Type: feature Target: workspaces Repos:
- workspaces Difficulty: large Autonomy: supervised Priority: high Status: formalised Epic: cluster-strong-lensing Phase: 9 Parent: draft/feature/autolens/source_cluster_arc.md Filed: 2026-08-19 (backfilled from git)
Part of the Source & Cluster arc (phase 9 of 12), gated on phases 1, 3, 5-8. User request (verbatim): "Make extra sure source_science.py cluster is robust, e.g. all examples of magnifications above but no mesh support yet."
Audit finding: scripts/cluster/ has NO source_science.py at all (only start_here,
modeling, simulator, plot, likelihood_function, csv_api, mass_parameterizations,
lenstool/). Cluster magnification today is just the PointSolver
magnification_threshold arg and the μ²-weighted source-plane χ² narrative in
likelihood_function.py:381-481.
Work — new scripts/cluster/source_science.py (point-source tier only; pixelized
cluster sources are phase 10):
- Point μ at each multiple image per source, multi-plane (per-source plane_redshift — the #678 final-plane-default trap must be spelled out in prose).
- Point μ at arbitrary (y,x) — the LEGGOS clump-centroid pattern.
- Area magnification of an arc segment via the LEGGOS Eq. 6-7 pixel-inversion sum, with the critical curve as segmentation boundary (phase 3/6 machinery).
- Errors on all of the above via posterior draws (phase 7 pattern, multi-plane).
- Magnification maps + uncertainty maps per source plane (phase 8 plotting).
- Robustness: run against the A2744 start_here fit and the simulated cluster; smoke-viable subset (cluster scripts are currently not smoke-able — keep the heavy draws behind the standard guard). Mirror the group/source_science.py structure for the regime ladder; cite LEGGOS II, Richard+17 (HFF magnification-map deliverables) and Atek+15 (magnification as source- science tool) in the prose. Coordinates with draft/docs/workspaces/cluster_regime_narrative.md (narrative alignment) — this phase delivers the science example; that draft delivers the regime prose.