The control derivative is benchmarked against classical PID and bang-bang on four plant models. The framework cost adds anomaly-suppression and cross-shadow-consensus terms a classical controller has no access to — the comparison shows when those terms matter.
Plants under test
- Reactor thermal — 4-thermocouple linear plant. The shield-friendly case: cross-thermocouple consensus.
- Aerodynamic boundary layer — 6-tap Stuart-Landau BL transition.
- Combustion (Rijke tube) — 3-transducer van-der-Pol oscillator.
- Tokamak (0-D) — ELM-trigger plasma with anomaly cost dominating during edge events.
Source: control/benchmarks/ · compare_vs_pid.py reproduces every result.