Acquisition temperature predictions may be using stale data from the dark model

Because the max temperature predictions are put in a cache by catalog, they basically are assigned the acquisition probability at time=t where t is the first good-pitch time day the combination of stars was available:

https://github.com/sot/aca_lts_eval/blob/3879648c054a677f3c51c8a40e5b79f27ffd7de3/aca_lts_eval.py#L116

This means that over a cycle, when the same combination of stars may be seen many times, the acquisition temperature prediction may be using a much earlier time/date for the prediction of a "later" catalog. The time is fed into get_warm_fracs as part of the acquisition probability determination, so the predictions on aca_lts_eval may be based on a dark model up to 18 months "too old", that is with up to 18 months of modeled CCD degradation missing.

The easiest way to improve prediction fidelity here would probably just be to index into the temperature prediction cache with both catalog and some binned version of the date (year + week number?).

sot / aca_lts_eval

Acquisition temperature predictions may be using stale data from the dark model #39