The proof ships with the claim. Every release is downloadable as an audit package: the corpus snapshot, the weights, the per-work breakdowns, and the one command that reproduces the ranking. If you cannot rebuild the rank from the package, the release is defective.
This release
version pilot-v0.1 · corpus_hash ecaac0049c25baa9fac29029 · method_version 0.1-pilot
Metrics: 174 (citation_count: 88, readership_persistence: 86); declared gaps: 76.
Downloads
The audit bundle below contains everything (code, weights, pinned data, all rankings, and every per-work breakdown). The individual files are listed for convenience.
Audit package
- audit-bundle.zip, the self-contained offline package: pipeline code, weights, pinned data, release outputs, and a one-command reproduce script. Rebuild this release with no repo and no network.
- release.json, the governance record
- coverage.json, declared gaps
Rankings (Top-50 per scenario)
The corpus, as open data
- books.json · books.csv, 573 books
- papers.json · papers.csv, 162 papers
- persons.json, 184 voices
- orgs.json, 133 organizations
- platforms.json, 90 platforms
Per-work breakdowns (one file per scored work) are inside the audit bundle.
Reproduce
make install && make assemble && make release && make verify-release
The last command rebuilds this release from the pinned inputs and asserts the corpus_hash and rankings are bit-identical. A mismatch means the release is defective, and we want the challenge.