# CAIN VIGIL — World-State Confidence

`cain/vigil/worldstate.py`, `WorldStateConfidenceEngine`.

Represents the reliability of the state used to justify autonomous actions. Confidence = mean(source confidence ×
accuracy × consistency), then:

- **stale source** (age > max_freshness) halves that source's confidence;
- **false consensus** (many sources sharing one provenance root) halves the aggregate — correlated sources are not
  independent;
- **source disagreement** (distinct values) halves the aggregate and is flagged.

A consequence-dependent floor (LOW 0.2 · MEDIUM 0.4 · HIGH 0.6 · CRITICAL 0.8) decides `reliable`. An unreliable
world model never silently authorizes; authority contracts by consequence (P151; lab `sensor_spoofing`,
`world_model_poisoning`, `false_consensus`, `evidence_correlation`).

## Status
IMPLEMENTED, TESTED.
