/full Chrome libraries were used; Python bytecode and Node compile-cache writes were disabled. FFmpeg and ffprobe processing used the supplied shared lock. No browser page or preparation was launched.

**Scratch-directory caveat:** one inherited Python unit explicitly creates and removes its empty temporary directory under `WORDMARK_TEST_OUTPUT`, not `TMPDIR`. That directory stayed inside the guarded new worker, but did not follow the narrower requested TMPDIR placement. No temporary directory or other new output was created in an input worker. This is an execution-scope caveat, not a verifier defect.

## Archive and patch

Both ZIPs were streamed, not extracted. The r02 archive has **279 members**: **272 unchanged**, five replaced, and two added; none removed. Every unchanged member matched r01, every replacement/addition matched the r02 source directory, and the generated package index matched all 278 payload members. The supplied patch exactly matched an independently recomputed unified diff of the seven affected members. All five delivery checksum entries matched.

- Replaced: `verify.py`, `test_verify.py`, `README.md`, `TESTS.json`, `PACKAGE_CONTENTS.json`.
- Added: `R02.md`, `R02-VERIFICATION.json`.
- `package_r02.py` is the separately supplied streaming packager, not an added ZIP member.
- r02 ZIP: **9,187,912 bytes**, SHA-256 `59d05a7917e431fa8a6d1991dbe84baaee3c0a522436de7864a271918eabcb41`.
- Patch: **42,152 bytes**, SHA-256 `35941e4e1721f1fb23ef0c20698039bc804700604e09da4100d6760ddc818f52`.

## Preservation and limits

Before/after checks cover **868 supplied files** across r01 code/worker, r02 code, and r02 delivery. File sets, SHA-256, byte counts, modification times, and modes are unchanged. `hashes.json` contains the complete root-relative input hash inventory and the frozen report identities. No source or repository file was edited; only the new recheck worker contains new files.

The global storage guard ran first with `--incoming-bytes 6291456`, without a punctuation argument, and passed again before testing/finalization. Any guard failure would stop writes rather than substitute a local ceiling. New evidence remained below the 3 MiB persistent / 6 MiB peak limits; final allocation is in the private seal receipt. No symlinks, dependency copies, downloads, media files, renders, or full ZIP extraction were created.

Source-footage approval **0**; moving-quality approval **0**; native Resolve approval **0**; source-critic resubmissions **0**. No stage/native comparison was run or credited. Source critics and aesthetic findings were not consulted; integrity hashing of frozen files is not a review of their contents. The valid manifest was only the required technical control. Closure is limited to this single-reviewer executable regression and packaging check, not an independent multi-reviewer certification.