System map
Five deterministic stages · one LLM judgement in the middle
- Ingest1/5
Stripe + HubSpot
waiting…
- Candidates2/5
deterministic · no LLM
waiting…
- Judge3/5
Claude Sonnet 4.6
waiting…
- Policy4/5
autonomy tiers
waiting…
- Executor5/5
write + read back
waiting…
No eval has run yet
Run an eval to score the agent against the 25 planted scenarios — 20 real discrepancies and 5 decoys it must leave untouched. Press ⌘K to start one.
Results
0 of 0Nothing judged yet
Start a run and rows will stream in here as the judge returns each verdict.
Activity
0No activity yet
Events stream here over SSE the moment a run starts.