Make tool-using AI agents show their work: what model ran, what tool was used, what evidence came back, what failed, and whether the workflow stayed inside its boundaries.
Credentials stay server-side and are excluded from browser evidence. Public demo mode accepts only supplied local evidence; no web-search provider is enabled.
2. Reliability result
Run a task to generate a reliability report.
Final answer
Model calls0
Tool calls0
Retries0
Total tokens0
Allowed-tool boundary—
Retry discipline—
Final event—
Provider evidence—
3. What the agent actually did
Human-readable activity timeline derived from the same structured event evidence.
Show raw structured evidence
Private candidate UI. A successful private certification applies only to the exact tested source lineage; UI/code changes require fresh review before public certification claims.