# Audit evidence bundle

Read the
[audit and canonical plan](../EVE_TASK_BACKLOG_RATIONALIZATION_2026-09-19.md)
first. This bundle is a review artifact, not a second executable task board.

The source snapshot is commit `e6a64111e3abb96af5a89d194d5c2c8c2d023724`. Every
open key maps to exactly one primary C01–C28 execution group in
`task-dispositions.tsv`; cross-group dependencies are described in the report.
Mapping is by reviewed logical group and explicit source/ID rules, not a claim
of individual implementation verification. Existing states and blockers are
recorded verbatim. No proposed disposition changes a board row.

`file-inventory.tsv` includes every checkbox-bearing file, including proposals,
inventories, templates and checked-only trackers. `document-catalog.tsv`
contains committed catalog entries other than generated slide/Docs Center
entries, so zero-checkbox plans remain discoverable. A catalog entry is not
automatically an adopted requirement or a freshly reviewed document.

`repeated-wording.json` is a conservative lexical screen of the first physical
checkbox line after removing its mark/leading identifier. Wrapped sentences can
share that line without being duplicates. The findings are candidates, never
automatic deletion decisions. `code-markers.tsv` similarly includes quoted
examples and fixtures and labels its confirmed runtime observations.

Run from the repository root, with a Node version supporting `node:sqlite`:

```sh
NODE_NO_WARNINGS=1 node docs/audits/eve-task-rationalization-2026-09-19/inventory.mjs --check
```

`--write` mechanically regenerates only the six evidence files in this
directory. Both modes read the board and catalog with SQLite `readOnly: true`,
assert unchanged board bytes, and reconcile all open task counts. A changed
upstream board should fail snapshot checking; it requires an explicit new audit
snapshot, not silently relabeling this one. The Markdown discovery count
excludes this report and its bundle.

The report's dispositions remain a proposal until adopted. In particular,
`retire-reimplementation-keep-proof` preserves missing acceptance tests,
`attach-acceptance` preserves per-route/per-slide evidence, and
`retire-duplicate-sweep-after-crosswalk` requires retaining any unmatched
obligation. Parent roll-ups and child work are not both counted as independent
deliverables in the proposed execution model.