test: every external field → fields= (al.MassField); one galaxy-attached legacy regression [draft until release] - #322
Conversation
…y-attached legacy regression kept Phase 3 of the MassField epic (autolens_workspace#559), re-scoped 2026-09-17: parity, JAX-likelihood, multi-dataset, substructure and interop scripts compose the shear / sheets as an al.MassField in `fields=`. misc/mass/galaxy_attached_legacy.py is the single script allowed to attach a field to a Galaxy (tracer parity + workspace-side identifier pin); coolest_round_trip.py keeps its legacy-peel case and gains a MassField case. Needs the PyAutoGalaxy #621 + PyAutoLens #742 release — draft PR, pending-release. Refs PyAutoLabs/autolens_workspace#559 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…pace
PyAutoLens #744 made the model's `fields=` slot accept a bare `al.MassField`
alongside a collection. This adopts the flat form everywhere in this repo, so a
model now reads
fields=field
instead of `fields=af.Collection(field=field)`, and prior paths read
`fields.shear.gamma_1` instead of `fields.field.shear.gamma_1`.
Sites. An AST walk of the branch tip (63c3323) found 66 model-side
`fields=af.Collection(field=...)` sites across 64 files. 65 are migrated here:
63 in the bulk pass (61 files) and 2 in the delicate files
(`multi_galaxy/composition_mge.py`, `multi_galaxy/model_fit.py`). The 66th is
deliberately retained -- see below. A re-walk of the tree confirms the census is
conserved at 103 `fields=` sites overall (66 -> 65 flat + 1 retained, 19
attr-chain, 16 tracer list-literals, 2 tracer `all_fields`), that no file mixes
the two forms, and that `fields.field` no longer appears anywhere in the repo,
docstrings included.
`multi_galaxy/composition_mge.py`. Three coupled edits: the model-composition
site, the path-grouping assertions at :159-:170 (the second path element under
`fields` is now the mass profile `shear`, with no intervening collection key),
and the identifier pin at :263. The pin was RECOMPUTED, not adjusted:
OLD b99831e66dd27eee314113e8e58235b6 (collection form)
NEW 29f82bd3de24984b94657c328b64c3be (flat form)
computed with the cwd at this repository's root (`autolens_workspace_test/`, the
same cwd every script here runs from), because an identifier is a function of the
model AND the active PyAutoFit configuration. The docstring at :237 now records
that. This script is on the smoke gate, so a wrong pin would go red immediately;
it passes (3.0 s) in a 32/32 local smoke run.
`misc/mass/galaxy_attached_legacy.py`. Its galaxy-attached pin
`35ebe9353118bcc0c7b2d577ce2639ee` is UNCHANGED -- that is the backwards-
compatibility regression and must not move. Its second model (:189) deliberately
KEEPS the `fields=af.Collection(field=...)` form: with every other model in
`autolens_workspace` and `autolens_workspace_test` now flat, it is the
workspaces' last regression witness that the library still accepts a collection
in that slot. Only a docstring saying so was added.
Left untouched. The 19 chained `fields=<result>.{model,instance}.fields` stages
in the four `misc/database/scrape/*_jax` SLaM scripts (chaining is shape-
transparent and carries the flat form unchanged); the 16 tracer-side
`al.Tracer(fields=[field])` list literals; the 2 genuine multi-field tracer sites
(`all_fields`); and `misc/interop/coolest_round_trip.py`, whose `.fields[0]`
reads are tracer-side list indexing.
Identifiers. Flat models compose differently, so they write to new `output/`
directories. Existing trees keyed on the old identifiers are orphaned and are not
resumed; no pin is aliased to an old value and no directory is renamed.
Phase 3 of the epic (the galaxy-attached -> collection migration) is also
unmerged on this branch, so this branch takes the repository from the
galaxy-attached form straight to the flat form and never ships the intermediate
collection form from here.
Verification (local stack: PyAutoLens 478213e78, PyAutoGalaxy 90e757d3,
PyAutoFit 7c0e79aab):
- Smoke: 32/32 PASS (979 s, `profile_smoke.yaml`), including `composition_mge.py`
and both latent integration smokes.
