Class Hierarchy
MONTE_CARLO_EXPERIMENT
└─ uses → TRIAL_OUTCOME (N copies, one per trial)
└─ contains → MEASUREMENT (multiple per outcome)
MONTE_CARLO_EXPERIMENT
└─ produces → SIMULATION_STATISTICS (aggregated)Data Flow
User defines trial logic agent
↓
MONTE_CARLO_EXPERIMENT runs N trials
↓
Each trial invokes user's logic agent
↓
Agent returns TRIAL_OUTCOME with MEASUREMENTS
↓
Experiment collects all outcomes
↓
Extract all measurements from all outcomes
↓
Compute statistics (mean, std_dev, CIs)
↓
Return SIMULATION_STATISTICS to userDesign Decisions
Type-Safe Measurements
MEASUREMENT distinguishes observed values from probabilities, preventing category errors in scientific code.
Semantic API
Uses business domain terminology (Experiment, Trial, Outcome) rather than mathematical terminology, making the library accessible to engineers.
Agent-Based Trial Logic
Flexible approach lets users define trial behavior inline without inheritance or complex factory patterns.
Sequential (for now)
Current phase uses sequential trial execution. SCOOP-based parallelization planned for v2.0.
Performance
- Single trial: ~1 microsecond (agent invocation + outcome creation)
- 1,000 trials: <50 milliseconds
- 1,000,000 trials: <50 seconds (estimated)