# Batch04 public companion: frozen static findings and packaging boundary

This builder-composed companion carries the findings from the frozen batch04
source audit. It is not a new independent critic review or an approval of the
packaging repair. The originals, source-first freeze, label votes and seals are
unchanged.

## Frozen evidence

- `batch04-source/SOURCE_ONLY_PUBLIC.md`, SHA-256 `58556c8ed4b2d99b6db41c88521a372b75facc02b9b1d3175c590c7639b67e7f`.
- `batch04-source/REPORT-R01.md`, SHA-256 `7d58fcd1a646fa793c6c4273fdb15f242d452b99387446861562e8d433d36f27`.
- `batch04-source/REPORT-R01.json`, SHA-256 `2a4cb931a8033461d73b535ac630e818663f1762646de47e217d95c425b8a37a`.

R01 was sealed on September 14, 2026 at 00:49:36 UTC. Its coverage remains
**1,087 unique source frames**, **2,174 candidate frames**, and **0 extra context
frames**. The label pass also viewed 1,087 repeated reference frames; these do
not add unique source coverage. This packaging revision adds no inspected frames.

The observations are static only, not moving perception or a source-blind vote.
Default and adaptation remain tied on the frozen evidence. There are **0 moving
blind wins, 0 native Resolve proofs, 0 source-quality completions and 0 whole-film
credits**. Ten already-rendered candidate exports are retained, not re-encoded.

## Five unresolved gaps

These are the exact gap summaries in R01's proposed batch review fields.

| Treatment | Remaining gap |
| --- | --- |
| archive-outline-montage | Full-width bands replace the advancing, line-wrapping opening highlight. |
| type-staggered-identity | The identity labels no longer identify a visible person. |
| voice-toggle-to-orb | The departing composer's visible width and crop change. |
| sidebar-editorial-callouts | First-box stroke cadence and right-edge closure are displaced. |
| agent-slide-build | Three-panel growth phases converge later and remain too widely separated. |

R01 locates the archive gap at local f0–15, the voice withdrawal at f74–82,
the sidebar's first thin closure at source-local f137 versus candidate f140,
and the slide-panel phase gap around f150–169. Packaging does not repair these
visual differences. Resolve construction remains manual guidance, not a tested
native project or import.

## Packaging evidence is separate

At its freeze, R01 recorded three integration blockers: absent review wiring,
rejection of the first three author archives' documented input pairs, and
missing companion documents/helpers. R01 explicitly did **not** test any actual
downloaded ZIP. Its local runtime checks must not be relabeled as delivered-kit
verification.

The public-r02 configuration pins all five author archives. The first three
carry exactly both recipe-derived input variants as well as the unchanged
recipe/SVG payloads. All five carry their overlay manifests; the first three
also carry provenance, source inspection and the documented Python/Node helpers.
The Python helper's only public adaptation replaces literal privacy-prefix
assertions with an equivalent regular expression. The original helpers remain
unchanged; `guides/editorial-next/PUBLIC-TESTS-R01.json` binds original and public
helper hashes. The runtime helpers are unchanged author files.

Any packaging closure belongs to separate actual-ZIP assembly/verification
receipts: unpack the delivered core first and active archive second, reproduce
the handoff and control CSV, and execute every bundled helper against that
extracted project. Runtime checks use pinned existing dependency modules, with
no installation, encoding or native Resolve execution. This composite grants
no critic approval; an independent critic must inspect the builder's repair
and its receipts. Historical authoring statements about unrendered work and
pending reviews remain snapshots, not claims that the ten retained exports are
missing or that their source fidelity is accepted.