In Heimdall v2, once lag calibration has converged, what mechanism is intended to guarantee that the five receivers remain aligned to the same physical sample epoch across later hardware/control transitions—especially calibration-noise on/off, periodic phase validation, frequency or gain changes, and other per-device control operations—and which of those events are supposed to invalidate the existing lag solution and force lag calibration to run again?
I’m asking because I have observed fresh, stable inter-receiver residual lags appearing while Heimdall still reports the previous lag state as CONVERGED. Several of those residuals fall very closely on multiples of about 94 complex samples, and manually resetting lag calibration drives them back to approximately zero. I have not established whether those relative offsets originate in the RTL devices, in stream alignment, or at a control/generation boundary.