[Jammy2211]

Summary

Two follow-ups to PyAutoLabs/PyAutoLens#712 (#480), which merged earlier today.

1. Unblocked an existing prompt (not a new filing)

draft/feature/workspaces/restore_multiple_sources_lensing_of_lens.md was already in the backlog, filed 2026-04-28 straight from #480's own text, so this updates it rather than filing a duplicate.

• Marked UNBLOCKED — fix: repos_sync check mode never checks root AGENTS.md routing table / WORKFLOW owner map #480 shipped in PyAutoLens#712 (c1bba66), issue closed, pointing at complete/2026/08/point-solver-magnification-plane-redshift.md.
• Difficulty: too-large → large — that rating carried the "first fix fix: repos_sync check mode never checks root AGENTS.md routing table / WORKFLOW owner map #480" leg, now done.
• Gave it a title. The first line was a truncated URL sentence (# Once https://github.com/.../480 is fixed (PointSolver), so it rendered on the dashboard with no readable name.
• Added step 7 — the no_run.yaml removal. The original six steps never mentioned it, yet it is the step that actually re-enables the example: both point_source/features/multiple_sources/{simulator,modeling} stay out of harness execution however good the code is until those entries go.
• Cross-linked with draft/bug/pyautolens/point_source_json_datasets_record_no_regime.md, which is parked behind exactly that step.

The validation it asked for and never got

The prompt recorded real doubt about #480's diagnosis — "I'm a bit unsure I totally buy it, so do some explicit tests which more directly remove the mass profile but also edit the magnification_threshold setting". Both were run, and the doubt pointed at something real: the bug is configuration-dependent, and removing source_0's mass profile masks it.

Solving for source_0 at plane_redshift=1.0 on the #480 tracer:

source_0	µ at z=1.0 (correct)	µ at z=2.0 (what the old code measured)	old code's result
with mass (R_E=0.2)	27.9 9.06 23.1 16.0	7e-3 2e-3 1e-3 2e-3	all rejected → 0 images
without mass	27.9 9.06 23.1 16.0	1.86 1.54 2.38 2.56	survive → 4 images

Dropping the mass was not incidental to the workaround — it removes the very condition that triggers the bug. That is why the simplified example worked, and why step 1 restoring that mass is exactly what would have re-broken it before #712. µ at z=1.0 is identical in both rows: correct physics, since a deflector at plane j does not affect the mapping to plane j.

2. New prompt: cross-validate multi-plane ray tracing

draft/test/autolens/cross_validate_multi_plane_ray_tracing.md (large, high, Model: Fable).

Multi-plane ray tracing has no second opinion. Single-plane quantities do — test_isothermal.py::test__shear_yx_2d_from__matches_via_hessian pins the analytic closed form against the Richardson Hessian of the deflections, and its docstring says why: the two are wrong in different ways, so agreement is evidence where either alone is only self-consistency. traced_grid_2d_list_from and its scaling factors are tested only against themselves, or against literals the same code produced.

Motivated by both bugs this session surfaced in that machinery. #480 lived four months in a core path, and the ray-traced Jacobian that finally settled it — agreeing with float64 JAX autodiff to 1.8e-09 — was a one-off rather than a fixture. The control arm of that same check then exposed the Richardson-step bug, filed separately.

The prompt pairs the implementation against the published formalism rather than re-deriving from our source, names the one certain reference (inspirehep.net/literature/419263, which lens_calc.py already cites for its equations 55 and 60), and explicitly marks the remaining references as leads to verify rather than established citations. It also flags the convention trap up front: deflections_between_planes_from is scaled against the final plane, so truncating a tracer to planes ≤ j gives 1.86 where the answer is 27.9 — measured during #480, a convention mismatch rather than a bug, and exactly what the guide exists to document.

Ends as an @autolens_workspace guide routed through /workspace; library tests and guide likely split into two PRs, library first.

Test Plan

• scripts/lifecycle.py check — OK after each change.
• pyauto-brain intake dashboard --check — "dashboard.md + dashboard.html are current" (136 prompts).
• Confirmed no duplicate prompt existed before updating (1) or filing (2).
• (2) re-homed draft/test/autogalaxy/ → draft/test/autolens/: the classifier resolved PyAutoGalaxy first, but the multi-plane machinery is PyAutoLens's tracer_util.

Prompts and generated pages only; no code, no registry state changes.