A day of bench regressions (crash at 2 A after three seemingly-safe control-path changes) shows every control-affecting change needs a pre-hardware regression check. Minimum viable: host-run the generated tim_isr domain against a scripted sensor feed with a simple motor model, asserting current tracking and no saturation rail; ngspice plant sim remains the longer-term version.
A day of bench regressions (crash at 2 A after three seemingly-safe control-path changes) shows every control-affecting change needs a pre-hardware regression check. Minimum viable: host-run the generated tim_isr domain against a scripted sensor feed with a simple motor model, asserting current tracking and no saturation rail; ngspice plant sim remains the longer-term version.