curie-data

Versioned nuclear-data releases and data tooling for curie, a Python toolkit for the analysis of experimental nuclear data.

This repository's Releases are releases of data, not software: each tag is an immutable-by-convention snapshot of the nuclear databases that a given range of curie versions consumes. The curie package itself ships no data — it carries a registry (curie/data_registry.json in the curie repo) mapping every file here to its SHA256 and download URL, fetches files on first use, and verifies every download against the registry checksum.

Release layout

One data generation spans several release tags, because GitHub caps a release at 1000 assets and the sharded libraries carry ~420 files each:

Tag	Contents
v1	The 9 whole database files + SHA256SUMS + data_manifest.toml
v1-endf	ENDF/B-VII.1 per-target shards (endf_<TABLE>.db, one table each)
v1-tendl	TENDL-2015 exclusive-reaction shards
v1-tendl-n-rp	TENDL-2015 neutron residual-product shards
v1-tendl-p-rp	TENDL-2015 proton residual-product shards
v1-tendl-d-rp	TENDL-2015 deuteron residual-product shards

The whole files and the shards are two representations of the same libraries: the shard tooling pins both to the same source file in the registry, so they cannot silently diverge. curie's lazy path fetches individual shards (a reaction lookup costs a fraction of a MB); ci.download() fetches whole files for offline preparation.

Databases that are not sharded (decay, ziegler, IRDFF, iaea_monitors) are fetched whole: decay.db is traversed globally by decay-chain calculations, IRDFF stores per-reaction rather than per-target tables, and the others are small.

Versioning and compatibility

• Every released curie version pins exactly one data generation through its registry: curie 0.1.0 → v1. Old tags are never modified or deleted, so old curie versions keep working indefinitely.
• A new data generation (e.g. the ENDF/B-VIII.1 / TENDL-2023 rebuild) is published as a new set of tags (v2, v2-endf, ...) alongside the old ones, and ships to users via a new curie release carrying the updated registry.
• A library whose shard count approaches the per-release asset cap splits by Z-range into multiple tags (v2-tendl-1, v2-tendl-2, ...); the registry's per-group base_url makes this transparent to curie.
• Historical note: curie 0.0.35–0.0.37 fetch whole files from the frozen data-v1 release on the curie repository itself; that release stays in place for them.

Tooling

scripts/shard_library.py splits a per-target library (endf, tendl, tendl_n_rp, tendl_p_rp, tendl_d_rp) into one small sqlite file per table and records per-shard SHA256s in the curie registry (default: the sibling ../curie checkout). Run it against the exact database file being published — it also pins the whole-file hash so the two representations stay coupled.

data_manifest.toml records the provenance (source library, size, SHA256) of the current data generation.

The build pipeline that regenerates the databases themselves from the upstream evaluations (ENDF/TENDL/ENSDF sources) will live here as it is developed.