Every value below is a committed measurement with a reproducibility command. Latencies are reduced-order kernel and simulated-HIL figures, not physical hardware-in-the-loop rig timings.
| Metric | Value | Reproducibility |
|---|---|---|
| Rust PID kernel latency | 0.52 µs P50 | validation/verify_10khz_rust.py |
| Closed-loop HIL latency (legacy row) | 24.5 µs P50 | validation/collect_results.py |
| Taskset-affinity sensor-to-control | 0.0534 ms P95 | scpn_end_to_end_latency.md (logical CPUs 10,11; not shielded cpuset) |
| Simulated 256-actuator HIL scaffold | 232.5 µs P95 | scpn_end_to_end_latency.py --strict (host ADC/DAC loop) |
| Rust full-order GS solve (33×33) | 413 µs SOR / 845 µs MG | cargo bench -p fusion-core --bench picard_bench |
| Rust vacuum field solve | 140 µs (33×33) / 489 µs (65×65) | cargo bench -p fusion-core --bench vacuum_bench |
| Metric | Value | Reproducibility |
|---|---|---|
| QLKNN-10D transport surrogate | test rel_L2 = 0.094 | weights/neural_transport_qlknn.metrics.json |
| FNO turbulence surrogate | val rel_L2 = 0.055 | weights/fno_turbulence_jax.metrics.json |
| Disruption rate (1,000-shot sim) | 0% (Rust-PID) | validation/stress_test_campaign.py |
| Whole-plant fault scenarios | 10/10 reduced-order lanes pass | whole_plant_fault_tolerant_scenario.py --strict (software evidence; not plant hardware / FEA-grade) |
| ITPA H-mode confinement set | 53 shots / 24 machines | validation/reference_data/itpa/ |
| SPARC GEQDSK validation | 8 public EFIT equilibria | benchmark_sparc_geqdsk_rmse.py (operator-source gate passes) |
0D scoping values from RESULTS.md; not a full integrated-scenario or engineering design.