Affected Repositories

• PyAutoNerves (library: board)
• PyAutoGalaxy (library: config promotion)
• autofit_workspace, autogalaxy_workspace, autolens_workspace, autocti_workspace (workspace deletes)

Branch Survey

Suggested branch: feature/workspace-config-cleanup, worktree ~/Code/PyAutoLabs-wt/workspace-config-cleanup/.

Parallel claim: PyAutoNerves is also claimed by eyes-organ-order (PyAutoMind#439). Its Nerves diff is only AGENTS.md, so it does not overlap with scripts/board.py. The human approved a parallel worktree on 2026-09-27.

Implementation Steps

1. PyAutoNerves scripts/board.py:
   • In build_snapshot/diff_against_stack, handle file↔directory equivalence: a library X.yaml section <f> corresponds to the workspace file X/<f>.yaml, and the reverse. The semantics mirror autonerves RecursiveConfig.
   • Add the NERVES_OWNED allow-list for general.yaml version.{workspace_version, minimum_library_version, workspace_version_check, python_version_check}. These are excluded from orphans and noted "owned by autonerves".
   • Add tests: the equivalence clears a defined key, an owned key is not flagged, and an unknown key is still flagged.
2. PyAutoGalaxy:
   • autogalaxy/config/notation.yaml: add label.ra/rs and label_format.ra/rs, using the autolens_workspace values.
   • autogalaxy/config/priors/mesh/delaunay.yaml: add areas_factor as a Constant 0.5, which matches the Delaunay(areas_factor=0.5) default.
   • Add unit tests: a Delaunay model has no missing prior, and the dPIE ra label is not the first-letter fallback.
3. Ship the library half (/ship_library).
4. Workspace edits:
   • Delete the DELETE keys, the top-level subplot_shape_to_figsize_factor (it equals the library default of mat_plot.figure.*) and the PROMOTE keys in autolens_workspace.
   • Fix fit_imaging: {}.
   • Leave the CTI and version keys alone.
   • Before each delete, re-grep the workspace's scripts and notebooks. Mirror a change into notebooks or markdown only where they carry config snippets. Remove any YAML comment that describes a deleted key.
5. Ship the workspaces (/ship_workspace), then /prm, then dispatch nerves_board.yml and re-check the published board.

Per-key verdicts (79)

Totals: DELETE 51, KEEP-WS 7, PROMOTE 5, FIX 16 (13 CTI board false positives, 2 × subplot_shape_to_figsize_factor, 1 fit_imaging typo).

Out of scope: library-side unused keys (human-review-gated, NOT removed here)

• PyAutoLens:
  • config/non_linear.yaml: the dead 57-key nest.DynestyStatic / nest.DynestyDynamic block.
  • general.yaml: output.fit_dill.
• PyAutoFit:
  • general.yaml: output.log_level, output.log_to_file, output.log_file, output.search_internal and profiling.repeats. These may be read dynamically.
  • non_linear/GridSearch.yaml: parallel, parallel.number_of_cores and parallel.step_size.
• PyAutoArray general.yaml: numba.use_numba.
• PyAutoGalaxy general.yaml: adapt.adapt_noise_limit.
• PyAutoCTI general.yaml: fits.flip_for_ds9, hpc.iterations_per_update, model.ignore_prior_limits, output.log_file, output.log_level and output.log_to_file.

Remove any of these only on explicit human confirmation, as a separate library prompt.

Key Files

• PyAutoNerves/scripts/board.py: board snapshot, stack diff, orphans
• PyAutoNerves/autonerves/directory_config.py, autonerves/conf.py: the merge semantics to mirror
• PyAutoGalaxy/autogalaxy/config/notation.yaml, autogalaxy/config/priors/mesh/delaunay.yaml
• The workspace config/ trees listed in the table

Workspace config cleanup: remove orphan config keys the Nerves board flags

Type: maintenance
Target: workspaces
Repos:

• autofit_workspace
• autogalaxy_workspace
• autolens_workspace
• autocti_workspace
  Difficulty: small
  Autonomy: supervised
  Priority: normal
  Status: formalised
  Consequence: glance
  Witness: after the change, a local Nerves board render (python3 scripts/board.py --json in PyAutoNerves) reports 0 workspace files with orphan keys (down from 14 files / 79 keys), except any key this task keeps with a recorded reason; each touched workspace's start_here/smoke scripts still load their config with no config-key error.
  Review-minutes: 3
  Unattended: ready
  Filed: 2026-09-27
  Epic: organ-cockpit
  Lane: local-dev

Why

The Nerves board (PyAutoNerves#175, shipped 2026-09-27 in the cockpit-followups bundle) now lists
workspace config keys that no library in the stack defines. On the 2026-09-27 render
(https://pyautolabs.github.io/PyAutoNerves/) the board is YELLOW with 14 workspace config files
carrying 79 orphan keys. They are dead settings that mislead users who edit them expecting an effect.

