Affected Repositories

• PyAutoLens (primary)

Branch Survey

Suggested branch: feature/padded-single-image-recovery

Implementation Steps

1. In autolens/analysis/result.py image_plane_multiple_image_positions (the if multiple_images.shape[0] <= 1: block), compute finite = np.isfinite(np.asarray(multiple_images)).all(axis=1); if finite.sum() <= 1 call image_plane_multiple_image_positions_for_single_image_from(); else return aa.Grid2DIrregular(values=np.asarray(multiple_images)[finite]). Keep the zero-images comment accurate and extend it to explain inf padding.
2. In image_plane_multiple_image_positions_for_single_image_from, apply the same finite-row count to the multiple_images.shape[0] > 1 test and return only finite rows.
3. Tests in test_autolens/analysis/test_result.py: monkeypatch PointSolver.solve to return (a) 1 finite row + inf rows → recovery helper called; (b) 2 finite rows + inf rows → exactly the 2 finite rows; (c) unpadded 2-row numpy array → unchanged.
4. Witness: run test (a) on unfixed source (fails) and with the fix (passes).
5. Run python -m pytest test_autolens/analysis -q and the full test_autolens suite.

Key Files

• autolens/analysis/result.py — image_plane_multiple_image_positions, image_plane_multiple_image_positions_for_single_image_from
• test_autolens/analysis/test_result.py — new tests

Pin risk: none for finite inputs (numpy solver output contains no non-finite rows, so the filter is the identity).

Links: PyAutoLens#765, https://github.com/orgs/PyAutoLabs/discussions/30

PyAutoLens: single-image recovery in Result.image_plane_multiple_image_positions_from is blind to inf-padded JAX solver output

Type: bug
Target: PyAutoLens
Repos:

• PyAutoLens
  Difficulty: medium
  Autonomy: safe
  Priority: normal
  Memory: wiki/galaxies/sources/massive-ellipticals.md; wiki/lensing/sources/dark-matter-substructure.md; wiki/galaxies/sources/light-profile-fitting.md
  Status: formalised
  Consequence: judge
  Witness: a new test in test_autolens feeding image_plane_multiple_image_positions_from a 1-image array padded with inf rows fails on main and passes with the fix
  Review-minutes: 20
  Unattended: ready

PyAutoLens: single-image recovery in Result.image_plane_multiple_image_positions_from is blind to inf-padded JAX solver output.

Found while reviewing community PR PyAutoLens#765 (@samlange04, discussion https://github.com/orgs/PyAutoLabs/discussions/23 sibling report #30). In autolens/analysis/result.py (around line 138) the inward-walk recovery for a single-image solve tests multiple_images.shape[0] <= 1. With a JAX-backed analysis the point solver (xp=jnp, remove_infinities defaults to xp is np since #759) returns a static-shape array padded with (inf, inf) rows (20 rows), so the shape test is always false and the recovery can never fire under JAX. After #765 (which drops non-finite rows at PositionsLH construction and raises PositionsException when fewer than 2 finite rows remain) a 1-image JAX solve now raises at construction instead of silently scoring zero penalty. Fail-loud is better than silent zero, but SLaM chains on JAX could start erroring where they used to run.

Fix: count finite rows (np.isfinite(...).all(axis=1).sum()) before the <= 1 check, or strip non-finite rows before it, so the recovery path triggers for padded output exactly as it does for the numpy solver. Add a unit test feeding a padded 1-image array and asserting the recovery path runs (and a 2-image padded array does not). Not yet reproduced with a real 1-image JAX solve; severity unknown. Check that the cached files/multiple_image_positions.json written from padded positions still round-trips.