Benchmarks¶
Recorded, reproducible benchmark reports for the packages whose implementation checklists require them. Each report is a standalone measurement you can re-run yourself; the scenario scripts live under benchmark/scenarios/.
What these measure - and what they don't
These are reference numbers for order-of-magnitude regression tracking, not controlled lab measurements. Two recurring results hold across every package:
- Repository
get()is O(1) - flat at ~0.08–0.11 µs across 10³–10⁵ populations, exactly as the design specifications require. - Throughput scenarios report honest sequential-vs-parallel numbers. For the trivial in-process fixtures, thread parallelism is GIL-bound and adds no speedup; real speedup materializes only when a concrete plugin backend (a native solver, an LLM call, a charting library) releases the GIL. The benchmarks prove contention-free correctness, not speedup.
Reports by package¶
- Decision -
RuleEnginethroughput,DecisionAuditTraillatency. - Digital Twin -
TwinRepositorylatency, cold-start replay cost. - Simulation -
SimulationRunRepositorylatency,SimulationStateCacheeffectiveness. - Optimization -
OptimizationRunRepositorylatency, sweep re-solve throughput. - AI Agents -
TaskRepositorylatency, task-dispatch throughput. - Visualization -
DashboardRepositorylatency, multi-widget render throughput.