Observability

The observability commands are model-free. They open a read-only view over the populated state directory (the durable receipt trail, world-model truth, and cached topology) and print projections. They run fully offline, with no key and with the model deps absent.

A read-only projection exits 0 even over an empty state directory: that is the honest quiet view. The two audit paths, receipts verify and inspect --strict, exit non-zero on a tampered or broken chain.

status

reactor status

Reports the standing compile cost beside the live run cost, plus the per-node dispositions. This is the at-a-glance answer to "what did this system cost, and what is it doing?"

topology

reactor topology

Prints the compiled DAG: the nodes (each with its wake source) and the resolved edges. It requires a compiled IR; without one it tells you to run reactor compile first.

inspect

reactor inspect <node> [--strict]

Inspects a single node: its topology position, its fingerprints, its last receipt, and its chain. A plain inspect is a pure read. With --strict, the command exits non-zero if the node's receipt chain does not verify, which makes it a CI gate.

logs

reactor logs [--node <node>]

Prints the receipt stream as compact log entries, optionally filtered to one node with --node.

trace

reactor trace [<node>]

Traces each node's receipt chain in chain order, from wake to disposition. With no argument it traces every node that has receipts; pass a node id to trace just one.

receipts

reactor receipts [list|verify|cost] [--node <node>]

Audits the receipt trail. The default subcommand is list.

Chain verification and tamper detection

receipts verify is the integrity check. It walks the receipt chain and exits non-zero the moment it finds a break, which is exactly what tampering or corruption looks like. The same chain check backs inspect --strict. Both are designed to drop straight into CI:

reactor receipts verify        # exit 1 on a broken chain
reactor inspect digest --strict

Reading cost

Every receipt carries a surprise_cause. reactor receipts cost and reactor status roll cost up by that cause, so a cost figure is always attributable to a specific change.

Because a node that re-wakes with unmoved inputs memo-skips at zero render cost, the standing cost of a quiet system trends to zero. A spike therefore means a real change propagated, and the rollup tells you which one. See cost scales with surprise for the model.

JSON output

Every observability command honors --json for machine-readable output, so you can pipe a projection into a script or a dashboard.

reactor status --json
reactor receipts cost --json

When a running daemon is bound with --http, the same projections are available over HTTP at /status, /cost, /topology, /receipts, and /nodes/<node>. See the HTTP surface.

Visualizing a run

The receipts, world-models, and topology a state-dir holds are exactly what Reactor DevTools replays. Point it at the same directory to watch the run animate -- nodes flashing on render, dim-pulsing on memo-skip, the cost meter spiking on a real change -- or read it headlessly with reactor-devtools <state-dir> --describe.