- `misc/latent/latent_integration_smoke_jax.py` -- the script whose CI leg was red
on this PR before PyAutoGalaxy 90e757d3 -- PASSES under the flat form (56.4 s),
writing exactly one `files/latent/latent_summary.json` with all 8 latents finite
and non-zero.
- Chained stages: `misc/database/scrape/slam_general_jax.py` hand-run (these four
scripts are SLOW-skipped in `no_run.yaml` and have no automated coverage).
All five stages carry a top-level `fields` under the flat form, with the
free/fixed split intact: `source_lp[1]` (seed), `source_pix[1]`
(`.model.fields`) and `mass_total[1]` (`.model.fields`) show
`fields MassField (N=2)` with the seed's own priors propagated;
`source_pix[2]` and `light[1]` (`.instance.fields`) show
`fields MassField (N=0)`.
- Equivalence: `prior_count` unchanged (21 == 21) and prior-path sets equal after
normalising `fields.field.` -> `fields.`, cwd at this repository's root.
- Every changed file byte-compiles.
No generated artifacts changed: this repo has no notebooks
(`config/build/visualise_notebooks.yaml` is `[]`), and the only generator
(`gallery/gallery_run.sh`) writes solely to `output/` and `scripts/*/images/`,
both gitignored with nothing tracked under either.
Issue: #561
Lands on PR #322 (draft), alongside the phase 3 commit 63c3323.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The flat
|
| Script | OLD | NEW |
|---|---|---|
scripts/multi_galaxy/composition_mge.py (:263) |
b99831e66dd27eee314113e8e58235b6 |
29f82bd3de24984b94657c328b64c3be |
scripts/misc/mass/galaxy_attached_legacy.py (:159) |
35ebe9353118bcc0c7b2d577ce2639ee |
UNCHANGED — the backwards-compatibility regression, must not move |
The composition_mge.py pin was recomputed, not adjusted: the baseline was reproduced first (old pin green pre-edit), then the new value read from the genuinely edited script's own assertion output. It was computed with the cwd at this repository's root, the same cwd every script here runs from, because an identifier is a function of the model and the active PyAutoFit configuration. The docstring at :237 now records that. composition_mge.py is on the smoke gate, so a wrong pin would go red immediately — it passes in 3.0 s.
That file took four coupled edits: the model-composition site, the path-grouping assertions at :159-:170 (the second path element under fields is now the mass profile shear, with no intervening collection key), the identifier pin, and the docstring/prose.
Evidence
- Smoke: 32/32 PASS (979 s,
profile_smoke.yaml), includingcomposition_mge.pyand both latent integration smokes.autolens_workspace: 38/38 scripts + 2/2 notebooks. - Equivalence 8/8:
prior_countunchanged,model.identifierchanged, prior-path sets equal after normalisingfields.field.→fields.. (composition_mge: 17/17, symmetric difference empty. Repo-wide witness: 21 == 21.) - Log-likelihood bit-identical for the same instance under both forms —
-37796.54513586828. - Chained stages,
misc/database/scrape/slam_general_jax.pyhand-run (these four are SLOW-skipped inno_run.yamland have no automated coverage): all five stages carry a top-levelfieldsunder the flat form with the free/fixed split intact —source_lp[1](seed),source_pix[1]andmass_total[1](.model.fields) showfields MassField (N=2)with the seed's own prior ids 25/26 propagated;source_pix[2]andlight[1](.instance.fields) showfields MassField (N=0), gamma 0.0.slam_start_here.pyhand-run PASS with all 5 stages carryingfields.shear. - Every changed file byte-compiles.
- No generated artifacts changed: this repo has no notebooks (
config/build/visualise_notebooks.yamlis[]), and the only generator (gallery/gallery_run.sh) writes solely tooutput/andscripts/*/images/, both gitignored with nothing tracked under either.
Local stack: PyAutoLens 478213e78, PyAutoGalaxy 90e757d3, PyAutoFit 7c0e79aab.
