Immutable run records for STAR-suite and Morphic recipe executions.
This repo records what actually ran: exact commits, rendered commands, parameters, input/output inventories, checksums, environment pins, logs, and handoff transfer status. It should not contain large data payloads.
Executable reproduction scripts may live in morphic-recipes. Provenance
records and dataset release notes should point to the exact recipe entrypoint,
commit, and workflow id used for reproduction.
runs/<project>/<run_id>/ One immutable production or handoff run record
dataset_releases/<project>/<release_date>/
Human-facing Globus/dataset release notes
schemas/ Machine-readable provenance schemas
templates/run/ Starter files for a new run record
docs/ Provenance policy and conventions
When reproducing a STAR-suite processing pipeline:
- Start from
runs/<project>/<run_id>/README.mdandrun.json. - Check the recorded reproduction entrypoint, usually in
morphic-recipes, when rerunning from a fresh root directory. - Check
dataset_releases/<project>/<release_date>/for what was actually uploaded to collaborators and any later release-specific notes. - Follow the recorded recipes repo commit and workflow id.
- Follow the recorded STAR-suite commit and binary checksum.
- Use the handoff packet only as a collaborator-facing convenience view.
Do not commit h5ad, h5, h5mu, BAM, FASTQ, Matrix Market payloads, or large logs. Reference them by path or URI with size, checksum, and retention policy.
Completed run records are append-only. If a mistake is found, add a
corrections entry to run.json and update the run README with the reason.
Do not rewrite history silently.