fix: correct unphysical ell_comps in tutorial 3 model-fitting exercise - #65

Summary

The manual model-fitting exercise at the end of tutorial_3_fitting shipped ell_comps=(0.5, 0.9) as its deliberately-poor starting guess. That pair has magnitude f = 1.0296, and the axis ratio is q = (1 - f) / (1 + f), so it implies q = -0.015 — an ellipse with a negative axis ratio, which has no geometric meaning.

The value was always unphysical. It only became visible when PyAutoGalaxy added the validate_ell_comps guard on EllProfile (the single base every elliptical profile inherits), which rejects f >= 1.

Changed to (0.5, 0.5): f = 0.707, q = 0.17. Still a wildly wrong starting point against the simulator's true bulge (axis_ratio=0.9, angle=45deg, i.e. ell_comps ~ (0.053, 0.0)), so the exercise keeps its "adjust these to improve the fit" pedagogy — the neighbouring centre=(1.0, 10) is left deliberately bad too. The profile is now geometrically valid.

Why fix the value rather than the guard

The guard is correct and correctly placed: q = (1 - f) / (1 + f) is a valid axis ratio in (0, 1] only while f < 1; at f == 1 the ellipse degenerates to q == 0 and beyond it q goes negative. Relaxing it would re-admit profiles that return a finite but meaningless image.

Scope

Applied to the script and to both generated mirrors, so the run_scripts and run_notebooks legs agree:

• scripts/chapter_1_introduction/tutorial_3_fitting.py
• notebooks/chapter_1_introduction/tutorial_3_fitting.ipynb
• markdown/chapter_1_introduction/tutorial_3_fitting.md

Evidence

Surfaced by PyAutoHeart Workspace Smoke run 31356506626 (2026-08-10). The howtogalaxy / chapter_1_introduction leg had exactly one failure — this script — with the other six passing:

scripts/chapter_1_introduction/tutorial_3_fitting.py ...   FAIL (19.0s)
ValueError: ell_comps must satisfy ell_comps[0]**2 + ell_comps[1]**2 < 1;
got (0.5, 0.9), whose magnitude is np.float64(1.0295630140987)

An AST scan of 454 ell_comps literals across autogalaxy_workspace, autolens_workspace, HowToGalaxy, HowToLens and both *_workspace_test repos found no other violating literal and no out-of-range axis_ratio conversion, so this is the complete literal surface.

Not in scope

The other red legs in that run are two different causes, neither fixed here:

• The guides legs fail on Nautilus prior draws (e.g. np.float64(-0.7446446619131553)), not literals — already fixed by PyAutoGalaxy#568, which makes ModelParameterException a FitException so searches resample. That run started 18 hours before #568 merged.
• The *_workspace_test aggregator legs fail on (1.0, 1.0) from PyAutoFit's mock scaffolding — filed as a PyAutoMind bug prompt.

🤖 Generated with Claude Code

Generated by Claude Code