The subhalo sensitivity fix (in #562)
Both hasattr guards in base_model_narrow_priors_from() were rewritten. Under the flat form the OLD guard silently skipped a deliberate shear prior re-centring — base_model.fields.field raises AttributeError inside hasattr, which returns False: no error, no log, the re-centring never ran. Proven to fire now: gamma_1 UniformPrior(-0.3, 0.3) → UniformPrior(0.0734, 0.1734), gamma_2 UniformPrior(-0.3, 0.3) → UniformPrior(-0.1067, -0.0067). Those scripts remain parked in no_run.yaml and out of smoke_tests.txt.
SEPARATELY — a pre-existing bug fix, not part of the migration (in #562)
scripts/multi_dataset/features/one_by_one/modeling.py re-collected af.Collection(dataset_model=..., galaxies=result.instance.galaxies) and dropped fields entirely, so its "Second Dataset Offset" stage — documented as keeping all lens model parameters fixed — silently fitted with no external shear. The author's own commented-out block above it already uses the correct fields=result.instance.fields. One line added. Pre-existing on main, approved by the human.
This PR's red CI leg is NOT this sweep
Bisected to a JAX 0.11.2 × unfixed-PyAutoGalaxy regression: TypeError: functools.partial ... is not a valid JAX type, raised in the zero-contour solver via the effective_einstein_radius latent. The unmodified main version of the script fails identically in a CI-matching venv. It is fixed by PyAutoGalaxy 90e757d3, which merged roughly 10.5 hours after that CI run, so a CI re-run should clear it.
The flat form passes under JAX 0.11.2 on current mains: misc/latent/latent_integration_smoke_jax.py — the exact script whose leg was red — PASSES under the flat form (56.4 s), writing exactly one files/latent/latent_summary.json with all 8 latents finite and non-zero.
PyAutoHeart ack
Heart was RED at ship time. Verdict, verbatim:
RELEASE READINESS ✗ RED score 60
✗ release validation FAILED (stage integrate)
! manifest drift: remote-session blocks (generated) — 2 mismatch(es) vs PyAutoMind/repos.yaml
Both conditions are pre-existing and unrelated to these two workspace repos. The human was asked:
"Heart is RED (release validation FAILED, stage integrate) plus a YELLOW manifest drift — both pre-existing and unrelated to these two workspace repos. Do you authorize pushing the two commits and opening/updating the PRs despite the RED?"
and answered:
"Yes — push and open/update PRs"
That authorization covers push and PR-open/update only. Merge and release remain human-only and are NOT authorized by it. This PR stays draft with pending-release until PyAutoLens #744 (478213e78) is on PyPI.
Cross-references
Issue autolens_workspace#561 · phase-3 issue #559 · merged autolens_workspace#560 (c79c8d3) · PyAutoLens #743/#744 (478213e78) · euclid pipeline #90 (9cdee7b)
🤖 Generated with Claude Code
1 participant
Summary
DO NOT MERGE until release — this PR needs PyAutoGalaxy #621 (
ag.MassField) and PyAutoLens #742 (Tracer(fields=), analysisfieldsslot,model_util.mass_field_from, LOSfields_from, COOLEST 1:1) on PyPI. The workspace follows the released stack, notmain; CI here is green only because the smoke workflow clones the librarymainbranches. Opened as a draft for that reason;gh pr ready+/prmafter the release.Phase 3 of the MassField epic (PyAutoMind
draft/feature/autogalaxy/mass_field_epic.md), re-scoped on the human's 2026-09-17 ruling: the user-facing API from here on is fields in the model as a separate thing. EveryExternalShear/MassSheet/ExternalPotentialthat was attached to aGalaxyinautolens_workspace_testnow lives in anal.MassFieldin its ownfields=slot of the model (fields=af.Collection(field=af.Model(al.MassField, redshift=…, shear=…))) and of the tracer (al.Tracer(galaxies=[…], fields=[field])). SLaM stages chainfields=<result>.model|instance.fields; results are read asresult.instance.fields.field.shear; aggregator paths arefields.field.shear.<param>. Numerically identical (deflections sum over planes — parity asserted, see Test Plan); committed datasets untouched.tracer.galaxiesnever holds a field, so positional galaxy access is unaffected (themulti_galaxy/shear_galaxygalaxy leaving the list is the one place indices moved; re-checked). A migrated example composes a different model and therefore gets a new result identifier; oldoutput/folders are not resumed. The library keepsal.Galaxy(shear=…)working and unwarned for users' own scripts.Heart readiness at ship: RED — "install verification FAILED (testpypi; checks F)"; "release validation FAILED (stage integrate)" (+ YELLOW manifest drift). Neither concerns this repo; the human acknowledged these exact reasons on 2026-09-18 for commit/push/draft PR-open only. Merge stays human and release-gated.
