Add nf-core/eager fixture + compare-and-contrast scenario against iwc#1234

[jmchilton]

Adds nf-core/eager to the Nextflow fixture corpus and a NEXTFLOW → GALAXY pipeline scenario that holds the journey against a human-authored answer to the same conversion.

The fixture

Pinned to the DSL2 rewrite on dev (f77d2464), deliberately not to a release. eager's latest tag (2.5.3) is DSL1 (nextflow.enable.dsl=1, monolithic main.nf), which [[summarize-nextflow]] short-circuits by design — a tag pin would die at phase 1 and the scenario would test nothing. tag is optional in fixtures.yaml and several entries are already HEAD-pinned. Bump the SHA rather than adding a tag until 3.0 ships; the entry's notes say so.

Tier large: ~110 .nf files, ~45 nf-core modules, ~23 local subworkflows. Materializes and passes make fixtures-verify.

The scenario

galaxyproject/iwc#1234 is a human-authored eager-style aDNA workflow targeting the same source. It's a reference case, not a matching test — eval.md keeps the gates, the comparison is advisory. The value is in where and why the two diverge.

Why it earns its keep: eager is a toolkit with ~15 independent run_*/skip_* branch groups, and expressing that optionality in Galaxy is the hard part of this conversion. The PR is a worked human example of paying that cost — 52 steps, 12 carrying a when, with 13 of 41 tool steps (~⅓) being pure param plumbing (7 map_param_value to build when booleans, 6 pick_value to rejoin branches). That ratio is the calibration this case exists to provide.

The expect: aims at that level of option coverage without prescribing the route: some meaningful part of the source's branch surface should reach the user as run-time choice, but which controls, how many, and by what mechanism are the journey's call. The reference's four controls are framed as calibration, not a quota — a different three, or the same coverage via optional inputs instead of when gating, passes; padding with toggles eager's schema doesn't motivate fails on intent-fidelity.

This also closes a loophole in the existing framing: "reach a stated scope decision" was satisfiable by declaring "scope: one linear path," passing the gate while producing a workflow no analyst would use.

Caveats recorded in the case

• Version boundary. The PR's README cites nf-co.re/eager/2.5.3/ — the human worked from eager 2.x while the fixture must pin 3.x. Tool vocabulary overlaps almost entirely, so the contrast holds, but it is not a controlled A/B and some per-step differences will be version artifacts.
• Not in the corpus. The PR is unmerged, so paleogenomics/adna-analysis doesn't exist at the pinned IWC SHA; compare-against-iwc-exemplar can't reach it and will settle on another exemplar. Supply the PR by URL per the URL-not-mirror principle — the .ga is not mirrored here.
• Human ≠ correct. The PR is open and unreviewed. It registers both Sex.DetERRmine branches (and both ANGSD paths) as workflow outputs, so any run yields necessarily-empty outputs; a journey that avoids that isn't wrong for differing.

Incidental finding

The case surfaces a real gap in the pipeline spine, noted but not fixed here: there is no find-test-data phase, so implement-galaxy-workflow-test takes test-data-refs with nothing upstream producing it — the validator already warns (phases[8]). aDNA reference data is heavy and non-obvious, so this is a likely place for that gap to bite.

Testing

npm run validate on the main base: 0 errors, 194 warnings — identical to the pre-change baseline. make fixtures-verify passes with the new pin. Scenario content is prose with no frontmatter; the fixture entry is YAML — no code touched.

🤖 Generated with Claude Code