Territory Grounder · Console tour

Inside the operator console

Seven views of the interface, using representative fixture data. These are illustrative screenshots, not the evidence from the lab recording. Select a view, then open its image to inspect it at full size. On a keyboard, use the arrow keys, Home and End to move between tabs.

The recorded lab demonstration shows actual executions on a disposable test machine.

The Command view with a session drawer open: the grounding ribbon pipeline (classified, predicted, approved, executed, verified), a map-laid-over-territory scatter with one node in deviation, the verifier's callout, and the proposed action bound to its hash. — **Command**: a session opened mid-triage, the map laid over the territory, and a deviation the acting model cannot overturn.

The Workflows view: an AI session as a typed run: ingest, correlate, classify, prediction commit, agent loop with its tool calls, prediction gate PASS, dry-run execute, and a final verify step stamped DEVIATION. — **Workflows**: an AI session as a typed, replayable run; the gate passes, the verifier still says `DEVIATION`.

The Signals view: nine telemetry charts drawing steel observed lines against indigo predicted envelopes, with deviation shading, event markers, and a 24-hour map-agreement score. — **Signals**: steel observed against indigo predicted; agreement is a number, not a feeling.

The Estate Depth view: a node inspector for a Kubernetes worker with golden signals, upstream and downstream dependencies, predicted blast radius, and the incident master-clock beneath. — **Estate**: golden signals, dependencies, predicted blast radius, the causal graph it reasons over.

The Ledger view: the tamper-resistant governance ledger, with sequence numbers, AI, HUMAN and SYSTEM actors, events like prediction.commit, floor.hold (never-auto) and kill-switch.arm, each row hash-chained to the previous. — **Ledger**: every decision on one SHA-256 chain, `AI`, `HUMAN` and `SYSTEM` on the same spine.

The Grounding scorecard: verifier match-rate, the falsifiability signal against a shuffled-graph control, and blast-radius precision and recall. — **Grounding**: the verifier's report card, the safety record as evidence, not assertion.

The Governance view: the graded-autonomy ladder with per-layer promotion, the mutation-off banner, and the non-configurable never-auto floor for irreversible operations. — **Governance**: the autonomy envelope and the never-auto floor.

Evaluate in Shadow Read the source