Add test-data-resolution to nextflow/cwl-to-galaxy with source-specific resolvers (#348)

[jmchilton]

Opened by Claude (AI assistant) on behalf of @jmchilton.

Fixes #348.

Problem

implement-galaxy-workflow-test requires test-data-refs, but the nextflow-to-galaxy and cwl-to-galaxy spines had no phase producing it — the phase was uncompletable by composition (the Pipeline validator warned on the unproduced consumed artifact).

Design

Rather than wire these spines straight to the generic find-test-data search mold, this restores the pattern paper-to-galaxy already uses (paper-to-test-data → find-test-data → user-supplied). find-test-data searches IWC/public sources — the wrong first move for nextflow/cwl, whose summaries already declare their own test fixtures:

• summary-nextflow.test_fixtures / nf_tests[] — each input a role + url/path + sha1 + filetype
• summary-cwl.tests[] — each case a job_path (naming inputs) + expected_outputs

So two new source-specific resolvers map those declared fixtures onto the Galaxy interface and emit test-data-refs, deferring to find-test-data (search) only for inputs they can't resolve:

• nextflow-to-test-data (axis: source-specific, source: nextflow)
• cwl-to-test-data (axis: source-specific, source: cwl)

test-data-resolution chains become:

nextflow-to-test-data → find-test-data → user-supplied
cwl-to-test-data      → find-test-data → user-supplied

This also dissolves a diamond-overlap: the resolver reads declared fixtures for data, the test-plan mold reads them for assertions — orthogonal, and exactly how paper already works.

Changes

• New molds nextflow-to-test-data, cwl-to-test-data (+ casts).
• Both test-data-resolution branch chains fronted with the source-specific resolver.
• Regenerated the two assembled harnesses + Index.md/Dashboard.md.
• Updated docs/HARNESS_PIPELINES.md and docs/MOLDS.md.

Verification

• npm run validate → 0 errors; the test-data-refs unproduced-artifact warning is gone for both spines.
• make check-generated → clean.
• assemble-pipeline --check for both spines → clean against committed cast state.

Notes

• New molds ship index.md + cast only (no eval.md/scenarios.md yet), matching the paper-to-test-data precedent; those come with first test-drives.
• update-interview-to-galaxy shows the same validator warning but is intentionally left out of scope — its harness_notes document test-data-refs coming from the edited workflow's baseline fixtures.
• Pre-existing doc staleness noted but not touched: docs/MOLDS.md's "34 candidate Molds total" is a frozen design-era count (actual dirs: 47), unrelated to this change.

🤖 Generated with Claude Code