Skip to content

Workflow: isolated catalogue registries, shear-only catalogues, narrower rerun triggers - #450

Open
cailmdaley wants to merge 6 commits into
developfrom
fix/workflow-mock-catalogues
Open

cailmdaley wants to merge 6 commits into
developfrom
fix/workflow-mock-catalogues

Conversation

@cailmdaley

Copy link
Copy Markdown
Collaborator

Lets the cosmo_val workflow run on a catalogue registry other than the checkout's cat_config.yaml, such as the catalogue-painted GLASS mocks in #449. It also stops cosmetic catalogue-config edits from rerunning expensive two-point products.

  • catalog_config: in the run config selects an isolated catalogue registry, including that registry's own paths.blinds, and every producer reads the same file. Configured catalogue names now also work as version wildcards, so a mock entry doesn't need an SP_v… name.
  • Shear-only catalogues skip the object-wise PSF regression and its plot instead of failing. Mocks have no galaxy PSF columns.
  • Rerun triggers: ξ±, ρ/τ, pseudo-Cℓ and its covariance now take the catalogue's load-bearing metadata (measurement_catalogue) as params: paths, columns, calibration, mask, and for ρ/τ the PSF and star entries. Changing a column or path reruns the product. Changing a plot style or a precomputed survey summary does not.
    • Existing run dirs see one rerun after this lands, because the params are new.

Tests: workflow/tests 16 passed, using the same command as the CI DAG tests; test_cosmo_val_optional_psf.py 1 passed in the image.

These ran the 4-entry seed-1 known-answer suite (clean, mock-blinded, PSF-leak, additive-B). Results: EE recovered, B nulls PTE 0.23/0.94, exact blind shift, injected additive B recovered at S/N 4.5.

Claude Opus 5.5 on behalf of Cail.

🤖 Generated with Claude Code

…orkflows

Real-data impact: opt-in catalog_config overrides reach every producer; default catalogue selection is unchanged.
Real-data impact: catalogues with configured galaxy PSF columns are unchanged; missing-column catalogues keep the rho/tau diagnostics and label the unavailable regression.
Real-data impact: edited column mappings, response or redshift paths invalidate the affected catalogue instead of silently reusing its products. Also documents isolated registries.
Real-data impact: plotting styles and cov_th edits do not rerun catalogue measurements. The first invocation after the catalogue param is introduced reruns these producers in existing run directories, including run-int1028-style directories; this is a one-time cost.
cailmdaley added a commit that referenced this pull request Oct 30, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JK7o9Gb8hcmKCbSUirKU76

This branch has not been deployed

No deployments
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