Rubricon 0.5.1

Added

• Property-based tests (Hypothesis) for alpha invariances and the noise model.
• OpenSSF Scorecard workflow and badge; CI, release and licence badges; ORCID in CITATION.cff; .zenodo.json archive metadata.

Changed

• Dependabot raises dependency ranges only when necessary.
• No change to any statistic or result.

Rubricon 0.5.0

Added

• rubricon.gates.attenuation (gate v2): claim checks that propagate gold-label noise
  into a paired comparison instead of blocking on a reliability floor.
  • eta_from_disagreement, eta_majority, attenuation: the binary noise model.
    Label noise leaves two systems' disagreement rate unchanged and shrinks the expected
    accuracy gap by 1 - 2 * eta_gold.
  • check_direction (the only blocking check, a paired z-test at a pre-set z*),
    attenuation_report (true-scale gap and MDE, never blocks), and
    check_contested_consistency (the gap reverses between unanimous and split items).
• Tests validating the noise model against exact enumeration and simulation.
• CI workflow (tests on Python 3.12 and 3.14, plus a byte-for-byte reproduction job),
  Dependabot, issue/PR templates, CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, CITATION.cff,
  and a pinned requirements-dev.lock.

Unchanged, on purpose

• SignalGate and its alpha floor keep their behaviour, so existing reports reproduce.
  A known-truth simulation in the companion audit repository found the floor costs power
  without reducing false claims at a matched false-claim rate; see that repository's paper.