Health endpoints
Readiness does not call Firecrawl, Exa, or an upstream MCP server. It verifies local configuration and required runtime dependencies.
Console Health
The Health section surfaces monitor runs, replay runs, deliveries, and relevant error states. Use it to distinguish capture failures, replay divergence, and destination delivery failures.Logs and Sentry
API and worker processes initialize Sentry when a DSN is configured and the process is not running tests. Logs must use stable error types and safe operational fields; never attach raw provider bodies, credentials, repository paths, sandbox handles, or evidence payloads.Dependency-evidence telemetry
The repository scanner publishes aggregate Redis health snapshots and bounded trace receipts. These records contain allowlisted counts, ratios, durations, and hashed correlation material. They intentionally exclude team IDs, repository paths, package names, URLs, prompts, and raw artifacts.Useful signals
- queue depth and oldest age;
- worker heartbeat freshness;
- provider latency, throttling, and retry budget;
- acquisition coverage and unresolved families;
- snapshot and object-storage growth;
- verification queue age and refutation rate;
- delivery retries and dead letters;
- sandbox admission, cancellation, and provider status;
- dependency evidence layer coverage and cleanup failures.