The compliance challenge
Regulators ask: "How did you decide to recommend this product?" or "Prove your vendor selection wasn't biased."
Traditional ML-based ranking gives you a black-box score. Even if you log features and weights, you can't show that the decision was:
- Optimal under stated criteria
- Non-discriminatory (criteria applied consistently)
- Auditable (level-by-level logic trail)
- Resistant to gaming (sellers can't boost rank with irrelevant features)
You need a system that produces decisions with built-in receipts.
Explainable by design
Every decision comes with a structured audit trail showing which criteria passed, which failed, and how dominance levels resolved ties.
Audit trail example: Vendor selection
Regulatory review of infrastructure procurement
Query logged:
"Select cloud vendor for regulated workload. Must meet SOC 2 Type II, HIPAA, data residency in EU. Prefer low cost, then low carbon, then vendor diversity."
Level 1 (hard constraints):
- SOC 2 Type II: ✓ Passed (3 vendors)
- HIPAA: ✓ Passed (3 vendors)
- EU data residency: ✓ Passed (3 vendors)
Candidates A, B, C all passed. Candidate D excluded (no EU region).
Level 2 (price buckets):
Candidate A: $0.08/GB (bucket P1) > B: $0.11/GB (P2) > C: $0.13/GB (P3)
Winner at this level: A (cheapest bucket dominates).
Level 3 (carbon preference):
Candidate A uses 60% renewable energy (tie-breaker if price were equal).
Not evaluated (A already won on price).
Level 4 (vendor diversity):
Candidate A is not over-represented in current portfolio (27% share < 30% cap).
Passed soft preference check.
Final decision (logged with timestamp):
Selected: Vendor A
Rationale: Met all hard constraints (SOC 2, HIPAA, EU residency). Lowest price bucket ($0.08/GB) dominated other compliant vendors. Carbon and diversity criteria satisfied but not decisive.
Co-optimal alternatives: None (A unique winner).
Non-discriminatory
Same criteria applied to all candidates. No hidden bias. Dominance levels ensure consistent priority across all evaluations.
Complete audit trail
Every decision logged with timestamp, input query, extracted criteria, level-by-level evaluation, and final rationale.
Legally defensible
Patent-backed methodology (US 2025/0298806 A1) with provably optimal results. Show courts or regulators the exact logic.
Gaming-resistant
Vendors or suppliers can't manipulate rankings by adding irrelevant features. Only meaningful improvements at the right dominance level matter.
Compliance use cases
AI systems & LLMs under EU AI Act
Builders of AI-based recommendation systems must provide transparent, explainable decisions under the EU AI Act. Our dominance-based approach delivers complete audit trails showing exactly how each recommendation was reached—meeting transparency requirements while remaining gaming-resistant. Perfect for high-risk AI systems requiring human oversight and regulatory compliance.
EU AI Act & GDPR
Meet "right to explanation" requirements. Show users and regulators how automated decisions were reached, with clear criteria and dominance logic.
Financial services
Show regulators how investment recommendations or loan decisioning meet fairness and transparency requirements. Audit trail proves no protected-class bias.
Healthcare & pharma
Document why a treatment or device was recommended. Prove compliance with clinical guidelines and formulary policies. Full traceability for FDA or EMA review.
Government procurement
Demonstrate vendor selection was fair, transparent, and optimal under stated criteria. Withstand FOIA requests and oversight audits with complete logic trails.