g6 was seen disintegration on day 36 due to beatriz's media change accidents.
per beatriz, claude and my final suggestion:
- removes it if it doesn't biologically recover.
- still remove it if it doesn't recover within 3 days (that includes day 36, which is really day 36+2 days).
- if it recovers very quickly, keep it, but it will be a small outlier in an otherwise moderately long longitudinal dataset (ideally 42 days of temporal visibility of the model, but could extend into future until day 120 [more features] or shorten [push toward ideal sample to feature ratio] to past if it underperforms from day 42.
and another bonus: if the day 1-42 temporal window for the ENR model is underperforming, the immediate action is to (again) drop g6. since the disintegration, being an area spike and circularity drop outlier, isn't caused by actual biology anyway.
triple safe.
let's see what happens by day 38, tuesday, 3/24.