Skip to content

Phase 4b: the eight feature ag_* skills - #6

Merged
Jammy2211 merged 1 commit into
mainfrom
feature/phase-4b-feature-skills
Aug 1, 2026
Merged

Phase 4b: the eight feature ag_* skills#6
Jammy2211 merged 1 commit into
mainfrom
feature/phase-4b-feature-skills

Conversation

@Jammy2211

Copy link
Copy Markdown
Contributor

Phase 4b of PyAutoLabs/PyAutoBrain#188 — the feature tier: basis profiles (linear/MGE/shapelets), pixelization, light-model extras (worked on the bundled real cutout), ellipse fitting, multi-dataset factor graphs, interferometer, multi-galaxy + cluster light, search chaining. 24 skills now live (17 galaxy-modelling).

Every skill execution-validated (test-mode fits completed; the multi-dataset recipe ran verbatim on the bundled 4-band JWST dataset; the cluster tier proven to cost exactly 1 free parameter). Every symbol verified against the released wheel.

Stale-claim fix on merged pages: result.model does not narrow priors on current stacks — narrowing lives on result.model_centred / model_centred_absolute/relative/max_lh_bounded. Three wiki pages corrected + re-stamped; ag_chain_searches documents the wheel truth and the workspace chaining.py staleness.

Upstream findings (report-only, logged in PENDING "Still open"): chaining.py's core prior-passing claim is stale (lines 209-289); ellipse scripts document the log-likelihood normalization wrong by 4× (−2.0×χ² vs actual −0.5×χ²); shapelets/modeling.py builds N=43 while claiming N=3 (unlinked ell_comps + a latent loop-variable bug); Ellipse.axis_ratio/.angle are methods not properties; interferometer transformer defaults differ between from_fits (NUFFT) and the simulator (DFT); "273 visibilities" is actually 190.

Validation (both stacks): --scope all 0 missing · citations 1044/0 · provenance 0 errors · --check-version clean · pytest 68 passed · link sweep 841/0 · sparse-cone sweep clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EVSbY4u1JLfqRMgXYtnN9v

ag_basis_profiles (linear+MGE+shapelets), ag_pixelization,
ag_light_model_extras (worked on the bundled real cutout),
ag_ellipse_fitting, ag_multi_dataset (AnalysisFactor/FactorGraphModel
only), ag_build_interferometer_model, ag_multi_galaxy_and_cluster
(light-only cluster tier), ag_chain_searches — each grounded on named
workspace scripts, every symbol wheel-verified, every skill's primary
path executed in test mode; the multi-dataset recipe ran verbatim on
the bundled 4-band dataset.

Stale-claim fix on three merged wiki pages: result.model does NOT narrow
priors on the released stack (narrowing lives on result.model_centred*);
pages corrected + re-stamped; ag_chain_searches carries the wheel truth.

Reconciled: skills/README (24 live), PENDING, skill_citation_map (24
rows), AGENTS/llms.txt/README counts, benchmark card MGE-routing rubric.

Validation both stacks: --scope all 43 files / 0 missing; citations
1044/0; provenance 0 errors; --check-version clean (wheel); pytest 68
passed; link sweep 841/0; sparse-cone sweep 0 violations.

Epic: PyAutoLabs/PyAutoBrain#188 (Phase 4b)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EVSbY4u1JLfqRMgXYtnN9v
@Jammy2211
Jammy2211 merged commit a226797 into main Aug 1, 2026
2 checks passed
@Jammy2211
Jammy2211 deleted the feature/phase-4b-feature-skills branch August 1, 2026 17:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant