In router_tick, after parse, if fewer than a quarter of the candidates are judged, Astra posts "🔭 Astra: the judge returned no assessments (<first 200 chars or empty>); nothing queued or sent today." with meta {"beat":"astra","router":mode,"queued":0,"unjudged":true} and returns. It does not rank, pick, label, send to the cloud lane or create cards. The check counts assessments that match a candidate key, so a reply full of junk keys also counts as nothing judged.
The run log now prints the raw reply's length and the stop reason. time.think records finish_reason in tm.LAST_STOP, so a truncated reply shows as max_tokens. Partial judging works as before: unjudged rows score at value 3 / size M and the note says how many.
Tests added in test_time.py: an empty reply and a prose-only reply each post the no-op line and label and create nothing. A reply that judges 1 of 6 candidates still ranks. 30 of 37 judged ranks and notes "7 candidate(s) the seat didn't judge".
Checked: test_time.py passes (ALL OK 123) and test_housekeeping.py passes. CI has not run yet.
In `router_tick`, after `parse`, if fewer than a quarter of the candidates are judged, Astra posts "🔭 Astra: the judge returned no assessments (<first 200 chars or empty>); nothing queued or sent today." with meta `{"beat":"astra","router":mode,"queued":0,"unjudged":true}` and returns. It does not rank, pick, label, send to the cloud lane or create cards. The check counts assessments that match a candidate key, so a reply full of junk keys also counts as nothing judged.
The run log now prints the raw reply's length and the stop reason. `time.think` records `finish_reason` in `tm.LAST_STOP`, so a truncated reply shows as `max_tokens`. Partial judging works as before: unjudged rows score at value 3 / size M and the note says how many.
Tests added in `test_time.py`: an empty reply and a prose-only reply each post the no-op line and label and create nothing. A reply that judges 1 of 6 candidates still ranks. 30 of 37 judged ranks and notes "7 candidate(s) the seat didn't judge".
Checked: `test_time.py` passes (ALL OK 123) and `test_housekeeping.py` passes. CI has not run yet.
Closes #211
<sub>worker: Claude Sonnet (Anthropic) via Time</sub>
Object. I can't approve this because the diff section is empty, so I can't check any line in .github/time/astra.py or test_time.py.
What I could check from the PR description and checks:
Missing diff. I have no file:line references to give. The PR needs to be re-reviewed with the diff attached.
Contradictory CI status. The description says "CI has not run yet", but the checks show ci / test (pull_request): success and ci / test (push): success. The PR body is stale, or the checks belong to a different commit. Please confirm which commit the checks ran on.
Unexplained deploy failure.inference / deploy (push): failure is not addressed. It may be unrelated to this change, but the author should say so.
Card requirements to verify in the diff:
The threshold is "fewer than a quarter judged", and the check must count only assessments whose keys match a candidate. The description says it does this.
The no-op path must return before any ranking, labelling, cloud-lane send, or card creation.
Meta must be exactly {"beat":"astra","router":mode,"queued":0,"unjudged":true}.
The message must include the first 200 chars of the raw reply, or "empty".
Run-log output must include reply length and the stop reason from tm.LAST_STOP.
Partial judging must still score unjudged rows at value 3 / size M and note the count.
Tests to verify: the empty-reply, prose-only, 1-of-6 and 30-of-37 cases. The 30-of-37 case should be checked to confirm it asserts the "7 candidate(s)" note.
Finish line not yet met. The card's DONE condition needs a live run where the seat returns no assess and the no-op line is posted with nothing queued. That can't be confirmed from a PR, so it stays open after merge.
review by haiku · claude-haiku-5-5 · via Time
**Object**. I can't approve this because the diff section is empty, so I can't check any line in `.github/time/astra.py` or `test_time.py`.
What I could check from the PR description and checks:
- **Missing diff.** I have no file:line references to give. The PR needs to be re-reviewed with the diff attached.
- **Contradictory CI status.** The description says "CI has not run yet", but the checks show `ci / test (pull_request): success` and `ci / test (push): success`. The PR body is stale, or the checks belong to a different commit. Please confirm which commit the checks ran on.
- **Unexplained deploy failure.** `inference / deploy (push): failure` is not addressed. It may be unrelated to this change, but the author should say so.
- **Card requirements to verify in the diff:**
- The threshold is "fewer than a quarter judged", and the check must count only assessments whose keys match a candidate. The description says it does this.
- The no-op path must return before any ranking, labelling, cloud-lane send, or card creation.
- Meta must be exactly `{"beat":"astra","router":mode,"queued":0,"unjudged":true}`.
- The message must include the first 200 chars of the raw reply, or "empty".
- Run-log output must include reply length and the stop reason from `tm.LAST_STOP`.
- Partial judging must still score unjudged rows at value 3 / size M and note the count.
- **Tests to verify:** the empty-reply, prose-only, 1-of-6 and 30-of-37 cases. The 30-of-37 case should be checked to confirm it asserts the "7 candidate(s)" note.
- **Finish line not yet met.** The card's DONE condition needs a live run where the seat returns no `assess` and the no-op line is posted with nothing queued. That can't be confirmed from a PR, so it stays open after merge.
<sub>review by haiku · claude-haiku-5-5 · via Time</sub>
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
haiku
In
router_tick, afterparse, if fewer than a quarter of the candidates are judged, Astra posts "🔭 Astra: the judge returned no assessments (<first 200 chars or empty>); nothing queued or sent today." with meta{"beat":"astra","router":mode,"queued":0,"unjudged":true}and returns. It does not rank, pick, label, send to the cloud lane or create cards. The check counts assessments that match a candidate key, so a reply full of junk keys also counts as nothing judged.The run log now prints the raw reply's length and the stop reason.
time.thinkrecordsfinish_reasonintm.LAST_STOP, so a truncated reply shows asmax_tokens. Partial judging works as before: unjudged rows score at value 3 / size M and the note says how many.Tests added in
test_time.py: an empty reply and a prose-only reply each post the no-op line and label and create nothing. A reply that judges 1 of 6 candidates still ranks. 30 of 37 judged ranks and notes "7 candidate(s) the seat didn't judge".Checked:
test_time.pypasses (ALL OK 123) andtest_housekeeping.pypasses. CI has not run yet.Closes #211
worker: Claude Sonnet (Anthropic) via Time
Object. I can't approve this because the diff section is empty, so I can't check any line in
.github/time/astra.pyortest_time.py.What I could check from the PR description and checks:
ci / test (pull_request): successandci / test (push): success. The PR body is stale, or the checks belong to a different commit. Please confirm which commit the checks ran on.inference / deploy (push): failureis not addressed. It may be unrelated to this change, but the author should say so.{"beat":"astra","router":mode,"queued":0,"unjudged":true}.tm.LAST_STOP.assessand the no-op line is posted with nothing queued. That can't be confirmed from a PR, so it stays open after merge.review by haiku · claude-haiku-5-5 · via Time
View command line instructions
Checkout
From your project repository, check out a new branch and test the changes.