Curation Workflow

New records should be staged through YAML before they are applied to production JSON.

The current curation workflow is a source-checkout workflow rather than an installed package command. From the repository root, install the contributor dependencies first:

python -m pip install -e ".[dev]"

  1. Copy one or more templates from curation/templates/.

  2. Fill in a staging file under curation/staging/.

  3. Generate a preview and review the report.

  4. Apply the staged records after human approval.

  5. Remove the staging YAML after the production update is committed.

Generate a preview:

python scripts/stage_records.py --staging curation/staging/example.yaml

Apply approved staged records:

python scripts/stage_records.py --staging curation/staging/example.yaml --apply

The staging templates intentionally expose the full available field set, including optional fields, so curators do not need to remember schema details from memory.

After an approved production data change, refresh the committed data baseline and inspect the diff:

python scripts/update_data_baseline.py
python scripts/check_curation.py

tests/baselines/production_data.json stores expected production inventory counts, known validation warnings, and curation-sensitive 2026 regression expectations. Regenerate it only after the data change has been reviewed and accepted. If a new curation batch changes a 2026 table, figure, or slide assertion, add that generated expectation to scripts/update_data_baseline.py instead of pinning a new literal directly in the regression script.

scripts/check_curation.py runs production-data validation plus the load, validation-baseline, and output regression tests that are sensitive to new secure molecules and detections.

References are maintained in Zotero and exported to astromol/data/references.bib. Do not hand-edit the BibTeX file; export the updated collection and commit the resulting file with the relevant data change.

Project-Computed Values

Some curated values may be calculated as part of this work when no suitable literature value is available. Keep the supporting notebook under docs/calculations/ and commit it with the data update. Do not commit local .ipynb_checkpoints/ directories.

Structured molecule reference fields resolve to BibTeX citekeys, so use those fields for the software, method, basis-set, or literature references that support the calculation. Record the local notebook path in the relevant note or history entry. If the calculation is later described in a citable publication, add that citekey through the normal Zotero export workflow.

Scientific data are human curated and human verified. LLM assistance may be used for code, staging support, or mechanical transformations, but production data are not accepted without curator review.