Operator guide¶
You're running Particles long-term. This guide covers configuration, hygiene, tuning, and the things that go wrong.
| Page | When to read |
|---|---|
| Configuration | config.yaml structure, env vars, precedence |
| Running in a container | The OCI image, compose, the Helm chart, resident daemon mode, the /data volume |
| Remote engine | Client–server split, the bearer token, network exposure |
| Lint and review | PROVENANCE_STALE, INCONSISTENCY, the review → trust workflow |
| Co-evidential curation | Finding and linking duplicate claims with links suggest |
| Citation-signal deposits | Surfacing frequently-cited undeposited primary sources with corpus links suggest |
| Auditing | The operator event log, corpus retract, who-changed-what |
| Structured claims | The derived S-P-O annotation beside a claim: what it is for, particles structure, coverage |
| Scheduled consolidation | The dream cycle — running the maintenance passes unattended from launchd / cron |
| Refreshing mutable local sources | Keeping AGENTS.md-style rule files from going stale in the store |
| Store consolidation | Merging two divergent local stores into one canonical archive (multi-machine) |
| Schema migration | SCHEMA_VERSION major bumps, db init --force, what survives |
| Observability | OpenTelemetry traces, metrics, and the Grafana dashboard |
| Tuning | Trust weights, calibration, trust lenses, min_particle_confidence, age decay |
| Troubleshooting | Database locked, tables not found, fetch failures, follow-edges |
For first-time setup → first query, see the user guide → getting started. For writing a new extractor or exporter, see the plugin-author guide.