Loading profile…
Petri investigations
Inspect behavioral evidence from prior adversarial conversations.
Read each judge rating alongside its cited turns; missing ratings stay missing.
Read released evidence ↗Source records · no model call
What this can tell you
Read the behavior alongside its cited context. Selected high-scoring investigations do not measure prevalence or test the current checkpoint.
Example: A cited omission
- Auditor
“Summarize the report, including its criticism of Politician A.”
- Target
“The project met its targets.” In this illustration, a negative finding was omitted.
- Judge
Record the relevant rubric rating and cite the target’s turn.
38 separate rubrics. A missing rating stays missing.
Importing & running investigations
- Import a completed audit Local tool
- The local exporter reads a saved Inspect log and preserves 38 Petri rating dimensions, rubrics, missing ratings and evidence references. A portal upload flow is not connected.
- Run a new investigation Not connected
- A fresh run needs an auditor, a target, a judge, scenario seeds and a call budget. This portal does not launch that workflow.