render: PointSolver refinement figure + thinner model cross

AGENTS.md

@@ -69,13 +69,23 @@ the full figure set. Every other producer tracks only `dataset.png` and `fit.png
 everything else. Producers with several sub-datasets write one pair per sub-folder
 (`multi_dataset` per band, the datacube per channel, `cluster` per point-source plane). Many
 of these figures are the galaxy-scale plots and look poor on point / cluster / weak data —
-bespoke figures are a follow-up through the Eyes dashboard. `cluster` calls the two
+bespoke figures are a follow-up through the Eyes dashboard. `cluster` calls the
 `PlotterPoint` methods directly instead of `VisualizerPoint`, skipping the untracked tracer and
-galaxy figures; the "Fit Positions" panel still draws the critical curves, which cost ~2 min per
-dataset on the 15-galaxy multi-plane lens (the PNGs match the visualizer's).
+galaxy figures; the "Fit Positions" and point-solver panels still draw the critical curves, which
+cost ~2 min per dataset on the 15-galaxy multi-plane lens (the PNGs match the visualizer's).
+
+**Point solver.** `point_source_image_chi_squared` (via `visualize(quick_update=False)`) and
+`cluster` (per plane) also keep `point_solver.png`: the model's real `PointSolver` refining onto
+its images (`fit_point_dataset: subplot_point_solver`, on in this repo's `plots.yaml`, off in
+the library). `point_source_image_chi_squared/images/visualization/teaching/` holds the same
+figure on a coarse teaching grid from `aplt.plot_point_solver`, plus its looping animation
+`point_solver.gif`.
 
 **What is tracked.** PNG figures under `scripts/<domain>/images/**`, `GALLERY.md` and
-`gallery/viz_manifest.yaml`. The FITS /
+`gallery/viz_manifest.yaml`, plus the teaching GIF animation
+(`point_source_image_chi_squared/images/visualization/teaching/point_solver.gif`). The manifest
+and `GALLERY.md` list PNGs only (the PyAutoEyes read contract), so the GIF is tracked beside its
+PNG still but is not a manifest entry; `render.yml` commits it with the PNGs. The FITS /
 CSV / JSON data products the visualizers also write are gitignored (bulky, not viewable on
 GitHub), as are `output/` and `dataset/**/lensed_source.fits`.
 

GALLERY.md

@@ -2,14 +2,14 @@
 
 <!-- Generated by gallery/gallery_build.py from scripts/<domain>/images/. Do not edit by hand: rerun `bash gallery/gallery_run.sh --all`. -->
 
-Rendered with PyAutoLens `2026.8.17.1` · 68 figures.
+Rendered with PyAutoLens `2026.8.17.1` · 72 figures.
 
 Every figure below is what the PyAutoLens visualizers (`VisualizerImaging`, `VisualizerInterferometer`, `VisualizerPoint`, `VisualizerWeak`) write to a model-fit's `image/` folder, rendered on this repo's datasets with the true model. See [README.md](README.md) for how to improve one.
 
 ## Contents
 
 - [cluster](#cluster)
-  - [visualization](#cluster--visualization) (4 figures)
+  - [visualization](#cluster--visualization) (6 figures)
     - [point_0/](#cluster--visualization--point_0)
     - [point_1/](#cluster--visualization--point_1)
 - [group](#group)
@@ -38,8 +38,9 @@ Every figure below is what the PyAutoLens visualizers (`VisualizerImaging`, `Vis
   - [visualization](#multi_galaxy--visualization) (2 figures)
     - [top level](#multi_galaxy--visualization--top-level)
 - [point_source_image_chi_squared](#point_source_image_chi_squared)
-  - [visualization](#point_source_image_chi_squared--visualization) (4 figures)
+  - [visualization](#point_source_image_chi_squared--visualization) (6 figures)
     - [top level](#point_source_image_chi_squared--visualization--top-level)
+    - [teaching/](#point_source_image_chi_squared--visualization--teaching)
 - [point_source_source_chi_squared](#point_source_source_chi_squared)
   - [visualization](#point_source_source_chi_squared--visualization) (4 figures)
     - [top level](#point_source_source_chi_squared--visualization--top-level)
@@ -61,6 +62,10 @@ Every figure below is what the PyAutoLens visualizers (`VisualizerImaging`, `Vis
 
 ![fit](scripts/cluster/images/visualization/point_0/fit.png)
 
+##### point_solver
+
+![point_solver](scripts/cluster/images/visualization/point_0/point_solver.png)
+
 #### cluster / visualization / point_1/
 
