Metric
Metrics are how you rate signals against the dimensions that matter to your investigation — impact, certainty, novelty, time horizon, anything you define. Each metric is one task in the workflow and produces one score per signal.
Defined on the workspace
A workspace declares its metrics: a name, a scale (numeric or ordinal), and a description that tells the agent what good and bad look like. Playbooks seed a starter set; you can add, remove, or rewrite metrics at any time.
Scored comparatively
The assess task sees every active signal at once and scores them against one metric in a single pass. Comparative scoring keeps the scale honest — the agent calibrates against the corpus rather than rating each signal in isolation.
Multi-model reconciliation
You can configure assess to run with several models and reconcile their outputs via median or mode. The reconciled score is what shows on the signal; the per-model scores remain available in the signal’s timeline.
Manual override
You can edit a metric score by hand at any time. Manual edits survive re-runs unless you force re-assessment, which restores the model-driven value.