Decision Analysis
Monte Carlo Simulation
1000 iterations with 10 perturbation types. Validates robustness of tactical recommendations under uncertainty. Wilson 95% confidence intervals on success rates.
Minimax Regret
Savage (1951). Minimizes worst-case decision regret across all possible opponent strategies. Payoff matrix construction with scenario enumeration.
MCDA Framework
Multi-Criteria Decision Analysis. Weighted Sum Model integrating survivability, mission success, tempo, risk, and logistics across competing options.
Sensitivity Analysis
One-at-a-time (OAT) parameter perturbation. Tornado diagrams show which factors most influence final recommendations.
Statistical Methods
Wilson Confidence Intervals
95% CI on success rates. More accurate than normal approximation for small samples and extreme proportions.
Kendall tau-b
Rank correlation coefficient measuring inter-agent agreement on COA ordering. Detects systematic ranking disagreements.
Jaccard Distance
COA similarity metric. Measures overlap between tactical approaches to detect redundancy and ensure genuine diversity.
Weighted Coverage Scoring
Deterministic security assessment. Hierarchical weighting (control → door → platform) with full decomposability.
Scoring & Validation
Flaw Severity Mapping
4-tier severity system: critical (-20), major (-10), moderate (-5), minor (-2). Category-mapped from adversarial analysis.
Risk Derivation
Risk scores derived from flaw categories via deterministic mapping. Not model-generated — computed from structured analysis output.
Consensus Tiebreaking
When scores and risk are tied, Monte Carlo success rate serves as tiebreaker. Eliminates alphabetical or arbitrary ordering.
Triple Modular Redundancy
Three independent assessments per control point. 2-of-3 majority for consensus. All-disagree triggers human review.
Classification
ConclAive's mathematical framework is classified as
Neurosymbolic Type 6 under the Kautz (2022) taxonomy.
Neural components (reasoning agents) and symbolic components (scoring, validation, decision analysis)
interact through three defined interface points.
Performance Metrics
9 formal performance metrics track system quality:
pipeline completion rate, success rate, average latency,
inter-agent agreement, Monte Carlo stability,
flaw detection accuracy, COA diversity score,
recommendation confidence, and audit trail completeness.