{{REPO_NAME}}

{{PAPER_TITLE}} — replication study. Reference paper: {{PAPER_DOI}}

This is a self-contained replication of the headline claim of the reference paper. It produces a reproducible computational pipeline, a Zenodo-archived release with a citable DOI, and a FORRT-tagged nanopublication chain on the Science Live platform.

---

Quick start

git clone https://github.com/{{REPO_ORG}}/{{REPO_NAME}}.git
cd {{REPO_NAME}}
pixi install
pixi run snakemake --cores 1

(Pixi resolves pixi.toml against the per-platform pixi.lock, installs the env under .pixi/, and provides pixi run for any task without needing an activate step.)

Or with Docker:

docker run --rm ghcr.io/{{REPO_ORG}}/{{REPO_NAME}}:latest

The Jupyter Book version is at https://{{REPO_ORG}}.github.io/{{REPO_NAME}}/.

Built from a template

This repository was created from sciencelivehub/forrt-replication-template. The template ships an operating manual for AI assistants (CLAUDE.md, AGENTS.md), domain conventions (DOMAIN.md), and reference docs (docs/) so that an AI working only inside this repository can guide a researcher from "paper PDF + GitHub repo" to "published FORRT chain + Zenodo DOI" with no other context.

If you are reading this in a fresh fork, run /init-template inside Claude Code to substitute the placeholder tokens with your details. (For other AI tools, see docs/ai-portability.md.)

After /init-template, do these one-time setup steps to enable the full CI/CD path:

• Enable GitHub Pages at Settings → Pages → Source: GitHub Actions. Until enabled, the Jupyter Book build runs but the deploy step is skipped (CI stays green).
• The CI workflows ship with scaffold-detection guards — they run end-to-end only after you implement Phase 2 (the notebooks/*.py files). Until then they exit early with an informative ::notice:: and the badges stay green.

Repository structure

.
├── CLAUDE.md / AGENTS.md       # operating manual for AI assistants
├── DOMAIN.md                   # domain flavour (current: biodiversity + earth observation)
├── USER_PREFERENCES.md         # per-user style (edit on first clone)
├── README.md                   # this file
├── LICENSE                     # MIT
├── CITATION.cff                # how to cite
├── codemeta.json               # software metadata (CodeMeta-2.0)
├── ro-crate-metadata.json      # research object packaging (RO-Crate 1.2)
├── pixi.toml + pixi.lock       # pinned dependencies (single source of truth; lockfile is per-platform)
├── Dockerfile                  # container build
├── Snakefile                   # pipeline orchestration
├── myst.yml + index.md         # Jupyter Book scaffold
├── paper/                      # the source paper PDF
├── data/                       # downloaded artefacts (gitignored)
├── notebooks/                  # jupytext .py pipeline (01–04)
├── nanopubs/                   # FORRT chain drafts + published-URI registry
├── docs/                       # reference material
├── figures/                    # curated figures used in the Jupyter Book
├── .github/workflows/          # CI, Jupyter Book, Docker
└── .claude/                    # Claude Code agents, skills, sandbox config

What you get

This template bakes in conventions that took multiple replications to discover. By using it, you inherit:

• FAIR4RS conformance — see docs/fair4rs-checklist.md for the principle-by-principle mapping.
• Self-contained data downloads — the first notebook fetches everything; no manual data prep.
• pixi.toml + pixi.lock as single source of truth — local dev, Docker, and CI all install the same per-platform-pinned env.
• prefix-dev/setup-pixi-based CI — caches the env, runs the pipeline with pixi run, executes notebooks via a glob, fails fast on a stale lockfile.
• Jupyter Book deployment — auto-deploys to GitHub Pages with BASE_URL set correctly. (Don't put base_url in myst.yml — MyST silently ignores it.)
• Docker + GHCR + Zenodo image archival — release trigger pushes to GHCR and (optionally) archives to Zenodo for long-term preservation.
• RO-Crate packaging — the entire repo is a navigable Research Object via ro-crate-metadata.json (Process Run Crate + Workflow RO-Crate profiles).
• Six-step FORRT chain workspace — nanopubs/drafts/ has a field-by-field skeleton for each step. nanopubs/PUBLISHED.md is the URI registry.
• Layered AI guidance — CLAUDE.md (universal) + DOMAIN.md (swappable per field) + USER_PREFERENCES.md (per-user). See docs/ai-portability.md for non-Claude AI tools.
• Sandbox by default — .claude/settings.json denies file ops outside the repo, so a fresh AI session can't accidentally read ~/.ssh/ or write to /etc/.

The six FORRT chain steps

A complete FORRT chain has six steps published on platform.sciencelive4all.org:

Quote-with-comment  →  AIDA  →  FORRT Claim  →  Replication Study  →  Replication Outcome  →  CiTO Citation

(For question-rooted chains with no upstream paper, replace step 1 with PICO or PCC. See docs/chain-decision-tree.md.)

Drafts live in nanopubs/drafts/ field-by-field. Published URIs go into nanopubs/PUBLISHED.md.

Optional further layers:

• Research Software nanopub — for reusable upstream tools (not demo repos). See docs/forrt-form-fields.md § Research Software.
• Research Synthesis nanopub — when this chain is part of a multi-chain story. See docs/forrt-form-fields.md § Research Synthesis.

After publishing

When the chain is live and the FAIR4RS checklist is green, drafting an announcement post is the next step. See docs/announcement-template.md for the structural template (vision-piece-first; the worked replication is the payoff, not the lead).

For lower-level nanopub work — retraction, superseding, batch publishing — see docs/programmatic-nanopubs.md.

Citation

If you use this work, please cite both:

• This software: CITATION.cff → DOI {{ZENODO_DOI}}
• The original paper: {{PAPER_DOI}}

Acknowledgements

This repository was built from sciencelivehub/forrt-replication-template, part of the Science Live platform. The template is licensed MIT and contributions (especially new domain flavours under docs/domain-flavours/) are welcome.