Public: anyone can read this; the crew can change it.
The release proof's log line and record say removed=true|false and why #12627
Check · Review · Land · Live done — Landed and live in production.
The release proof now says whether its probe forge was removed and, if not, why. The release proof log line adds removed=true|false and, when false, removereason="…", next to the existing ok= and tries=. Before this, a leftover probe forge only showed up on the needs-you card.
shipped production is serving it deploy e5f3b40
The release proof now says whether its probe forge was removed and, if not, why. The release proof log line adds removed=true|false and, when false, remove_reason="…", next to the existing ok= and tries=. Before this, a leftover probe forge only showed up on the needs-you card.
What changed
src/release/journey.ts: the removal step now reports a reason along with its true/false result.JourneyResultgainsremoveReason, set only whenremovedis false. The reasons are plain sentences:
- "the host said 503 to the removal" (whatever status the host gave) - "the removal timed out" - "the removal could not reach the host" - "no way to remove it was given" - "the host could not remove it" (in-process removal returned false) - "the removal failed on the host" (in-process removal threw) - "the name is not a probe forge's"
The reason comes from the last try after the existing backoff. The leftover sweep still only uses the true/false result.
src/release/prove.ts: theProofrecord hasremovedandremoveReasonwhenever a walk ran, taken from the last walk. The walk's ok/failed verdict is computed as before, so a failed removal still doesn't change it.
Tests
test/release-prove.test.tshas two new tests. Both would fail without the change, and each covers a removed and a not-removed walk: one checks the stored record, the other checks the exact log fields.test/release-journey.test.tsnow checks the reason in two cases: no token given, and a host that answers 503.bun scripts/gate.tsandbun x tsc --noEmitpass. The prove and journey suites pass (59 tests).scripts/fast.tssaid the diff was too broad to pick tests from, so I ran the otherrelease-*, leftover and probe suites myself. The only failure there istest/release-handover-keepalive.test.ts, which also fails with my change stashed, so this diff didn't cause it. I didn't look into why it fails; the forge's full check on the push will show whether it fails there too.
Left out
- The needs-you card doesn't show the reason yet.
- I didn't keep a plan on the issue: no
keep_planorestimatetool was available in this run.
🤖 Generated with Claude Code
The release proof now says whether its probe forge was removed and, if not, why. Before this, a leftover probe forge only showed up on the needs-you card.
What changed
src/release/journey.ts: the removal step now reports a reason along with its true/false result.JourneyResultgainsremoveReason, set only whenremovedis false. The reasons are plain sentences:
- "the host said 503 to the removal" (whatever status the host gave) - "the removal timed out" - "the removal could not reach the host" - "no way to remove it was given" - "the host could not remove it" (in-process removal returned false) - "the removal failed on the host" (in-process removal threw) - "the name is not a probe forge's"
The reason comes from the last try after the existing backoff. The leftover sweep still only uses the true/false result.
src/release/prove.ts: theProofrecord hasremovedandremoveReasonwhenever a walk ran, taken from the last walk. Therelease prooflog line addsremoved=true|falseand, when false,remove_reason="…", next took=andtries=.- Verdict unchanged: a walk whose forge was left behind can still pass, and a test checks this.
Tests
test/release-prove.test.ts: two new tests check the stored record and the exact log fields. Both would fail without the change, and each covers a removed and a not-removed walk.test/release-journey.test.ts: now checks the reason in two cases, no token given and a host that answers 503.
Check fix: the earlier check failed on four lines I added. The two Bun.sleep(0) waits in the new log-line test now carry // real-timer: notes, like the file's other waits. The two new doc comments in prove.ts are rewrapped to 100 columns. The branch is merged with main at b9ab715. bun scripts/gate.ts and bun x tsc --noEmit pass, and the prove and journey suites pass (59 tests).
Not checked
scripts/fast.tssaid the diff was too broad to pick tests from, so I ran the otherrelease-*, leftover and probe suites myself. The only failure there wastest/release-handover-keepalive.test.ts, which also fails with my change stashed, so this diff didn't cause it.
Left out
- The needs-you card doesn't show the reason yet.
- I didn't keep a plan on the issue: no
keep_planorestimatetool was available in this run.
🤖 Generated with Claude Code
Clean and small. The removal now returns a plain-sentence reason, Proof carries removed/removeReason, and the log line adds removed= and remove_reason=. Reasons hold only a status code or fixed text, so nothing leaks. A failed removal still stays out of the walk verdict, and the tests cover both removed and kept walks.
The release proof now says whether its probe forge was removed and, if not, why. The
release prooflog line addsremoved=true|falseand, when false,remove_reason="…", next to the existingok=andtries=. Before this, a leftover probe forge only showed up on the needs-you card.What changed
src/release/journey.ts: the removal step now reports a reason along with its true/false result.JourneyResultgainsremoveReason, set only whenremovedis false. The reasons are plain sentences:- "the host said 503 to the removal" (whatever status the host gave) - "the removal timed out" - "the removal could not reach the host" - "no way to remove it was given" - "the host could not remove it" (in-process removal returned false) - "the removal failed on the host" (in-process removal threw) - "the name is not a probe forge's"
The reason comes from the last try after the existing backoff. The leftover sweep still only uses the true/false result.
src/release/prove.ts: theProofrecord hasremovedandremoveReasonwhenever a walk ran, taken from the last walk. The walk's ok/failed verdict is computed as before, so a failed removal still doesn't change it.Tests
test/release-prove.test.tshas two new tests. Both would fail without the change, and each covers a removed and a not-removed walk: one checks the stored record, the other checks the exact log fields.test/release-journey.test.tsnow checks the reason in two cases: no token given, and a host that answers 503.bun scripts/gate.tsandbun x tsc --noEmitpass. The prove and journey suites pass (59 tests).scripts/fast.tssaid the diff was too broad to pick tests from, so I ran the otherrelease-*, leftover and probe suites myself. The only failure there istest/release-handover-keepalive.test.ts, which also fails with my change stashed, so this diff didn't cause it. I didn't look into why it fails; the forge's full check on the push will show whether it fails there too.Left out
keep_planorestimatetool was available in this run.🤖 Generated with Claude Code
Closes #12574