Monitoring
Monitors run saved code on a cadence against logs and events — provisioning results, job runs, API latency, storage usage. Classifications decide, record by record, whether to act, review, or pass.
Monitor · Employee IT workspace
Account provisioning failures
Counts app assignments that failed for new and changed accounts, per app, over a rolling 24-hour window.
Runs · 14 days
1,344
every one executed
Records examined
18,210
41 in the last run
Telemetry lag
4 min
inside one cadence
Last raised
22h ago
3 failures in Slack
Failure needs IT action
Operational triage1 act2 review38 passExpected retry, self-heals
Control verification0 act0 review41 passRecord · 22h ago
Slack assignment failed for a new hire: license limit reached
Failure needs IT action
Verdicts
Every run ends in one of three verdicts, and each one keeps the records that produced it.
Clear
The monitor examined its records and found nothing that meets its condition.
Raised
The condition was met. The run keeps the matching records for classification and follow-up.
Undetermined
The evidence couldn't decide — no records, stale telemetry, or a failed read. Reported exactly that way.
Classifications
Classifications read the records a monitor keeps and decide what each one needs, with a confidence — so people spend time on the one record that matters, not the forty that don't.
Classification · Latest run
Failure needs IT action
on Account provisioning failures · 41 records
Slack · license limit reached for a new hire
confidence 0.91
Jira · group mapping missing
confidence 0.62
Zoom · SCIM request timed out
confidence 0.55
Google Workspace · retry succeeded
confidence 0.97
Okta · duplicate assignment ignored
confidence 0.94
GitHub · seat already assigned
confidence 0.93
+35 more
Each record gets a decision and a confidence — kept with the run it came from.
41 → 1
records examined, one that needs a person
Schedule
See every run coming in the next 24 hours, and move new monitors from draft to preview to published when they're ready.
Monitoring
Next 24 hours
3 active monitors · 11 drafts in flight
Account provisioning failures
Every 15 minutes
API p95 latency within budget
Hourly
Nightly warehouse load finished by 06:00
Daily at 06:05
Drafts in flight
Nothing goes live until it’s published
121
runs in the next 24 hours, each one recorded
Run time
Evaluate now executes the saved code once, with no AI in the loop. What runs on schedule is code you can read.
Records examined, telemetry lag, and what the run concluded — kept for every run, not just the ones that raised.
Telemetry delay is tracked per monitor, so a quiet monitor and a blind one never look the same.
Examples
Works with
Start with the logs you already collect. Panaptico turns them into monitors with verdicts you can defend.