Summary

Nextflow plugin that writes a FAIRSCAPE EVI RO-Crate for each pipeline run — the EVI ontology provenance model emitted natively, rather than converted from another provenance format after the fact.

The crate conforms to RO-Crate 1.2, the FAIRSCAPE profile 0.1 and EVI, with PROV-O typing over schema.org.

Each successful task becomes an EVI:Computation under a run-level Computation. Files become EVI:Dataset entities with generated/generatedBy edges; the script, the engine and each process become EVI:Software. Full mapping: docs/FAIRSCAPE.md.

Plugin version	Minimum Nextflow
0.1.x	25.10

Get Started

Requires Nextflow 25.10 or later. Enable the plugin in nextflow.config and Nextflow fetches it from the plugin registry on the first run:

plugins { id 'nf-fairscape@0.2.0' }

Until the first registry release lands, build and install it locally instead — same result, one extra step (CONTRIBUTING.md):

make install
nextflow run <pipeline> -plugins nf-fairscape@0.2.0

A configured run looks like this:

plugins { id 'nf-fairscape@0.2.0' }

outputDir = params.outdir

fairscape {
    file      = "${params.outdir}/ro-crate-metadata.json"
    overwrite = true
    author    = 'Jane Doe'
    keywords  = ['genomics', 'my-project']
    license   = 'https://spdx.org/licenses/MIT'
}

No pipeline changes needed. On success you get:

results/
  ro-crate-metadata.json     # the crate
  provenance-graph.json      # evidence graph rooted at the crate
  provenance-graph.html      # interactive, self-contained viewer
  ro-crate-datasheet.html    # datasheet
  ai_ready_score.json        # AI-Readiness rubric behind it
  ro-crate-linkml.yaml       # crate root as a D4D document
  workflow/                  # the script and configs that produced all of it

Everything has a fallback (manifest, then a generated value), so the crate is valid with no fairscape block at all.

One thing that is on your pipeline rather than the plugin: if the crate is going to be archived or shared, publish with mode: 'copy'. Nextflow's default publish mode is symlink, and a symlinked published file gets a crate-relative contentUrl like any other — the path is right, but the bytes are still in work/, so zipping the crate ships dangling links. nf-core pipelines already set publish_dir_mode = 'copy'. Files the crate describes but does not contain (work-directory intermediates) carry localPath instead of a contentUrl, so the graph keeps them without claiming they are retrievable.

Examples

Every example is a self-contained pipeline plus the config that switches the plugin on. Install the plugin, then run one:

make install
cd examples/letters-chain && nextflow run . -plugins nf-fairscape@0.2.0

That writes results/ro-crate-metadata.json and the derived artifacts next to it — open results/provenance-graph.html to see the run's evidence graph.

Runnable demos: examples/reverse-list (minimal), examples/letters-chain (multi-step chain). New to Nextflow plugins? docs/WALKTHROUGH.md.

Against real pipelines: examples/nf-core runs five released nf-core pipelines unmodified and checks the crates they produce — the regression set the plugin is hardened against. Because they are unmodified, none of them can carry an ext fairscape: block; examples/fastquorum-like and examples/bamtofastq-like are the other half of that — real bioinformatics (fgbio duplex UMI consensus calling, and samtools BAM→FASTQ conversion) after nf-core/fastquorum and nf-core/bamtofastq, flattened into one annotated main.nf each, on their own test data. examples/cycle-repro and examples/edge-cases are the two smallest failures those runs turned up, reduced to something that finishes in seconds.

Per run — fairscape.metadata is merged onto the root crate entity. Any field the FAIRSCAPE profile declares works, including the Croissant RAI rai:* properties:

fairscape {
    author       = 'Jane Roe'
    organization = 'Example Institute'          // -> publisher
    license      = 'https://spdx.org/licenses/CC-BY-4.0'

    metadata = [
        identifier           : 'https://doi.org/10.1234/example',
        principalInvestigator: 'Jane Roe',
        funder               : 'NIH Bridge2AI (OT2OD032742)',
        associatedPublication: 'https://doi.org/10.1234/example-paper',
        conditionsOfAccess   : 'Non-commercial research use only.',
        'rai:dataLimitations': 'Not validated for clinical use.'
    ]
}

Per process — ext fairscape describes the tool a process actually runs:

process REVERSE {
    ext fairscape: [
        softwareName   : 'tac',
        softwareVersion: '8.32',
        softwareUrl    : 'https://www.gnu.org/software/coreutils/tac',
        softwareAuthor : 'Jay Lepreau, David MacKenzie (GNU coreutils)'
    ]
    ...
}

The same map works from config (process { withName: 'REVERSE' { ext.fairscape = [...] } }), which is how you annotate a pipeline you don't own.

License

Apache-2.0 — full text in LICENSE. Parts of the plugin framework derive from nf-prov, also Apache-2.0; NOTICE records what and from where.