ADR-002: Policy seeds the baseline

Status

Accepted

Context

Pure static rules are too brittle for enterprise behavior analytics, while pure unsupervised learning often fails to encode what the business actually expects.

Decision

SenseGNAT will use policy to define expected constraints and then use observed telemetry to construct and refine behavior baselines.

Consequences

• policy drift becomes detectable
• baseline quality improves with observed reality
• operators can distinguish forbidden, expected, and merely unusual behavior