H8, H10, and the entire col 11 are showing signs of plateuing and mild decline, not severe enough to trigger suggestive flags for exclusions, but severe enough to a human eye, looking at the heatmaps.
what is important to note is that they are likely to survive after the transfer, and that means the covariate might be lost if, when screening for covariates beyond day 21, we only consider the overall trajectory that was completely provisional because the physical positioning will shift completely. it would be a paradigm shift.
what i propose are the following and i think there's no harm besides the P>N problem, which is also mild enough:
- the 96w screen is computed regardless of being beyond or within the initial 21d window. if beyond, d21's covariate's screening result will persist + carried over to whatever endpoint above d21.
- the 24w screen produces a separate covariate based on the new plate geometry. it is completely independent of the 96w screen, to the point where for example, covariate screening calculates base on d1-d42; the two independent screens that don't object but complement each other would then be d1-d21 screen and d22-d42 screen. then the result of each, eg cov_col/row#_96w (however many of these are) and cov_col/row#_24w (however many of these are) are both populated inside the final dataset, without z-score standardization.
to kind of visualize this:
- d1-d2: screen for d1-d2 based on 96w geometry
- d1-d17: screen for d1-d17 based on 96w geometry
- d1-d21: screen for d1-d21 based on 96w geometry
- d1-d22: screen for d1-d21 based on 96w geometry; separately screen for d22 based on 24w geometry
- d1-d(22+n): screen for d1-d21 based on 96w geometry; separately screen for d22-day(22+n) based on 24w geometry