The one script allowed to attach a field to a
Galaxyis the newscripts/misc/mass/galaxy_attached_legacy.py(the human's "maybe an autolens_workspace_test integration test"): tracer parity vs itsMassFieldtwin (deflections 4.4e-16, convergence 0, potential 1.8e-15) and the workspace-side identifier pin35ebe9353118bcc0c7b2d577ce2639eefor the galaxy-attached lens+shear/source model (computed from this repo'sconfig/; the library's own pin of the same model under its shipped config isfef2697b5c32ba56bb18a7baecb7b0f6— identifiers depend on the active config).coolest_round_trip.pykeeps its legacy galaxy-attached case as phase 2's legacy-peel regression. Two judgement calls worth a glance:composition_mge.py's identifier constant is re-anchored because its model now composes the shear as a field (a different model by design), and itsunique_prior_pathsround-trip is compared as a set becausefrom_dictre-orders prior ids for any model with a second top-level collection (reproduced withextra_galaxies=; pre-existing PyAutoFit behaviour, identifier unaffected). Three files (multi_dataset/simulator.py,multi_galaxy/model_fit.py,multi_galaxy/jax_likelihood/lp.py) were already non-black at HEAD and were not reformatted. Substructure tests build per-planeMassFields;substructure_util.galaxies_to_halo_arraysaccepts them unchanged.Scripts Changed
scripts/imaging/jax_grad/delaunay.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_grad/knn.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_grad/lp.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_grad/mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_grad/pixelization.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_grad/regularization.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/delaunay.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/delaunay_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/delaunay_near_caustic.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/lp.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/mge_group.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/multipole.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular_dspl.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular_dspl_rtu.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular_mge_rtu.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/rectangular_rtu.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/jax_likelihood/smbh.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/simulator/dspl.py—al.MassField(redshift, shear=...)passed asTracer(fields=[field])(dataset bit-identical)scripts/imaging/simulator/simple.py—al.MassField(redshift, shear=...)passed asTracer(fields=[field])(dataset bit-identical)scripts/imaging/substructure/subhalo.py— per-planemass_sheetgalaxies → per-planeMassFields infields=scripts/imaging/substructure/test_batched_simulate.py— per-planemass_sheetgalaxies → per-planeMassFields infields=scripts/imaging/substructure/test_scan_multiplane.py— per-planemass_sheetgalaxies → per-planeMassFields infields=scripts/imaging/substructure/test_simulate_e2e.py— per-planemass_sheetgalaxies → per-planeMassFields infields=scripts/imaging/visualization/modeling_visualization_delaunay_jit.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/visualization/modeling_visualization_jit.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/imaging/visualization/modeling_visualization_rectangular_jit.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/datacube/delaunay.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/datacube/rectangular.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/datacube/shared_preloads.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_grad/gradient.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/delaunay.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/delaunay_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/lp.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/mge_group.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/rectangular.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/rectangular_dspl.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/rectangular_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/jax_likelihood/rectangular_sparse.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/nufft.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/simulator/dspl.py—al.MassField(redshift, shear=...)passed asTracer(fields=[field])(dataset bit-identical)scripts/interferometer/simulator/simple.py—al.MassField(redshift, shear=...)passed asTracer(fields=[field])(dataset bit-identical)scripts/interferometer/visualization/modeling_visualization_jit.