Agent performanceSections

Agent performance

How long execution takes, where attempts fail, and which items repeat.

Model time · p50
2m 43s
p90 4m 45s · n=135,811
Failed attempts · 60m
1
0 terminal · 1 retryable
Repeated claims · 24h
0
items with >24 claims

Review run time · 30d

Lower is faster. If p90 rises while p50 stays flat, investigate the slower runs.

0s3m 20s6m 40s10m 00s13m 20s
Oct 17Oct 23Oct 29Nov 3Nov 9Nov 15
p50 · typical runp90 · slower runs

30 of 30 complete days with timings · Actions through Nov 16, 14:12 UTC. Model-only time above uses a separate latest-review snapshot.

Where attempts fail

  • Source preparation1100.0%

Start with the largest stage, then check retry and backoff reasons.

1 affected targets · last 60 minutes · snapshot Nov 16, 14:13 UTC

Repeatedly claimed items

Showing 0 of 0 flagged targets. Bounded public sample, at most 10; other targets may be outside the sample or ineligible for public display. No private-target count is inferred.

No targets over the claim threshold.

Claims are not completed reviews. Revision, trigger, model execution and next retry are not recorded by claim history. Workflow/job URLs are unavailable; item links are canonical GitHub references.

Repeated claims alone do not prove wasted execution. Counts and recent history appear only when observed; revisions and triggers remain unknown.

esc