refactor: mass profiles spring clean — xp threading and consistency

[Jammy2211]

Overview

Code quality sweep across the mass profiles module after the MGE/CSE fallback mechanism landed (#449). Audit utility files for hardcoded np that should be xp, fix method signature inconsistencies, and verify decorator stacking order. Phase 4 of the mass profiles refactoring epic (#445).

Plan

• Audit and fix xp threading in dark/mcr_util.py and dark/ludlow16.py (4+1 bare np.* calls)
• Add xp=np to PowerLawMultipole methods (deflections_yx_2d_from, convergence_2d_from, potential_2d_from)
• Verify decorator consistency across all mass profile methods
• Document parameter naming inconsistencies (mass_to_light_ratio vs mass_to_light_ratio_base)
• Run full mass profile test suite

Detailed implementation plan

Affected Repositories

• PyAutoGalaxy (primary)

Work Classification

Library

Branch Survey

Repository	Current Branch	Dirty?
PyAutoGalaxy	main	modified CLAUDE.md only

Suggested branch: feature/mass-profiles-spring-clean
Worktree root: ~/Code/PyAutoLabs-wt/mass-profiles-spring-clean/

Key Files

• dark/mcr_util.py — 4 bare np.* calls to fix
• dark/ludlow16.py — 1 bare np.* call
• total/power_law_multipole.py — missing xp=np on 3 methods

Original Prompt

Click to expand

Spring clean the mass profiles module. xp threading audit, dead code removal, decorator consistency, parameter naming standardization.

[Jammy2211]

Audit Complete — 2026-05-26

Status

No code changes needed — closing without a PR.

Audit Findings

xp threading: Already complete. nfw_hk24_util.py has 0 bare np.* calls. ludlow16.py is fully xp-native. mcr_util.py has 4 bare np.* in the Duffy function — intentionally numpy-only (the Ludlow path has full xp support).

Decorator consistency: All profiles follow legitimate patterns — some use @aa.decorators.transform (MGE-based, computation in profile frame), others handle transformation internally. Both work correctly (406 tests pass).

Method signatures: All concrete convergence_2d_from, potential_2d_from, deflections_yx_2d_from accept xp=np and **kwargs. PowerLawMultipole already had xp=np (multi-line def, initial grep missed it).

Dead code: No dead zero-returning methods remain — all replaced in Phase 3 (#449).

Parameter naming: mass_to_light_ratio vs mass_to_light_ratio_base (GaussianGradient vs SersicGradient) is a known asymmetry but changing it would be a breaking API change — documented, not fixed.