Monitoring
Is the agent trustworthy today?
Every generation is logged with its path, eval results and cost. A rising fallback or repair rate is the alert signal. In production these events go to the warehouse and page on-call.
Pages generated
0
First-pass eval rate
–
Repair rate
–
Fallback rate
–
p50 / p95 latency
–
Tokens used
0
Eval pass rate by check
Generate a page to see results.
Offline regression suite
Run npm run eval to push the fixed brief set through the full pipeline and fail CI if the first-pass rate drops below the threshold.
Recent runs (0 audience, 0 page)
| When | Kind | Campaign / prompt | Path | Evals | Latency | Note |
|---|
No runs yet in this browser.