Rules¶

Rules are the atomic unit of compliance evaluation in Opencomplai.

Every rule today belongs to the EU AI Act, and rule sources are EU AI Act only. Other frameworks are assessed from the same evidence: NIST AI RMF by re-projecting the EU AI Act rows, and a natively evaluated framework through its own requirements map (see Adding framework packs).

What a rule does¶

Each rule is a deterministic check that:

1. Receives an AssessmentInput (model metadata + optional answers).
2. Returns a RuleResult with a passed boolean, rationale string, and a reference to the article of the framework the rule belongs to.

Rule ID naming convention¶

Rule IDs follow the pattern EU_AIA_<ARTICLE>_<DESCRIPTION>:

The live rule set¶

Rules are implemented in packages/core/src/opencomplai_core/rules.py. Each Rule subclass implements the evaluate(input: AssessmentInput) -> RuleResult method.

The rule engine (in packages/core/src/opencomplai_core/engine.py) discovers all registered Rule subclasses, runs them against the input, and aggregates results into a RiskResult.

Adding new rules¶

See Adding Rules for the step-by-step guide.

Python model reference¶

from opencomplai_core.models import RuleResult

class RuleResult(BaseModel):
    rule_id: str
    rule_name: str
    passed: bool
    rationale: str
    reference: str  # EU AI Act article or clause reference