Remove five documents nothing reads

[vrv-panw]

The repository is 32 tracked files and all seven scripts are live, so the dead weight was documents rather than code. Tracked size 1.5 MB → 868 KB.

Removed

File	Why
PROSE-INVENTORY.csv	677 KB, 44% of the tracked repository. Frozen at the end of Phase 1 because it could only be derived by diffing against the base, and 3,618 of its 4,221 pointers dangled — they named operations and components the drops deleted. It stored digests and lengths, never text, so nothing was recoverable from the 603 that still resolved. The worklist it was meant to be is overlays/docs-prose.yaml, which has one stub per authoring site and is checked on every build
brief.md	The original handoff. Referenced nowhere; superseded by the README and PLAN.md
_project/docs-note.md	The five docs asks, all completed in Phase 1.5 and recorded there
_project/drop-list.md	A generated inventory nothing regenerated. It restated planes.yaml, and its job — being the tick-list for choosing the drops — ended when the choices landed in drops.yaml
_project/gen_drop_list.py	Generated the above

The drop list is the one worth a second look. It was not stale when removed, but nothing in CI regenerated it and it had already drifted silently once — Phase 1 task 7 found it reporting 95 unclassified paths instead of 23, without failing. A derived file no gate checks is a file that will eventually lie. The alternative was to wire it into build.py --check; removing it was chosen because planes.yaml already carries the content under a gate.

Kept, deliberately

• The KB machinery — scripts/drift.py, scripts/emit_change_event.py, webhooks/*.schema.json, kb-drift.yml, notify-kb.yml. On hold, not cancelled, and built to stay exercised while inert.
• _project/classification.yaml — marked historical, but still a live reference target: planes.yaml cites it by name in the how: field of 23 paths.
• PLAN.md and TODO.md whole. A cleanup pass is the wrong place to decide which reasoning is worth keeping.

Reference fixes

So nothing points at a deleted file: the PROSE-INVENTORY.csv row and paragraph in the README, the drop-list.md mention under Known gaps, a stale "five files under _project/" that was already wrong (there are four decision files), and comments in build.py, drops.yaml and classification.yaml. The only surviving mention of a deleted file is a deliberate historical note.

Verification

check.py, build.py --check and lint.py are green either side of the change. Nothing published changed — openapi.yaml, docs-navigation.json and build-report.txt are byte-identical.

This PR also gives breaking-changes.yml a run, which is the first item in TODO.md's Residue. It does not. The workflow is filtered to paths: ['openapi.yaml'], and this PR does not touch that file, so it was correctly skipped. The Residue item stands unchanged: the workflow has still never graded a real diff, and getting that audit trail needs either a workflow_dispatch run against the pre-drop revision or the next drop batch landing through a PR.

🤖 Generated with Claude Code