Files
fusion/packages
gsxdsm 9c1c6f7479 docs(engine): close the unproven-sites ledger — one entry was wrong, the rest need two named lanes (#2544)
Comment-only change to the ledger. No test or production code moves.

## Why this is a PR and not a note

The ledger is the artifact that keeps *"the E2E covers the conversion"*
honest. It gets the same treatment as the code: claims verified by
mutation, not by reading.

## Correction: one entry was wrong

`core/task-store/reads.ts` was listed as **unproven**. It isn't. Core's
`store-stale-paused-renamed-hold.pg.test.ts` is a real-store test that
drives `listTasks` against a renamed hold column — and forcing the
hydration back to the `todo` literal **fails exactly that file's renamed
case**.

I had listed it as unproven because I assumed a separate E2E was needed.
Verified *before* removing it, since "already covered somewhere else" is
precisely the assumption that lets a gap hide.

## What remains, and why it is not another table row

**Lane 1 — real git.** `merger.ts`'s `resolveMergerLifecycleColumn` and
`executor.ts`'s `resolveReboundColumnFor` are module-private helpers
whose only callers sit inside merge/session machinery needing a real
worktree, branch and squash; `merger-ai.ts` is the same. Re-checked with
the lens that freed `auto-merge-finalization` and both self-healing
rebounds — **these genuinely need the lane.** The earlier over-broad
claim doesn't retroactively excuse them.

**Lane 2 — dashboard HTTP.** The four `register-task-workflow-routes`
sites sit behind `registerTaskWorkflowRoutes(ctx, deps)`, needing a full
`ApiRoutesContext` plus twelve injected deps. Standing that up is the
mock-the-world shell FN-5048 says not to add. The narrower alternative —
exporting the two private resolvers — yields **unit** evidence while
looking like E2E.

Deliberately not done rather than done badly and overclaimed.
`live-agent-count`'s `columnIsIntakeOrHold` is the same lane: only the
*waiting* predicate reads it, and its consumers are dashboard-side.

## Running total

**10 of 15 census sites proven end to end** across six suites; 5 remain,
each named with the lane it needs.

## Verification

- lifecycle suite 20/20; engine `tsc --noEmit` clean; `pnpm test:gate`
green (414 + 10 + 71)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Tests**
* Updated end-to-end test coverage documentation to accurately reflect
verified workflow and task-store behavior.
* Clarified coverage gaps for live agent-count logic and dashboard
workflow routes.
  * Added a two-lane breakdown describing remaining coverage work.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-07-29 20:56:04 -07:00
..
2026-07-26 18:11:47 -07:00