Skip to content

v2.0.0 — LLM-assisted reproducibility and scientific-alignment update

Latest

Choose a tag to compare

@reblocke reblocke released this 27 Jul 23:09
· 1 commit to main since this release
88098c6

v2.0.0 — LLM-Assisted Reproducibility and Scientific-Alignment Update

This is the current citable release of the public repository.

What Changed

  • Applied human-adjudicated corrections to published case-definition
    boundaries, sensitivity-analysis exclusions, testing-strategy subgroups,
    selected upstream aggregate checks, and figure filenames.
  • Added a guarded restricted-data runner, adjacent input approval contract,
    provenance capture, complete artifact checks, and a value-suppressing
    comparison workflow.
  • Added a portable PNG-to-TIFF cohort-diagram workflow, locked public checks,
    structured metadata, and release-aware reader and machine documentation.
  • Reorganized the README around release selection, article identifiers, the
    data boundary, quick starts, and a paper-to-code map.

LLM Assistance and Scientific Authority

LLM tools assisted repository review, reproducibility hardening, testing, and
documentation. Scientific decisions were made and adjudicated by the repository
maintainer and study author.

Validation

  • Release commit: 88098c6e89f8084114d4bf72e787340b0422523b
  • GitHub Actions passed on both main and the v2.0.0 tag.
  • The public suite contains 100 tests, a public-surface audit, CFF validation,
    and an executed diagram/TIFF smoke test.
  • The exact scientific-alignment implementation
    9f8fecd2384edd3ae3fd4b0c8ecbc8536769bc19 was executed twice against the
    same approved restricted input. The candidates were artifact-for-artifact
    repeatable, and historical differences were recorded as correction impact.
  • The current scientific body remains separately test-locked.

Important Boundaries

  • This release is not equivalent to v1.0.0; it includes documented scientific
    corrections.
  • SA-002, SA-007, SA-008, and SA-009 remain unresolved.
  • Public checks do not reproduce the restricted Stata analysis or validate the
    article estimates.
  • The repository does not reproduce upstream TriNetX construction.
  • No restricted data, generated results, logs, workbooks, figures, or other
    release assets are attached.

See docs/SCIENTIFIC_ALIGNMENT.md, docs/VALIDATION.md, and
docs/REPRODUCIBILITY.md for the technical evidence and limitations.