Reading results

Logs has three tabs. Sessions shows every judge. Traces shows judges that run at trace level or finer. Spans shows span-level judges. Columns that could never fill on a tab are hidden.

The Sessions tab of Logs with numbered callouts on waiting, not evaluable, score, past data, and not backfilled cells — The Sessions tab. Session-level judges only appear here.

Session waiting. Sessions wait longer (30 minutes by default) because a conversation can continue. Score now is available here too.
Not evaluable. A roll-up: some traces in this session could not be scored.
92%. A session-level score.
Past data. Coherence was run over older sessions.
Not backfilled and Not sampled sit side by side for sessions from before a judge was on.

The Spans tab of Logs with numbered callouts on the judge columns, queued cells, and a pass cell — The Spans tab shows only span-level judges.

PII Leakage and Toxicity. Only span-level judges appear on this tab.
Queued. New spans wait for the judge worker, then fill in on their own.
Pass. One verdict per span.

What each cell means

You see	What it means	What to do
Pass, Fail, or a percentage	The judge scored it. Hover for the explanation.	Nothing. Results are final once produced.
Settles 6:10 PM and Score now	A trace or session is waiting for a quiet period before it is judged.	Wait, or choose Score now.
Scoring now	You chose Score now and the judge is running.	Wait a few seconds. The cell updates by itself.
Queued	Accepted, waiting for the judge worker.	Wait. Span judges usually finish within seconds to a few minutes.
Backfill running	A past-data run has not reached this row yet.	Wait.
Waiting for data	Some of the data the judge needs has not arrived.	Wait for the rest of the trace to be sent.
Retrying, a cause, and “Retry 2/5”	The judge call failed for a temporary reason. BotGauge tries again on its own.	Nothing, unless it keeps failing.
Failed and “Gave up”	Every retry failed.	Check the cause shown, then choose Score now if available.
Failed and “Needs attention”	Retrying can’t help. For example, the judge’s credentials were rejected.	Fix the cause, for example the judge key or model setup.
Not evaluable	The judge ran but could not decide, for example the input was missing.	Open the cell to read the reason.
Sample limit reached	The onboarding sample logs only run up to 5 different judges per project. This judge would be the sixth.	Nothing, or run it on your own data. Sample judges don’t use your quota.
Not sampled	The rule scores a percentage of rows and this one was not picked.	Nothing. Raising sampling later does not re-score it.
Not eligible	The row does not match the rule’s filter, or it is a span type this judge doesn’t grade, for example a tool call for Correctness.	Nothing.
Not backfilled	The row is older than the moment you turned the judge on.	Past data can only be added when a judge is first turned on. Duplicate the judge to include it.

When one cell covers many items

A trace or session summarizes its spans. Scores are averaged. Booleans fail if any span failed. Categories show the most common label. A cell with a mix of results shows how many could not be evaluated.

Read your evals in Logs

Open BotGauge and visit the Logs page to see pass, fail, and score cells for each judge.

Try BotGauge →

If a cell shows a cause, see Failures and troubleshooting. To change what gets scored, see Rules and settings.

Get started

Judges

Results

Reference

What each cell means

When one cell covers many items

​What each cell means

​When one cell covers many items

​Related

What each cell means

When one cell covers many items

Related