Validation data and examples (cjfp-data)#

cjfp-data is the public companion repository for CosmicJellyfish validation artifacts, posterior chains and HPC workflows. The source repository keeps runnable tools and small configuration examples; the data repository holds published outputs. A real metadata-only example is shown in Likelihood sampling with Nautilus.

Get the repository and selected data#

Keep the checkouts next to one another:

git clone https://github.com/santiagocasas/cosmicjellyfish.git
git clone https://dagshub.com/santiagocasas/cjfp-data.git
cd cjfp-data
python -m pip install -r requirements.txt

The repository is public. Downloading DVC-backed artifacts may require a DagsHub account and authentication; follow the repository’s access instructions if prompted. Git contains scripts and metadata; large arrays and chains need a separate DVC download. Choose a target rather than downloading every suite:

dvc pull validation/fisher.dvc
# Alternatively, fetch only the ns-response bundle:
dvc pull scripts/validation_results/ns_derivative_lhs.dvc

Location in cjfp-data	Contents and instructions
`chains/`	Selected completed Nautilus runs (`chains.dvc`).
`validation/fisher/`	CAMB-versus-CLASS Fisher comparisons, logs and dashboard.
`validation/syren_deaglan_lhs_opt*/`	SYREN comparisons; targeted-download guide.
`scripts/validation_results/ns_derivative_lhs/`	ns-response arrays, figures and provenance; sharing guide.
`jobs_punch/`	JURECA launchers, sampler YAMLs and campaign notes.

Run a CAMB-versus-CLASS Fisher campaign#

Install the source environment first, then run the data-side preflight from the source checkout:

cd ../cosmicjellyfish
uv sync --locked
bash ../cjfp-data/scripts/run-fisher-validation.sh --source "$PWD" --check

For a fresh 25-case campaign (adjust concurrency to your CPU allocation):

OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  bash ../cjfp-data/scripts/run-fisher-validation.sh --source "$PWD" \
  --run --all --jobs 2 --omp-threads 4

The launcher records the source revision, package versions and inputs, and writes to a new cjfp-data/validation/fisher/numpy2_<revision>_<timestamp>_<pid>/ directory. Keep the checkout unchanged until it finishes. The validation compares CAMB and CLASS; consult the per-case comparison reports and dashboard/index.html inside that campaign. For setup and troubleshooting see the workstation runbook.

Nautilus and cluster workflows#

From the source root, preview the maintained Euclid sampling campaign with:

bash scripts/likelihood/nautilus/run_euclid_mp_nautilus.sh --dry-run

Remove --dry-run to sample after reviewing the resource settings in Scripts reference. This campaign defaults to timestamped outputs under ../cjfp-data/chains/. CFP_DATA_REPO selects another data checkout for maintained source runners; pass --source explicitly to the data-side Fisher launcher when the source checkout is elsewhere. The generic YAML sampler still defaults to ./chains/chains_<name>/ unless output_dir is set.

For Slurm production runs, use the configurations and site-specific instructions in jobs_punch and its Nautilus findings. Review scratch paths, environment activation and CPU allocation for your site.

Citation and reproducibility#

When publishing results, link to cjfp-data and record both the source and data Git revisions plus the selected campaign directory. See the source repository’s CITATION.cff for software citation metadata. A Zenodo DOI is planned for the CosmicJellyfish GitHub release; a separate data DOI or mirror is deferred.