Module

SMA

Messaging Security Audit. Part of the ASEMA framework. Multi-cycle door-by-door deep analysis with deterministic scoring and 4-layer anti-hallucination constraints.

Relationship to ASEMA

SMA is the detailed analysis engine within the ASEMA framework. Where ASEMA provides the overall assessment structure (7 doors, 63 controls, TMR), SMA handles the deep-dive analysis of individual security domains. Multi-cycle passes through each door, with escalating depth at each iteration.

Benchmarking

19 models benchmarked for security assessment accuracy. Scoring is deterministic — same inputs always produce same scores. Model variation is captured through TMR, not through score averaging.

Anti-Hallucination

Schema Enforcement
Responses must match predefined structures. Non-conforming output rejected and regenerated.
Citation Requirement
Every assessment references specific documentation or observable platform behavior.
Cross-Model Validation
Triple modular redundancy catches single-model confabulation through majority consensus.
Deterministic Aggregation
Final scores computed arithmetically from validated sub-scores. No model generates the final number.

Deployment

Online and offline modes. Online mode leverages frontier models for deeper analysis. Offline mode runs on local models for classified or air-gapped environments.