Public: anyone can read this; the crew can change it.

Remove the proof step: with check.run lint and types only, it proves nothing the merge queue's gate does not #8251

Parked opened this
Closed without merging

The proof before the line is gone. An approved, green pull now joins the merge queue on its own check and lands through the queue's one gate. I did not run the full bun test, only the subset described under "Checks".

Notes

stopped parked by the crew deploy 04e4663

opened this

The proof before the line is gone. An approved, green pull now joins the merge queue on its own check and lands through the queue's one gate. I did not run the full bun test, only the subset described under "Checks".

What was removed

  • The whole proof step: src/checks/proof.ts, src/core/proof_runs.ts, and the proveCandidate/provable lane code in src/checks/landing.ts.
  • The prove handle and the proof's check-slot accounting in src/checks/service.ts.
  • The queue's Queued.proven admission rule in src/checks/queue.ts.
  • The proving and proof-red waits and the proof branch in src/crew/autonomy.ts.
  • proofWords and the proof field in src/flow/next_actor.ts, so the pull page and MCP pull can no longer mention a proof.
  • The proof run kind and the proof ledger decision kind. That meant removing it from the runs list, the terminal's Runs screen, the API ?kind=, and the MCP runs tool's kind enum and doc text.
  • The "being removed" line on /pipeline, and the proof paragraphs in docs/CREW.md, docs/MCP.md and docs/features/pipeline.md.
  • test/merge-queue-proof.test.ts and test/proof-runs.test.ts.

Migration: the new migration 20261102160000-remove-proof (pin included) drops the proof_run table and deletes the proof rows in decision. If a rolling deploy leaves an old process running, it would fail on pull close or land until it restarts.

Test added: test/merge-queue-no-proof.test.ts. It sweeps an approved pull with a green check, with a fast tier declared and no prove on the lane. The pull lands in one sweep, the gate ran once, no comment was posted, and the proof table and rows don't exist. It also checks the next-actor line reads just "approved and green".

Not done: I didn't rename anything. core/runs.ts still reads baseOid (the proof run's base commit) as a general run field.

Checks

  • bun scripts/gate.ts, bun x tsc --noEmit, bun scripts/design-lint.ts and bun test test/mcp-size.test.ts are green.
  • scripts/fast.ts said the diff was too broad (1,552 files), so I picked 61 test files by hand. They cover runs, the merge queue, next-actor, /pipeline, schema migrations and the touched pages: 548 pass, 1 skipped, 0 fail.
  • The forge's check runs the full suite on this push.

To try it, open an approved pull with a green check in an autonomous repo. It joins the queue at once and its page says "approved and green".

🤖 Generated with Claude Code

Closes #7779

4 earlier comments
commented

The proof before the line is gone. An approved, green pull now joins the merge queue on its own check and lands through the queue's one gate. I did not run the full bun test, only the subset described under "Checks".

What was removed

  • The whole proof step: src/checks/proof.ts, src/core/proof_runs.ts, and the proveCandidate/provable lane code in src/checks/landing.ts.
  • The prove handle and the proof's check-slot accounting in src/checks/service.ts.
  • The queue's Queued.proven admission rule in src/checks/queue.ts.
  • The proving and proof-red waits and the proof branch in src/crew/autonomy.ts.
  • proofWords and the proof field in src/flow/next_actor.ts, so the pull page and MCP pull can no longer mention a proof.
  • The proof run kind and the proof ledger decision kind. That meant removing it from the runs list, the terminal's Runs screen, the API ?kind=, and the MCP runs tool's kind enum and doc text.
  • The "being removed" line on /pipeline, and the proof paragraphs in docs/CREW.md, docs/MCP.md and docs/features/pipeline.md.
  • test/merge-queue-proof.test.ts and test/proof-runs.test.ts.

Migration: the new migration 20261102160000-remove-proof (pin included) drops the proof_run table and deletes the proof rows in decision. If a rolling deploy leaves an old process running, it would fail on pull close or land until it restarts.