 ##### dataset
@@ -71,6 +76,10 @@ Every figure below is what the PyAutoLens visualizers (`VisualizerImaging`, `Vis
 
 ![fit](scripts/cluster/images/visualization/point_1/fit.png)
 
+##### point_solver
+
+![point_solver](scripts/cluster/images/visualization/point_1/point_solver.png)
+
 ## group
 
 ### group / visualization
@@ -359,6 +368,16 @@ Every figure below is what the PyAutoLens visualizers (`VisualizerImaging`, `Vis
 
 ![image_with_points](scripts/point_source_image_chi_squared/images/visualization/image_with_points.png)
 
+##### point_solver
+
+![point_solver](scripts/point_source_image_chi_squared/images/visualization/point_solver.png)
+
+#### point_source_image_chi_squared / visualization / teaching/
+
+##### point_solver
+
+![point_solver](scripts/point_source_image_chi_squared/images/visualization/teaching/point_solver.png)
+
 ## point_source_source_chi_squared
 
 ### point_source_source_chi_squared / visualization

README.md

@@ -30,11 +30,11 @@ mirroring `autolens_workspace` (a variant gets its own folder):
 | `imaging` | `visualization.py` | full set: parametric + Delaunay source |
 | `interferometer` | `visualization.py` | full set: parametric + Delaunay source |
 | `interferometer_datacube` | `visualization.py` | `dataset.png` + `fit.png` per channel |
-| `point_source_image_chi_squared` | `visualization.py` | `dataset.png` + `fit.png` |
+| `point_source_image_chi_squared` | `visualization.py` | `dataset.png` + `fit.png` + image-with-points pair + `point_solver.png`; `teaching/point_solver.png` + `.gif` (coarse grid) |
 | `point_source_source_chi_squared` | `visualization.py` | `dataset.png` + `fit.png` |
 | `multi_galaxy` | `visualization.py` | `dataset.png` + `fit.png` |
 | `group` | `visualization.py` | `dataset.png` + `fit.png` |
-| `cluster` | `visualization.py` | `dataset.png` + `fit.png` per point-source plane |
+| `cluster` | `visualization.py` | `dataset.png` + `fit.png` + `point_solver.png` per point-source plane |
 | `multi_dataset` | `visualization.py` | `dataset.png` + `fit.png` per band |
 | `weak` | `visualization.py` | `dataset.png` + `fit.png` |
 

_viz_cli.py

@@ -220,12 +220,14 @@ def keep_dataset_and_fit(
 
 
 def mirror_to_scratch(domains: list[str] | None = None) -> Path | None:
-    """Copy freshly rendered PNGs to the opt-in Eyes scratch folder; return its lib folder.
+    """Copy freshly rendered PNGs (and GIF animations) to the opt-in Eyes scratch folder.
+
+    Returns the scratch lib folder.
 
     For on-the-fly figure review before any PR or CI. When ``$PYAUTO_EYES_SCRATCH`` is
     unset this is a silent no-op returning ``None`` (so CI and other machines are
     untouched). Otherwise each domain's ``$PYAUTO_EYES_SCRATCH/<EYES_LIB>/<domain>/`` is
-    deleted and refilled with every PNG under ``scripts/<domain>/images/`` (subpaths
+    deleted and refilled with every PNG and GIF under ``scripts/<domain>/images/`` (subpaths
     below ``images/`` kept), so no stale figure survives a re-render. ``domains=None``
     mirrors every domain that has an ``images/`` folder (``misc`` never does).
     """
@@ -243,12 +245,12 @@ def mirror_to_scratch(domains: list[str] | None = None) -> Path | None:
             continue
         dest = lib_dir / domain
         shutil.rmtree(dest, ignore_errors=True)
-        pngs = sorted(images.rglob("*.png"))
-        for png in pngs:
-            target = dest / png.relative_to(images)
+        figures = sorted([*images.rglob("*.png"), *images.rglob("*.gif")])
+        for figure in figures:
+            target = dest / figure.relative_to(images)
             target.parent.mkdir(parents=True, exist_ok=True)
-            shutil.copy2(png, target)
-        print(f"  [eyes-scratch] {len(pngs)} PNGs -> {dest}")
+            shutil.copy2(figure, target)
+        print(f"  [eyes-scratch] {len(figures)} figures (PNG/GIF) -> {dest}")
     return lib_dir
 
 

config/visualize/plots.yaml

@@ -62,7 +62,8 @@ fit_interferometer:                        # Settings for plots of fits to inter
 point_dataset:                             # Settings for plots of point source datasets (e.g. PointDatasetPlotter).
   subplot_dataset: true                    # Plot subplot containing all dataset quantities (e.g. the data, noise-map, etc.)?
 