The orphan keys (workspace side — in scope)

• autofit_workspace/config/general.yaml (L14, 6): hpc.live_visual_update, hpc.quick_update_background, test.check_preloads, test.preloads_check_threshold, version.minimum_library_version, version.workspace_version_check
• autofit_workspace/config/visualize/plots_search.yaml (L6, 1): mle.corner_cornerpy
• autogalaxy_workspace/config/general.yaml (L26, 5): hpc.live_visual_update, hpc.quick_update_background, test.disable_positions_lh_inversion_check, version.minimum_library_version, version.workspace_version_check
• autogalaxy_workspace/config/notation.yaml (L24, 2): label.label.contribution_factor, label_format.format.contribution_factor
• autogalaxy_workspace/config/visualize/general.yaml (L14, 14): subplot_shape.1, subplot_shape.100, subplot_shape.12, subplot_shape.16, subplot_shape.2, subplot_shape.20, subplot_shape.36, subplot_shape.4, subplot_shape.49, subplot_shape.6, subplot_shape.64, subplot_shape.81, subplot_shape.9, subplot_shape_to_figsize_factor
• autogalaxy_workspace/config/visualize/plots.yaml (L30, 1): fit_imaging {}
• autolens_workspace/config/general.yaml (L25, 5): hpc.live_visual_update, hpc.quick_update_background, version.minimum_library_version, version.python_version_check, version.workspace_version_check
• autolens_workspace/config/notation.yaml (L24, 10): label.label.contribution_factor, label.label.ra, label.label.rs, label.label.sigma_scale, label.superscript.hyperbackgroundnoise, label.superscript.hypergalaxy, label.superscript.hyperimagesky, label_format.format.contribution_factor, label_format.format.ra, label_format.format.rs
• autolens_workspace/config/priors/mesh/delaunay.yaml (L2, 1): delaunay.areas_factor
• autolens_workspace/config/visualize/general.yaml (L14, 14): subplot_shape.1, subplot_shape.100, subplot_shape.12, subplot_shape.16, subplot_shape.2, subplot_shape.20, subplot_shape.36, subplot_shape.4, subplot_shape.49, subplot_shape.6, subplot_shape.64, subplot_shape.81, subplot_shape.9, subplot_shape_to_figsize_factor
• autocti_workspace/config/general.yaml (L2, 6): analysis.n_cores, analysis.preload_attempts, profiling.should_profile, test.check_preloads, test.disable_positions_lh_inversion_check, test.preloads_check_threshold
• autocti_workspace/config/visualize/general.yaml (L5, 2): general.subplot_ascending_fpr, general.symmetric_cmap_value
• autocti_workspace/config/visualize/plots.yaml (L1, 11): combined_only, dataset.data, dataset.data_binned, dataset.data_logy, dataset.fpr_non_uniformity, dataset.subplot_dataset_regions, fit.data_logy, fit.fits_fit, fit.residual_map_logy, fit.subplot_fit_regions, subplot_format
• autocti_workspace/config/visualize/plots_search.yaml (L6, 1): mle.corner_cornerpy

Before deleting each key, grep the workspace itself (scripts, notebooks, start_here.py, any
workspace-local helper) for a reader: the board only checks library code, so a key read by the
workspace's own code (possibly the version.* keys via the version handshake in autonerves, or
subplot_shape.* read by an integer lookup) is not an orphan. Keep any such key and record why.
Mirror the removals into the matching notebooks/markdown only if they carry config snippets.

The typo

autogalaxy_workspace/config/visualize/plots.yaml:30 has fit_imaging {}: — a malformed key (the
{} looks like a leftover flow-mapping), so the fit_imaging section it was meant to open is not
read. Fix it to the intended fit_imaging: section (compare with autolens_workspace's plots.yaml),
not just delete it.

Separate, human-review-gated: library-side unused keys (NOT in scope to remove without review)

The same scan flags library config keys no library code reads. These need a human to confirm
before removal, because a key may be read dynamically (string-built lookups the scan cannot see):

• PyAutoLens config/non_linear.yaml: a dead 57-key nest.DynestyStatic / nest.DynestyDynamic
  block (Dynesty searches live in PyAutoFit's config; PyAutoLens's copy is not read).
• PyAutoLens general.yaml: output.fit_dill.
• PyAutoFit general.yaml: output.log_level, output.log_to_file, output.log_file,
  output.search_internal, profiling.repeats (may be read dynamically — needs review);
  non_linear/GridSearch.yaml: parallel, parallel.number_of_cores, parallel.step_size.
• PyAutoArray general.yaml: numba.use_numba; PyAutoGalaxy general.yaml: adapt.adapt_noise_limit.
• PyAutoCTI general.yaml: fits.flip_for_ds9, hpc.iterations_per_update, model.ignore_prior_limits,
  output.log_file, output.log_level, output.log_to_file.

Present this list to the human; remove only what they confirm, as a library change shipped before
the workspace edits (library-first), or re-file it as its own library prompt.

Related prompts

• draft/maintenance/workspaces/config_key_mirror_drift.md — the opposite direction (library keys
  missing from workspace configs); do both in one pass if picked up together, and do not re-add a
  key this task removes.
• draft/maintenance/workspaces/sync_remaining_workspace_config_priors_copies.md.