test: Keck-AO acceptance checks 3-4 — astrometry + Einstein-radius invariance #13

Activity

1. Jammy2211 commented on Jul 9, 2026

   

   Jammy2211

   on Jul 9, 2026

   ContributorAuthor

   More actions

   Interim results + parked fit (laptop reset)

   Check 3 — plate-scale audit (finding, fix gated on review): the adapter's narrow-camera native_scale (9.942 mas, from SHARP paper prose) is wrong for both epochs. Truth: 9.952 mas pre-2015 (Yelda 2010 — B1938 raw headers agree: PIXSCALE 0.009952) and 9.971 mas post-2015 (Service 2016). Effect on shipped phase-4 mosaics: output pixels are truly 10.010 mas, so a naive θ_E reads 0.10% low. Proposed fix: epoch-aware native_scale selected alongside the distortion solution.

   Check 3 — geometry leg: no ACS/WFC3 archival imaging of B1938 exists (only 1997–98 WFPC2/NICMOS), so no own-pipeline HST comparison; and the ring is below direct azimuthal-profile detection in the 7,020 s stack (annulus medians at the noise floor) — the model fit carries the geometric test, as in Lagattuta+12.

   Check 4 — fit CHECKPOINTED, resumable: Nautilus (SIE+shear+Sersic lens, SersicCore source, 20 mas Chen-style binning, plate-scale-corrected pixels, tier-A PSF) ran ~1 h on CPU; posterior not yet converged; 94 MB checkpoint.hdf5 on disk.

   Resume after reset (Nautilus resumes from its checkpoint via the identity-hashed output path):

   cd ~/Code/PyAutoLabs/PyAutoReduce
   NUMBA_CACHE_DIR=/tmp/numba_cache MPLCONFIGDIR=/tmp/matplotlib \
   PYAUTO_SKIP_WORKSPACE_VERSION_CHECK=1 \
   ~/venv/PyAuto/bin/python prototypes/b1938_lens_fit.py

   Inputs it needs are all on disk under prototypes/output/b1938_keck/ (dataset regenerated post-merge) and output/b1938_keck_acceptance/ (checkpoint). Success criterion: θ_E consistent with published ~0.45″.
2. Jammy2211

   added a commit that references this issue on Jul 10, 2026

   docs: Keck per-exposure frame products — feasibility verdict GO, gated (

   ...

   34866b7