chore(openspec): archive four completed changes and write their specs

[tsenoner]

/opsx:archive is what makes openspec/specs/ true, and it had not been run for four changes whose work is long since on main — the oldest merged ten weeks ago. Until it runs, the behaviour those changes shipped is simply absent from the source of truth, so the next change reads openspec/specs/ and sees a repo that does not do things it has done since June.

Split out of #391 rather than folded into it: this writes six new capability specs into the source of truth, which deserves its own review rather than being buried in a 42-file docs cleanup.

Archived

Change	Merged to main	Capabilities added
add-eat-visualization	082b8b75	eat-annotation-overlay, eat-provenance-connectors
organize-predicted-annotations	ee707064 (10 wks)	annotation-metadata, annotation-presentation
optimize-e2e-suite	d15d73f0 (5 wks)	e2e-validation
fix-numeric-binning-e2e	adeb2dd1 (8 wks)	query-builder-filtering

Each was checked to be genuinely finished, not merely ticked: its last commit is an ancestor of origin/main, and no open PR still touches its change directory — the twelve open PRs each carry their own new change dir. All six capabilities are new, so none collides with an existing spec.

Ordering note: add-eat-visualization is archived ahead of #427, whose improve-eat-reliability-filter builds on eat-annotation-overlay and wants that capability to exist first.

Deliberately not archived

merge-protspace-monorepo (16/30) and unified-visibility-model (20/22). Both need reconciling against what actually shipped rather than archiving.

merge-protspace-monorepo is the clearer case. Its delta would land a bundle-contract capability directly alongside the bundle-format-contract that already exists, leaving two specs on the same subject with nothing to say which is authoritative. And its tasks.md is wrong in both directions: 4.3 archive the old repo is unticked but done (tsenoner/protspace-legacy is archived: true), while tasks 5.2–5.6 call for a packages/bundle-contract/ with a schema.json that does not exist anywhere in the repo — the contract work shipped by a different route, as bundle-contract.yml running pnpm test:contract on every PR shows. Archiving would freeze a superseded plan as current behaviour.

Purpose placeholders

openspec archive writes TBD - created by archiving change X. Update Purpose after archive. into each new capability, and that placeholder passes validate --strict — so it survives unless somebody looks. Two from an earlier archive (fix-pdb-annotation-without-uniprot) were still sitting there, which is how it goes unnoticed. All eight now describe what the capability actually covers, written from each spec's own requirements.

Verification

openspec validate --all --strict → 19 passed, 0 failed, re-run after prettier reformatted the spec markdown in precommit (it rewraps text, and --strict only scans a requirement's first line for the normative clause). openspec list now shows only the two genuinely incomplete changes. pnpm precommit green.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Ar6ME3o6VYLYsie3J4kxE5