[Jammy2211]

Summary

Phase 3 of the model-figures epic, PyAutoFit half (PyAutoLens#736; phases 1–2: #1606, #1614). PyAutoFit cannot know which lens parameters a fit solves rather than samples, and solved_paths= could only re-tag rows that already exist — a linear light profile has no intensity slot at all. This PR adds a class-declared protocol: a class sets __solved_parameters__ = ("intensity",) and graph_spec appends a solved row for each declared name that is not a slot (re-tagging it when it is), with provenance solved-by-fit and in_model_info=False. No lens class name appears in PyAutoFit; PyAutoGalaxy and PyAutoArray declare the attribute on LightProfileLinear, PointSolved and Pixelization in the sibling PRs. Because the per-search model.png calls ModelPlotter(model) bare, it inherits the semantics automatically.

Also: solved_paths= is now additive (an unmatched path is synthesised on its owner, never silently ignored); rule R7 drops af.Model(cls) for any int/float subclass so a free Redshift renders as an ordinary pill on its galaxy card; the presentation shows name · solved and a footer N parameters solved during fitting; counts["solved"] added.

Bug fix found on the way: Model.constructor_argument_names filtered only self, so for a class with a custom __new__(cls, …) (autogalaxy.Redshift(float)) cls leaked in as a parameter and af.Model(Redshift) overwrote its own .cls with a ConfigException. The bound first argument is now dropped whether it is self or cls.

API Changes

Added the __solved_parameters__ class-attribute protocol consumed by af.GraphSpec / af.ModelPlotter; solved_paths= became additive; af.Model of an int/float subclass is a row on its owner in the figure; af.Model(cls) with a custom __new__ no longer mis-parses cls as a parameter. No symbol removed or renamed. See full details below.

Test Plan

• test_autofit/graph_spec 66 passed (14 new in test_solved.py), test_autofit/model_figure 80 passed (4 new), test_autofit/non_linear/paths/test_model_figure_output.py 9 passed — serial.
• Full test_autofit 2732 passed / 2 skipped; black + pyflakes clean.
• Verified in the task worktree against the real autogalaxy.Redshift: af.Model(ag.Redshift).cls is ag.Redshift, config prior resolves, instance_from_prior_medians().redshift == 1.5.
• CI: lib-tests 3.12 / 3.13 / no-jax, docs.

Full API Changes (for automation & release notes)

Added

• Protocol: a class attribute __solved_parameters__: tuple[str, ...] on any modelled class; af.GraphSpec.from_model appends a solved row (provenance solved-by-fit, in_model_info=False) per declared name that is not a slot and re-tags a declared name that is.
• Provenance.kind value solved-by-fit.
• GraphSpec.counts["solved"] (solved scalars on the uncollapsed tree).
• Presentation: name · solved pill text; footer phrase N parameter(s) solved during fitting.

Changed Behaviour

• GraphSpec.from_model(..., solved_paths=): an entry matching no row is synthesised as a solved row on its owner node (root when the owner is not a component) instead of being ignored.
• Rule R7 (Model(int)/Model(float) is a row on its owner) now applies to any int/float subclass (e.g. autogalaxy.Redshift).
• Model.constructor_argument_names drops the bound first argument whether it is self or cls — fixes af.Model of a class with a custom __new__.

Migration

• None. Downstream libraries opt in by declaring __solved_parameters__ on their classes (PyAutoArray / PyAutoGalaxy sibling PRs).

Generated by the PyAutoLabs agent workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_0178yLU9v19GtACcRPFggjGa

[Jammy2211]

Sibling PRs (merge order)

1. PyAutoFit — feat: __solved_parameters__ protocol for model figures + Model cls fix (PyAutoLens#736) #1615
2. PyAutoArray — feat: declare Pixelization.__solved_parameters__ for model figures (PyAutoLens#736) PyAutoArray#550
3. PyAutoGalaxy — feat: solved parameters on linear profiles + PointSolved; model cookbook figures (PyAutoLens#736) PyAutoGalaxy#616
4. PyAutoLens — feat: model figures phase 3 — real-class lens acceptance + cookbook/overview figures (#736) PyAutoLens#737