ngmix v2.0 — a correct, validated v2.0 shape catalogue

[cailmdaley]

Tracking issue (epic) for the ngmix v2.0 shape-measurement upgrade — the keystone of the UNIONS tomographic round. PRs merge and fall away; this issue is the durable home for the goal, and the single place to see every moving piece, across both shapepipe and sp_validation.

The goal

A v2.0 shape catalogue we trust end-to-end: ngmix 2.4.0 + Lucy Baumont's new ngmix classes, PSF/shape columns that mean what their names say, the known shape-measurement bugs fixed, and the downstream validation consuming the new columns.

Pieces

Shape measurement — shapepipe

• Ngmix v2.0 (CI mirror of #740) #741 — Ngmix v2.0 migration (the keystone PR)
• ngmix v2.0: one column grammar across estimators (PSF source-of-truth, single size, E↔G) #761 — PSF columns: single source of truth (original vs reconvolved PSF)
• [BUG] ngmix module passes the galaxy shape prior to the PSF fit, which is prior-dominated #749 — PSF-prior bug: the PSF fit is prior-dominated → biased PSF columns (tracked as a sub-issue below)

Downstream — sp_validation (cross-repo)

• Migrate ngmix PSF/shape column names (tracking shapepipe#761) sp_validation#201 — adopt the renamed v2.0 PSF/shape columns. The column rename in ngmix v2.0: one column grammar across estimators (PSF source-of-truth, single size, E↔G) #761 forces a matching change in validation, so this lives in the other repo but is part of the same initiative.

Not in this epic

• General ShapePipe test guardrails (star shear-response, m-bias) — these test ShapePipe broadly, not v2.0 specifically; tracked separately.
• ngmix config knobs (metacal_psf / metacal_step / galaxy_model) — deferred; not on the board yet.

— Claude on behalf of Cail