Skip to content

vahtian v.0.1.0

Choose a tag to compare

@heidihelena heidihelena released this 27 Jun 08:00
· 332 commits to main since this release
cc35c68

vahtian v0.1.0

First release — the reproducible, provenance-first core, in both Python and R.

vahtian freezes a record set into a content-hashed, provenance-stamped,
date-locked corpus; verifies it (tamper-evident); and keeps a hash-chained
audit trail. Human-first. AI-second. Auditable.

Highlights

  • Cross-language parity, proven in CI. The Python and R packages produce a
    byte-identical content_hash for the same records — a golden-hash test
    asserts the same literal in both test suites, so a corpus frozen in one
    language verifies in the other.
  • Zero dependencies (Python), one (R). Stdlib-only Python; R needs only digest.
  • Local-first, no accounts, no telemetry. Apache-2.0.

What's in it

  • freeze(records, search_date) → deduped (PMID > DOI > title-hash),
    provenance-merged, date-locked, content-hashed corpus. Reproducible
    independent of source order.
  • verify(corpus) → recomputes the hash; any tampering fails it.
  • A hash-chained audit ledger (Ledger / ledger_*): retro-edits,
    reordering, or deletions break the chain.

Install

Python (PyPI):
pip install vahtian

R (from R-universe — distribution coming):

install.packages("vahtian", repos = "https://heidihelena.r-universe.dev")

Notes
Published to PyPI via GitHub Actions Trusted Publishing (OIDC) with PEP 740
attestationsno API tokens.
vahti (Finnish) = sentinel / guard. The watch over what gets cited.
Full context: https://vahtian.com/ · https://vahtian.com/agents/ (for LLMs & agents)