Skip to content

Algorithm Fidelity: PHREAK vs the Reference Evaluator

Prerequisites: The FluxRules Engine.

FluxRules ships one production engine (PHREAK). To guard against defects in shared code, the test suite checks PHREAK against an independent, dependency-free reference evaluator (fluxrules.services.reference_evaluator.ReferenceEvaluator). Because the two implementations share no matching code, agreeing on results is strong evidence of correctness.

Verification approach

Test dataset: cross_border_payment_risk_triage (3 facts, shared use case)

Test rules: All 31 example rules with varying complexity: - Simple conditions (1 field) - Complex AND/OR logic - NOT operators - Nested groups

Verification method: 1. Load the same ruleset into PHREAK and the reference evaluator 2. Evaluate identical facts 3. Compare fired_rules and actions 4. Assert complete parity

The same harness also compares PHREAK's stateless and streaming modes to confirm a fact fires the same rules on first sight in either mode.

Results

✅ COMPLETE PARITY - PHREAK matches the reference evaluator on all test cases.

Test Case PHREAK Reference Match
Simple condition ✅ ✅ ✅
AND groups ✅ ✅ ✅
OR groups ✅ ✅ ✅
NOT operator ✅ ✅ ✅
Nested groups ✅ ✅ ✅
Multiple rules ✅ ✅ ✅

Conclusion: PHREAK's optimized matching is semantically equivalent to the naive reference evaluator across the supported condition surface.

Next Steps