-fit_point_dataset: {}                      # Settings for plots of fits to point source datasets (e.g. FitPointDatasetPlotter).
+fit_point_dataset:                         # Settings for plots of fits to point source datasets (e.g. FitPointDatasetPlotter).
+  subplot_point_solver: true               # PointSolver refinement steps at the best fit (point_solver.png); off by default in the library.
 
 weak_dataset:                              # Settings for plots of weak lensing shear catalogues (e.g. PlotterWeak).
   subplot_dataset: true                    # Plot subplot containing all dataset quantities (e.g. the shear field, noise-map, etc.)?

gallery/viz_manifest.yaml

@@ -8,7 +8,7 @@ rendered_with:
   autogalaxy: 2026.8.17.1
   autoarray: 2026.8.17.1
   autofit: 2026.8.17.1
-figure_count: 68
+figure_count: 72
 figures:
 - file: scripts/cluster/images/visualization/point_0/dataset.png
   producer: scripts/cluster/visualization.py
@@ -20,8 +20,14 @@ figures:
   producer: scripts/cluster/visualization.py
   domain: cluster
   source: point_0
-  bytes: 68264
-  sha256: c330570f6373ef9055c8d4dd5b7be3fa8b3f096cdb0117d2aec4b3d5987e43fa
+  bytes: 68170
+  sha256: a77734d4569f512ec14657e5a6fb2c1308a1e49c7b2153e00bf83f2c83f9e860
+- file: scripts/cluster/images/visualization/point_0/point_solver.png
+  producer: scripts/cluster/visualization.py
+  domain: cluster
+  source: point_0
+  bytes: 646052
+  sha256: b0f21b073ab19c91c629504940ecd8fb631dda85107edabfd6345c60e7fbc1b4
 - file: scripts/cluster/images/visualization/point_1/dataset.png
   producer: scripts/cluster/visualization.py
   domain: cluster
@@ -32,8 +38,14 @@ figures:
   producer: scripts/cluster/visualization.py
   domain: cluster
   source: point_1
-  bytes: 65979
-  sha256: aaea68e094f6ff3d142c9a24cb92e750781b43b7768c82f3ff527356150c0f35
+  bytes: 65871
+  sha256: 813846220f1a47d03f88084b92cb93507a91b6410ae72dd0138cf13eda54c667
+- file: scripts/cluster/images/visualization/point_1/point_solver.png
+  producer: scripts/cluster/visualization.py
+  domain: cluster
+  source: point_1
+  bytes: 631672
+  sha256: 357d5dff2da880f3bd0908dba09624d2ee66d4211548f27dd79807452b153299
 - file: scripts/group/images/visualization/dataset.png
   producer: scripts/group/visualization.py
   domain: group
@@ -368,8 +380,8 @@ figures:
   producer: scripts/point_source_image_chi_squared/visualization.py
   domain: point_source_image_chi_squared
   source: ''
-  bytes: 59950
-  sha256: dcf7b7a60281c49244264678c6c15ab0a5636be9c66fd4e555bba2b9fdecab40
+  bytes: 59992
+  sha256: 75919308cfe263625a6d71450f2eb83acc2b13a0b08f1edc3ee97720c4f5f75a
 - file: scripts/point_source_image_chi_squared/images/visualization/fit_image_with_points.png
   producer: scripts/point_source_image_chi_squared/visualization.py
   domain: point_source_image_chi_squared
@@ -382,6 +394,18 @@ figures:
   source: ''
   bytes: 66518
   sha256: 2653419882e5f55d2afe24b067e8d5c9865081aee51517dc05520a12d1787468
+- file: scripts/point_source_image_chi_squared/images/visualization/point_solver.png
+  producer: scripts/point_source_image_chi_squared/visualization.py
+  domain: point_source_image_chi_squared
+  source: ''
+  bytes: 809716
+  sha256: 92252350649f9d7a8698a775afcbf02cc0bc0291f842b5365f7ecea0fb4ec25a
+- file: scripts/point_source_image_chi_squared/images/visualization/teaching/point_solver.png
+  producer: scripts/point_source_image_chi_squared/visualization.py
+  domain: point_source_image_chi_squared
+  source: teaching
+  bytes: 495158
+  sha256: 84f0312c1415db43b8d8b901a2977a24d706697df70523aef122744a7e00a451
 - file: scripts/point_source_source_chi_squared/images/visualization/dataset.png
   producer: scripts/point_source_source_chi_squared/visualization.py
   domain: point_source_source_chi_squared
@@ -392,8 +416,8 @@ figures:
   producer: scripts/point_source_source_chi_squared/visualization.py
   domain: point_source_source_chi_squared
   source: ''
-  bytes: 60925
-  sha256: b9af86e928d6a1e395cc81aca38cffc7b94bb5b3af82484180d37f989b245472
+  bytes: 60766
+  sha256: 386aaf5cf28aa8b69b71a53962a3edde8d5673769ab7eb7f7ae81f1fe2fe9399
 - file: scripts/point_source_source_chi_squared/images/visualization/fit_image_with_points.png
   producer: scripts/point_source_source_chi_squared/visualization.py
   domain: point_source_source_chi_squared

