Policies

Policies express requirements that govern language elements.

A policy can describe concerns such as reliability, performance, security, governance, data protection, or confidence. It can then govern a capability, effect, event, outcome, or lifecycle. Observability is modelled with observe blocks rather than an observability policy family in DCL v1.0.

Policies keep cross-cutting requirements visible in the model instead of burying them in prose or implementation-specific configuration.

Confidence policies use grouped family block syntax:

policy MinimumAnswerConfidence {
  confidence {
    threshold 0.8
  }
}

The threshold is a number between 0 and 1. It defines a boundary for accepting a decision, effect, outcome, or tool result; it is not a full LLM evaluation or grounding model.