mind: file autofit mock all-ones ell_comps bug from the workspace smoke triage

[Jammy2211]

Summary

Files one new bug prompt from triaging the 2026-08-10 Workspace Smoke ell_comps failures, plus the regenerated dashboard.md.

• draft/bug/autofit/mock_all_ones_parameters_break_ell_comps_guard.md
• dashboard.md (regenerated with pyauto-brain intake --apply dashboard; backlog 135 → 136)

Why

The smoke failures split three ways. Two are already resolved:

• Sampler prior draws (guides legs) — fixed by PyAutoGalaxy#568, which makes ModelParameterException a FitException so searches resample instead of terminating.
• One genuinely unphysical shipped literal — ell_comps=(0.5, 0.9) in HowToGalaxy tutorial_3_fitting, corrected in fix: correct unphysical ell_comps in tutorial 3 model-fitting exercise HowToGalaxy#65.

The remaining seven aggregator legs across the two *_workspace_test repos fail on ell_comps=(1.0, 1.0). PyAutoGalaxy#568 does not help there, because the aggregator rebuilds instances from stored samples outside any likelihood call, so there is no resample path. That is what this prompt captures.

What the prompt records

Narrowed to PyAutoFit's mock scaffolding, which fills every parameter with a blanket 1.0 — a safe placeholder for most parameters and an invalid value for any ell_comps:

• MockSamples.default_sample_list → kwargs={path: 1.0 for path in self.model.paths}
• MockSamplesSummary.__init__ → self._kwargs = {path: 1.0 for path in self.model.paths}

The exact call site is not pinned — closing it needs an environment that can run the full autogalaxy stack, which a cloud session cannot (autoarray/jax/numba will not install). The prompt says so explicitly rather than implying the diagnosis is complete.

It also records a plausible-but-disproven hypothesis so it is not re-tread: the parameter_list_with_physical_ell_comps helper copy-pasted into all seven scripts looks broken, but ell_comps is a TuplePrior, so its path contains a bare 'ell_comps' element and the check matches. Reproduced against installed autofit — it yields (0.1, 0.1). Changing that helper would be a no-op.

Finally it flags the blast radius (roughly 55 call sites in PyAutoFit alone, plus the PyAutoGalaxy and PyAutoLens suites) so nobody patches shared mock infrastructure without running all three.

Notes

• Filed under draft/ only. active.md, active/ and complete/ are untouched — those belong to $start-dev / $create-issue and the ship skills.
• lifecycle.py check reports pre-existing drift (active/pixelized_prodigy_laptop_gpu_phase_1_compatibility.md has no registry entry) from an earlier commit. Unrelated to this change and left alone; lifecycle_drift.yml is path-filtered to active/** / complete/**, none of which this PR touches.

🤖 Generated with Claude Code

---

Generated by Claude Code