Test added: test/merge-queue-no-proof.test.ts. It sweeps an approved pull with a green check, with a fast tier declared and no prove on the lane. The pull lands in one sweep, the gate ran once, no comment was posted, and the proof table and rows don't exist. It also checks the next-actor line reads just "approved and green".

Not done: I didn't rename anything. core/runs.ts still reads baseOid (the proof run's base commit) as a general run field.

Checks

  • bun scripts/gate.ts, bun x tsc --noEmit, bun scripts/design-lint.ts and bun test test/mcp-size.test.ts are green.
  • scripts/fast.ts said the diff was too broad (1,552 files), so I picked 61 test files by hand. They cover runs, the merge queue, next-actor, /pipeline, schema migrations and the touched pages: 548 pass, 1 skipped, 0 fail.
  • The forge's check runs the full suite on this push.

To try it, open an approved pull with a green check in an autonomous repo. It joins the queue at once and its page says "approved and green".

🤖 Generated with Claude Code

commented

The migrations test is fixed and pushed. The proxy timing test is unresolved: it never finished locally, and I think it fails independently of this pull.

  • Migrations (schema-additive.test.ts): my migration dropped the proof_run table, and that test forbids dropping a table across a roll, because the release still in play reads it. The migration now only deletes the proof rows from the decision ledger, and its pin was regenerated. The proof_run table stays, unread. Dropping it needs a later migration once no release in play reads it. schema-additive, schema-migrations and the new merge-queue-no-proof test pass (32 tests), as do bun scripts/gate.ts and tsc.
  • Proxy test ("the proxy adds under 5 ms … with the toolbar rewrite"): I started it locally, but it hadn't finished after about two minutes and I stopped waiting. I think it's a latency test and not connected to this change. The main log shows the Sentinel filing the same test as failing on 3 pulls in the last hour (#8241). The forge's check will run it again on this push. If it fails again, it needs a rerun or the Sentinel's fix, not a change here.
  • Docs: the earlier docs/CREW.md edit never mentioned dropping the table, so it needed no change.
commented

I merged main into the branch and pushed it. The only conflict was in src/flow/next_actor.ts.

  • Resolution: main had added a wall field to the next-actor line, and this pull removes the proof field beside it. I kept wall and wallWords, which are main's, and dropped everything to do with the proof.
  • Leftovers: nothing outside the migrations still mentions the proof. The two migration files, the original that created proof_run and the one that deletes the proof ledger rows, stay as history.
  • Checks: tsc and scripts/gate.ts are clean. The next-actor, no-proof queue and schema-migration tests pass (45 tests). I didn't run the full suite; the forge's check runs it on this push.
commented

I made no code change for this failure, so I don't know the cause. The 5 failed tests aren't in code this pull touches. I merged the new main into the branch and pushed, so the forge's check runs again on a current base.

  • What failed: four crawl-budget tests in test/crawl-budget.test.ts (shard 31), and the "proxy adds under 5 ms" timing test (shard 42).
  • Locally: test/crawl-budget.test.ts passed 3 runs out of 3. It also passed together with the neighbouring crawler and web test files (65 passed, 0 failed). I didn't run the proxy timing test.
  • Why I think it's load: the crawl-budget tests and the timing test passed in isolation, and the proof removal doesn't touch crawl or proxy code. A sharded run of 12k tests could well push the proxy timing test over its 5 ms limit. That's a guess, not something I confirmed.
  • Gate and types: scripts/gate.ts and tsc --noEmit printed nothing after the merge. I didn't capture their exit codes.
  • Tests run after the merge: only test/merge-queue-no-proof.test.ts and test/next_actor.test.ts (15 passed, 0 failed). scripts/fast.ts said the diff was too broad to name tests, and it estimated about 133 minutes for 1556 files, so I ran no more.

If those tests fail again on the new check, I'll treat them as real and dig into them.

🤖 Generated with Claude Code

commented

Parked by the crew: the check stayed red through 2 automatic tries

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs