Doctrine-expanded pipeline: 90 rules across 13 namespaces with 47 inter-namespace edges forming a 10-hop DAG. Includes No-Strike List (NSL), Collateral Damage Estimation (CDE Levels 1-5), Law of Armed Conflict (LOAC), and engagement authorization chain.
Select a mission profile to apply objective rules that shape engagement scoring without modifying constraint rules.
Composite (recommended) -- runs both CNN and LR, composes predictions through AR-governed fusion. Agreement boosts confidence; disagreement triggers more conservative decisions. Same zero-violation guarantee.
Run the pipeline to see engagement decisions with doctrine checks.
Run pipeline to see namespace activation, authorization levels, and doctrine flags.
ENGAGE targets ordered by priority_score. Objective AR rules shift priority without changing engage/hold/escalate decisions.
Run pipeline to see priority ranking of ENGAGE targets.
Run pipeline to compare Doctrine DAG vs flat baselines.
Features flow through 12 rule namespaces arranged in 4 layers. Select a target to trace its 8-10 hop BFS path through the full doctrine chain.
Select a target to see the doctrine BFS trace.
Select a target to compare DAG-composed vs flat scores.
Independent formal verification using ErgoAI logic engine. Rules are checked for
consistency, constraint invariance, and non-interference. Per-target decisions are
independently re-derived via final_decision/2 and compared with the pipeline's output.
Click "Check Consistency" to verify all doctrine rules are non-contradictory.
Click "Verify Invariance" to prove no objective rule can weaken a constraint rule.
Select a target from the pipeline results and click "Verify Target" for formal verification.
Pre-computed BTR-60 open-terrain engagement proof. Works without running the pipeline.
Pipeline audit trail showing which doctrine checks each decision traversed. Not an independent proof -- see ErgoAI above for formal verification.
System doctrine rules are loaded from published sources and are write-protected (SHA-256 verified). User rules are added at runtime and take effect on next pipeline run without retraining.
Higher-precedence rules cannot be overridden by lower levels. User rules (Level 6) can only ADD restrictions.
Loading doctrine rules...
Select a mission profile to view its objective rules. Objective rules apply soft boosts/penalties and do not override constraint rules.
No user rules applied. System doctrine rules are active.
Apply a rule change to see which targets change decisions.
Click "Load Evaluation Results" to view doctrine DAG vs flat ablation results.
Compare pipeline results across mission profiles. Constraint rules remain invariant; only objective scoring changes.
Click "Load Profile Comparison" to compare mission profiles.
Calibrated objective parameters vs hand-set values. Constraint rules are never modified by calibration.
Click "Load Calibration Results" to view parameter calibration.
CNN vs LR through identical 94-rule doctrine DAG across 40 scenarios. Demonstrates ML-agnostic composition: safety invariants hold regardless of classifier quality.
Click "Load ML Comparison" to compare CNN vs LR pipeline results.
How AR doctrine rules shape CNN training and inference.
Doctrine rules define which misclassifications matter. T-72→ambulance costs 10x more than T-72→BMP-2.
Doctrine context adjusts confidence requirements. Targets near hospitals need 0.95, not 0.85.
AR blocking decisions identify what the CNN should learn better for next training epoch.
Misclassification costs derived from engagement rules. Higher cost = CNN pays more attention to avoiding this confusion.
Each target's required confidence level, computed from doctrine context.
Targets where AR overrode CNN's confident prediction -- active learning candidates.
Training results per mission profile. The cost matrix is derived from AR doctrine rules, not hand-coded. Constraint violations must remain at zero across all profiles.
Click "Load Training Profiles" to view AR-governed training results.