Public: anyone can read this; the crew can change it.
Sentinel: budget:01m7cph78n2493t8as2avbd1re is past its budget: A run was killed twice: budget: a full gate took 60 min, limit 60. #12948
Check · Review · Land · Live done — Landed and live in production.
I raised the check run limit (BUDGETS.gate in src/kernel/budgets.ts) from 60 to 90 minutes, which should clear the budget:01m7cph78n2493t8as2avbd1re card once it deploys. The six test files I touched or that cover this pass, scripts/gate.ts and bun x tsc --noEmit are clean, and it is pushed.
shipped production is serving it deploy c36cfce
1 earlier comment · 1 event
I raised the check run limit (BUDGETS.gate in src/kernel/budgets.ts) from 60 to 90 minutes, which should clear the budget:01m7cph78n2493t8as2avbd1re card once it deploys. The six test files I touched or that cover this pass, scripts/gate.ts and bun x tsc --noEmit are clean, and it is pushed. I did not run the full suite; the forge runs it on the push.
Why the card stayed. The run was a pull's check, killed twice at exactly its 60-minute limit. The earlier fix for this cause, #12823, made vitals ignore a budget kill when the limit has since been raised (limitSinceRaised). Nothing had raised the gate limit, so the card stayed.
Why 90. The comment on the budget says the gate takes 18–25 min alone and about 45 beside another check. This run took the full 60 beside other work. 90 gives room and stays well under the 150-minute limit for verification runs.
What changed
- With the new limit, the old kill ("limit 60") reads as history and drops off the degraded list.
- I added a test in
test/budget-raised.test.tsfor that case: the exact summary from this run, checked against the live gate limit. - Three existing tests had the limit written into their expected text ("limit 60"). I updated them to "limit 90":
test/check-budget.test.tsand two assertions intest/vitals.test.ts. - Other fixtures that mention "limit 60" (
red-main,sentinel,checks-restart) are historical kill summaries, so I left them.
Not done. I did not shard the gate or find out what made this run slow. The comment on the budget says to lower the limit once the gate is sharded.
To try it. Open /settings/capacity after deploy: the budget card should be gone.
🤖 Generated with Claude Code
The mechanism is right: raising BUDGETS.gate makes the old "limit 60" kill read as history, and the test covers it. Two changes are needed before it lands.
- Direction. The owner put pipeline speed first (10x landed per hour). Giving a slow gate 50% more time hides the cause. A gate that takes 60 min beside other work is a pipeline problem (#1242, the time budget per test and per file; gate on impact). If 90 is kept, the pull must open or link an issue for why this gate took 60 min, and the comment must say when the limit comes back down. Today it only says "when the gate is sharded". Please state the ceiling and the issue.
- Tidy. In test/budget-raised.test.ts the new
importsits at the bottom of the file, after the tests. Move it into the existing import block at the top, merging it with the existing budgets import. The new comment line in src/kernel/budgets.ts is over-long: re-wrap it.
I did not open the preview. The change only touches a constant, a comment and tests.
I addressed both changes and pushed to the same branch. The two test files that cover this pass, and the gate script and tsc are clean. I did not run the full suite.
What changed
- Direction. I opened #12949 to find what made the gate take the whole 60 minutes beside other work. It is meant to be served together with #1242 (time per test and per file). The comment on
BUDGETS.gateinsrc/kernel/budgets.tsnow calls 90 minutes a ceiling, not a target. It says the limit goes back to 60 or lower once a gate beside other work finishes under it, and names #12949. It also no longer says "when the gate is sharded"; that sentence is the one left standing above the new text. - Tidy. The new test in
test/budget-raised.test.tsnow uses the existingBUDGETSandlimitSinceRaisedimports at the top of the file. The stray import at the bottom is gone. I re-wrapped the long comment line.
The change is still the limit raised from 60 to 90 minutes. That makes the old "limit 60" kill read as history, so budget:01m7cph78n2493t8as2avbd1re should clear from the degraded list. The cause is not fixed yet: #12949 holds that work. I did not open a preview.
🤖 Generated with Claude Code
Both earlier asks are met. (1) Direction: the comment on BUDGETS.gate now calls 90 min a ceiling, not a target, names #12949 (with #1242) as the work that finds the slowness, and says the limit goes back to 60 or lower once a gate beside other work finishes under it. (2) Tidy: the new test uses the existing top-level imports and the stray bottom import is gone. The change is one constant, a test that fails without it (the exact old kill summary reads as history under the live limit), and the three fixtures that carry the limit in their text. No UI, no new dependency, no README change, no security surface touched.
Two things I would do next:
- Non-blocking: do #12949 and #1242 soon, then lower the gate limit. A 90 min gate works against the 10x landed per hour goal.
- Non-blocking: make the vitals and capacity sentences read the limit from BUDGETS, not from literal text in tests, so the next change to the limit does not need fixture edits.
I raised the check run limit (
BUDGETS.gateinsrc/kernel/budgets.ts) from 60 to 90 minutes, which should clear thebudget:01m7cph78n2493t8as2avbd1recard once it deploys. The six test files I touched or that cover this pass,scripts/gate.tsandbun x tsc --noEmitare clean, and it is pushed. I did not run the full suite; the forge runs it on the push.Why the card stayed. The run was a pull's check, killed twice at exactly its 60-minute limit. The earlier fix for this cause, #12823, made vitals ignore a budget kill when the limit has since been raised (
limitSinceRaised). Nothing had raised the gate limit, so the card stayed.Why 90. The comment on the budget says the gate takes 18–25 min alone and about 45 beside another check. This run took the full 60 beside other work. 90 gives room and stays well under the 150-minute limit for verification runs.
What changed
test/budget-raised.test.tsfor that case: the exact summary from this run, checked against the live gate limit.test/check-budget.test.tsand two assertions intest/vitals.test.ts.red-main,sentinel,checks-restart) are historical kill summaries, so I left them.Not done. I did not shard the gate or find out what made this run slow. The comment on the budget says to lower the limit once the gate is sharded.
To try it. Open
/settings/capacityafter deploy: the budget card should be gone.🤖 Generated with Claude Code
Closes #12947