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/interferometer/visualization/visualization_jax.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/database/scrape/general.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/database/scrape/multi_analysis_jax.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/database/scrape/scaling_relation.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/database/scrape/slam_general_jax.py— SLaM stage chainsfields=<result>.model|instance.fieldsinstead of a galaxyshear=kwargscripts/misc/database/scrape/slam_multi_one_by_one_jax.py— SLaM stage chainsfields=<result>.model|instance.fieldsinstead of a galaxyshear=kwargscripts/misc/database/scrape/slam_pix_jax.py— SLaM stage chainsfields=<result>.model|instance.fieldsinstead of a galaxyshear=kwargscripts/misc/interop/coolest_round_trip.py— legacy galaxy-attached case kept (phase 2 legacy-peel regression, post-import checks follow the shear intotracer_back.fields); newMassFieldcasescripts/misc/jax_assertions/delaunay_nn_caps.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/jax_assertions/delaunay_walk.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/latent/latent_integration_smoke.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/latent/latent_integration_smoke_jax.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/misc/mass/galaxy_attached_legacy.py— NEW — the one script allowed to attach a field to aGalaxy: tracer parity vsMassFieldtwin + workspace-side identifier pin35ebe9353118bcc0c7b2d577ce2639ee+ both models fit under test modescripts/misc/mass/sheets.py— new__MassField__section: field tracer equals the direct profile sumscripts/multi_dataset/jax_likelihood/dataset_model.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/delaunay.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/delaunay_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/lp.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/mge_group.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/rectangular.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/rectangular_mge.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/rectangular_mge_rtu.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/rectangular_rtu.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/jax_likelihood/shared_preloads.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/simulator.py—al.MassField(redshift, shear=...)passed asTracer(fields=[field])(dataset bit-identical)scripts/multi_dataset/visualization/imaging.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_dataset/visualization/interferometer.py— galaxy-attached shear →fields=af.Collection(field=af.Model(al.MassField, ...)); prose names theMassFieldscripts/multi_galaxy/composition_mge.py— shear →fields=; identifier pin re-anchored5a3c480d…→b99831e6…(model changed by design);unique_prior_pathsround-trip compared as a set (PyAutoFitfrom_dictid-ordering, not MassField-specific)scripts/multi_galaxy/jax_likelihood/lp.py— shear →fields=(JAX pytree leg;ENV: jax full_datasetskept; literal likelihood unchanged)scripts/multi_galaxy/model_fit.py— shear →fields=; prior-count assertions now require equal lens counts and a 2-priorfields.field.shearUpstream PR
Test Plan
main(PyAutoGalaxy 33714b80 / PyAutoLens 71973806),run_python.py --listwith the smoke profile (PYAUTO_TEST_MODE=2), from the task worktree:autolens_workspace_test subset — 13/13 PASS, 0 FAIL, 0 TIMEOUT (the two
jax_likelihood/lp.pyentries declareENV: jaxand ran with JAX: NumPy == JIT likelihood 626.2235846074775 and 1206.80145178, the pre-migration literals):imaging/jax_likelihood/lp.pyimaging/jax_likelihood/delaunay.pyimaging/jax_likelihood/mge.pyinterferometer/jax_likelihood/rectangular.pymulti_dataset/jax_likelihood/mge.pymulti_galaxy/composition_mge.pymulti_galaxy/model_fit.pymulti_galaxy/jax_likelihood/lp.pymisc/jax_assertions/delaunay_walk.pymisc/latent/latent_integration_smoke.pymisc/mass/sheets.pymisc/mass/galaxy_attached_legacy.pymisc/interop/coolest_round_trip.pyfields=):imaging/simulator/simple.pytracer max abs diff 0.0;galaxy_attached_legacy.pyandmisc/mass/sheets.pyassert field-vs-galaxy parity to ~1e-15.grep -rnE "al\.Galaxy\(.*(shear|mass_sheet|potential)="andgrep -rn shear_galaxyoverscripts/return exactly the sanctioned sites:misc/mass/galaxy_attached_legacy.py(×2) and the legacy case inmisc/interop/coolest_round_trip.py(AST walk over everyGalaxy(...)/af.Model(al.Galaxy, ...)call).notebooks/in this repo;black --checkclean on 74 of 77 changed scripts (3 pre-existing non-black files untouched).main)Closes PyAutoLabs/autolens_workspace#559 (both PRs of the task reference it; the issue closes at
/prm).Generated by the PyAutoLabs agent workflow.
🤖 Generated with Claude Code