Review: original-PSF epoch-averaging in ngmix.py (also: write an average reconvolved PSF?)

[cailmdaley]

The #761 PSF-column refactor restored the original-PSF fit (lost when the reconvolved-only bootstrapper landed) and now epoch-averages the original PSFs the same way as the reconvolved ones — average_original_psf shares _average_psf_fits with average_multiepoch_psf (src/shapepipe/modules/ngmix_package/ngmix.py:1145–1225), weighting by the galaxy inverse-variance. This is the one science change in the refactor (the rest is renaming/standardisation), and the 2026-06-23 tomography call flagged it for expert review.

Two things to confirm:

1. Is averaging the original PSFs across epochs the right scheme (galaxy-inverse-variance weighting, failed-fit epochs dropped)?
2. Should an average reconvolved PSF also be written?

Meeting record assigns this to Fabian / Axel. Related: #761 (where it landed), #741 (keystone), epic #762.

— Claude on behalf of Cail

[cailmdaley]

Closing — the original-PSF epoch-averaging review doesn't need a standalone issue; it rides as a review thread on #741/#761 where the change landed. (Can't delete on write access; closing as not-planned.) — Claude on behalf of Cail