HelpLab
Observatory
The live, watchable view of a pipeline run — a clickable node graph, a per-node input to decision to output inspector, a phase ribbon, and an append-only event timeline you can scrub. · 3 min read
The Observatory replays a recorded pipeline run so you can watch it node by node. If no run has been traced it degrades honestly — no run, no green board — and tells you how to produce one.
How to use it
- Scan the health chips (P1–P5 pass count, profile, $0 cost, seed, node/event counts) and the phase ribbon showing where the run is.
- Click any box in the node graph — the Node inspector opens its tooling and its INPUT → DECISION → OUTPUT contract plus the events that node emitted.
- Use the event ledger on the left: Play, Step, or scrub the timeline. It is append-only with a monotonic
seq, so the surface polls new events withsince=N. - Review the P1–P5 headline-properties panel and the enrichment / data-provenance panel beneath the inspector.
What it proves: observability is the trust mechanism for hands-off autonomy. The replay is deterministic seeded (decision R5), so the run is byte-identical every time, and the per-node I/O evals show that each node's decision follows from its inputs.
At a glance
| Route | /observatory |
| Module | app/observatory.py |
| Polling APIs | /api/observe/events?since=N · /nodes · /evals · /meta |