scripts/cluster/visualization.py

@@ -2,7 +2,7 @@
 Visualization: Cluster
 ======================
 
-Renders the two figures ``VisualizerPoint`` writes for each point-source dataset of a
+Renders the figures ``VisualizerPoint`` writes for each point-source dataset of a
 multi-plane cluster fit (two point sources at z=1.0 and z=2.0 behind a cluster lens), so the
 latest version of each lives in git (``GALLERY.md``). A cluster model-fit has one
 ``AnalysisPoint`` per point-source dataset (combined in a factor graph), each visualized
@@ -13,14 +13,18 @@
 
     scripts/cluster/images/visualization/
         point_0/dataset.png, point_0/fit.png   <- PlotterPoint.dataset_point, PlotterPoint.fit_point
-        point_1/dataset.png, point_1/fit.png
+        point_0/point_solver.png               <- PlotterPoint.point_solver (the model's real
+                                                  multi-plane PointSolver refining onto this
+                                                  source's images, steps 0, 1, 2 and final)
+        point_1/dataset.png, point_1/fit.png, point_1/point_solver.png
 
 **Plotter, not Visualizer.** ``VisualizerPoint.visualize`` also renders the tracer and galaxy
-figures this repo does not track, so this producer calls the same two ``PlotterPoint`` methods the
+figures this repo does not track, so this producer calls the same ``PlotterPoint`` methods the
 visualizer calls (``dataset_point(solver=...)`` in ``visualize_before_fit``;
-``fit_point(quick_update=True)`` in ``visualize``), writing the figures the visualizer would. The
-"Fit Positions" panel draws the tracer's critical curves, which ``fit_point`` computes on a 100x100
-grid covering the solver grid: for this 15-galaxy multi-plane lens that costs ~2 min per dataset.
+``fit_point(quick_update=True)`` and ``point_solver`` in ``visualize``), writing the figures the
+visualizer would. The "Fit Positions" and point-solver panels draw the tracer's critical curves,
+computed once per dataset (as the visualizer does) on a 100x100 grid covering the solver grid: for
+this 15-galaxy multi-plane lens that costs ~2 min per dataset.
 
 The point plotter writes ``dataset_point.png`` (renamed to ``dataset.png``); every other file
 the visualizer writes is deleted.
@@ -62,6 +66,8 @@ def _repo_root() -> Path:
 
 import autofit as af
 import autolens as al
+from autolens.imaging.plot.fit_imaging_plots import _compute_critical_curve_lines
+from autolens.lens.plot.critical_curves import critical_curve_grid_from
 from autolens.point.model.plotter import PlotterPoint
 
 from _viz_cli import ensure_dataset, keep_dataset_and_fit
@@ -121,11 +127,22 @@ def _repo_root() -> Path:
     t0 = time.perf_counter()
     plotter = PlotterPoint(image_path=paths.image_path, title_prefix=analysis.title_prefix)
     plotter.dataset_point(dataset=analysis.dataset, solver=analysis.solver)
+    fit = analysis.fit_for_visualization(instance=instance)
+    ip_lines, ip_colors, sp_lines, sp_colors = _compute_critical_curve_lines(
+        fit.tracer, critical_curve_grid_from(dataset=fit.dataset, solver=solver)
+    )
     plotter.fit_point(
-        fit=analysis.fit_for_visualization(instance=instance),
+        fit=fit,
         quick_update=True,
+        image_plane_lines=ip_lines if ip_lines is not None else [],
+        image_plane_line_colors=ip_colors if ip_lines is not None else [],
+        source_plane_lines=sp_lines,
+        source_plane_line_colors=sp_colors,
+    )
+    plotter.point_solver(fit=fit, image_plane_lines=ip_lines if ip_lines is not None else [])
+    keep_dataset_and_fit(
+        point_path, dataset_png="dataset_point.png", extra_pngs=("point_solver.png",)
     )
-    keep_dataset_and_fit(point_path, dataset_png="dataset_point.png")
     print(f"visualize [{dataset.name}]: {time.perf_counter() - t0:.1f}s")
 
 n_png = len(list(image_path.rglob("*.png")))