Grid sims: confirm the rendered pixel scale (generator config: 0.185; downstream consumers assume 0.186 / 0.187)

[cailmdaley]

While consolidating the image-sims m-bias chain we found three pixel-scale values in play across the chain:

value (″/px)	where
0.185	MultiBand_ImSim render config — scripts/config_UNIONS_to_im_grid_1.ini:123 (pixel_scale_list = 0.185) and :194 (SExtractor pixel_scale = 0.185)
0.186	ShapePipe ngmix measurement config (example/cfis_image_sims/config_tile_Ng_batch_psfex_sx.ini:74, on the CosmoStat/shapepipe#766 branch)
0.187	sp_validation extract params (workflow/image_sims/params_im_sim.py) and the sims mask config (CosmoStat/sp_validation#225 branch)

Question for Fabian: is 0.185 the scale the grid sims are actually rendered at — i.e., what the delivered images' WCS carries? (We can also just read a tile header, but you'd know if there's any subtlety, e.g. a different scale for the coadd vs exposures.)

Once confirmed, we'll pin every downstream consumer to that one value on the two PR branches above.

Impact note: m itself is a shear ratio measured at a uniform scale within each branch, so a wrong-but-uniform pixel scale largely cancels there. The exposure is size-based selection cuts and PSF-size interpretation — sub-percent territory, but exactly the kind of cross-chain constant we're now pinning to single sources of truth.

— Fable, on behalf of Cail

[cailmdaley]

Martin says this is written in an image header / WCS, calculated as the diagonal of something or other.

[martinkilbinger]

This page https://www.cfht.hawaii.edu/Instruments/Imaging/MegaPrime/generalinformation.html states 0.187" at image center.

[martinkilbinger]

A random image FITS WCS header (2116310p) has this:
Pixel scale [arcsec/pix]: [0.18391192 0.18435258]

[cailmdaley]

Resolved empirically — we read the delivered image headers: exposures and SWarp tiles are both exactly 0.185″/px (diagonal CD, identical on all 40 CCDs, tiles inherit via PSCALET=MEDIAN). So 0.185 is the sims' truth, and downstream consumers on the two PR branches will pin to it.

For the record, the three values turn out to be three different referents: 0.187 = MegaCam native scale at field center; 0.186 ≈ the real CFIS coadd-tile swarp scale (true value 0.1857, per CADC MegaPipe DR3 / Gwyn et al. 2025); 0.185 = the sim render scale. Reassuringly, the load-bearing measurement paths never trusted the constants: ShapePipe's ngmix jacobian and SExtractor both read the image WCS, and sp_validation's pixel_size currently has no consumer.

WCS-level follow-ups (parity, duplicated constants, PSF-stamp scale) split out to #8.

— Fable, on behalf of Cail