fix: Sph profile transform startswith/endswith name-check asymmetry

[Jammy2211]

Overview

The two mirror grid-transform methods in GeometryProfile use inconsistent spherical-profile name checks: transformed_to_reference_frame_grid_from checks endswith("Sph") while transformed_from_reference_frame_grid_from checks startswith("Sph"). Spherical profiles are suffix-named (IsothermalSph, NFWSph, …) and no concrete class starts with "Sph", so the inverse transform's spherical branch never fires — every Sph profile takes the elliptical path with a pointless identity rotation. Numerically benign today (Sph profiles have angle = 0), but it is an asymmetric latent trap and wasted work on the hot path.

Plan

• Fix the name-check asymmetry in transformed_from_reference_frame_grid_from so both transforms take the translation-only path for spherical profiles.
• Evaluate replacing both string checks with an isinstance-based check so subclasses like IsothermalSphMLR (name ends "MLR", missed by both string checks) are also covered.
• Add a regression unit test round-tripping a grid through both transforms for a spherical profile (NumPy only, no JAX).
• Run the profile/geometry test suite and ship via the standard library flow.

Detailed implementation plan

Affected Repositories

• PyAutoGalaxy (primary)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoGalaxy	main	clean

Suggested branch: feature/sph-transform-name-check

Implementation Steps

1. autogalaxy/profiles/geometry_profiles.py:392 — transformed_from_reference_frame_grid_from checks self.__class__.__name__.startswith("Sph"); its mirror transformed_to_reference_frame_grid_from (line 371) checks endswith("Sph"). Minimal fix: change line 392 to endswith("Sph") for symmetry.
2. Preferred fix (evaluate first): replace both name checks with an isinstance(self, SphProfile)-style check if the class hierarchy supports it cleanly. This also covers IsothermalSphMLR (autogalaxy/profiles/scaling_relations.py:16, subclasses mp.IsothermalSph but its name ends "MLR" so both string checks miss it). Precedent for pairing the name check with isinstance exists at autogalaxy/profiles/light/linear/abstract.py:131.
3. Add a regression unit test: for a spherical profile, round-trip a grid through transformed_to_reference_frame_grid_from → transformed_from_reference_frame_grid_from and assert the translation-only path is taken both ways / results are exact. NumPy only — no JAX in unit tests.
4. Run the geometry/profiles tests, then ship_library.

Key Files

• autogalaxy/profiles/geometry_profiles.py — the asymmetric checks (lines 371 / 392).
• autogalaxy/profiles/scaling_relations.py — IsothermalSphMLR, the subclass both string checks miss.
• autogalaxy/profiles/light/linear/abstract.py — existing endswith("Sph") or isinstance(...) precedent.

Notes

• Numerical impact currently nil: spherical profiles have angle = 0, so the elliptical path applies an identity rotation. This is a latent-trap/perf fix; expected zero downstream workspace impact.

Original Prompt

Click to expand starting prompt

The following line in @autogalaxy/profiles/geometry_profiles.py is probbaly a bug

Type: bug
Target: PyAutoGalaxy
Difficulty: small
Autonomy: safe
Priority: normal
Status: formalised

The following line in @autogalaxy/profiles/geometry_profiles.py is probbaly a bug:

• if self.class.name.startswith("Sph"):

• Can you check you agree and apply a fix.

[Jammy2211]

Shipped and merged: PyAutoGalaxy#557 (merge 66fe21c). The inverse transform now uses endswith("Sph") matching its mirror; regression test control-verified. pending-release — reaches PyPI with the next release. Merge authorized by human 2026-08-06.

[Jammy2211]

Shipped + Smoke Test Results — 2026-08-06

PR #557 merged (66fe21c). Full test_autogalaxy suite: 1028 passed. Regression test control-verified (fails on pre-fix source).

Workspace	Passed	Failed	Skipped	Total
autofit_workspace	7	0	1	8
autogalaxy_workspace	13	0	0	13
autolens_workspace	34	0	1	35
autolens_workspace_test	22*	0	2	24
euclid_strong_lens_modeling_pipeline	6	0	0	6
HowToLens	50	0	0	50

* autolens_workspace_test initially reported failures in the jax_likelihood script family, traced to the smoke harness not honouring the scripts' in-file ENV: jax full_datasets declarations (capped datasets + JAX disabled against full-resolution reference assertions), compounded by a concurrent smoke run on the same checkouts. Controls with the canonical per-script env (build_env_for_script) on merged main all pass:

• imaging/jax_likelihood/mge.py — PASS, log-likelihood matches reference -86283.10393 exactly
• imaging/jax_likelihood/lp.py — PASS
• imaging/jax_likelihood/rectangular.py — PASS
• imaging/jax_likelihood/smbh.py — PASS

The same scripts failed identically in the concurrent run on pre-#557 code, independently confirming the failures were environmental, not from this change.

Generated by the PyAutoLabs agent workflow.