Changelog

All notable user-facing changes to GazeAudit are documented here.

The 0.1.0 entry below defines the first stable release candidate metadata. A repository state containing this entry is not, by itself, evidence that v0.1.0 has been tagged or that a GitHub Release, PyPI distribution, DOI, or archival deposit exists. Those are separate publication actions performed only after the release-candidate gate succeeds.

Unreleased

Documentation and website

Align public documentation, homepage installation guidance, and provenance surfaces with the verified v0.2.0 GitHub Release and PyPI publication.
Add a sensitivity-analysis learning path spanning conceptual rationale, protocol design, a worked sampling/missingness example, and a study-level workflow.
Improve plot and documentation discovery while preserving all frozen scientific validation outcomes and their original evidence boundaries.

0.2.0 — 2026-09-21

Scientific and software qualification

Reach 100% executable statement coverage across the GazeAudit Python package.
Qualify the complete test matrix on Python 3.10, 3.11, 3.12, and 3.13.
Expand deterministic fail-closed tests for source intake, provenance, archive verification, execution contracts, missingness, AOI uncertainty, freeze records, reporting, readiness, and publication infrastructure.
Preserve the frozen GazeBase, Korthals, and Pedrotti scientific outcomes without changing estimators, cohorts, thresholds, endpoints, protocols, or interpretation rules.
Normalize Korthals v2 companion paths for portable Windows/POSIX execution.

Release engineering

Generalize GitHub Release and PyPI workflows so new releases are version-driven rather than hard-coded to the historical 0.1.0 publication files.
Preserve exact-tag rebuilding, checksum verification, clean-wheel installation, GitHub Release round-trip verification, and PyPI Trusted Publishing.
Require one source distribution and one universal wheel for the qualified payload.

0.1.0 — 2026-09-14

Scientific MVP

Validate GazeAudit against known-truth tests and three canonical real-data cases.
Preserve the GazeBase multi-detector audit as incomplete after its frozen completeness gate failed rather than removing failing detectors post hoc.
Preserve the Korthals target-tracking AOI audit as robust_negative under the frozen measurement-error propagation protocol.
Preserve the Pedrotti/de Chambrier sampling/missingness audit as materially_fragile under the frozen recovery rule.
Bind protocols, source identities/source locks, controlled executions, fingerprints, checksums, and archived artifacts with archive-before-reveal provenance.

Reproducibility and publication support

Add an authoritative validation matrix linking all three canonical real-data cases.
Add the archived Pedrotti/de Chambrier sampling/missingness result record.
Add a publication-facing validation summary and machine-readable reproducibility index.
Add citation, reporting, interpretation-boundary, and reuse guidance.
Add repository citation metadata through CITATION.cff.
Add regression tests binding canonical scientific identities and indexed repository paths.

Packaging and release engineering

Mark the package as Alpha.
Qualify clean source and wheel distributions in CI with metadata/content validation.
Install the built wheel in a fresh environment and verify bundled scientific data.
Add a fail-closed stable metadata/tag/changelog release gate.
Add a release-candidate workflow that rebuilds and archives candidate distributions without automatically publishing them.
Require exact-head PR qualification followed by exact-main tests, live interoperability, distribution, and Pages certification before tagging.

Scientific MVP baseline — 2026-09-14

The scientific MVP was completed before stable-release preparation.

GazeBase multi-detector audit: canonical classification incomplete.
Korthals target-tracking AOI measurement-error audit: canonical classification robust_negative.
Pedrotti/de Chambrier sampling/missingness audit: canonical classification materially_fragile.

These scientific outcomes are post-outcome frozen. Later documentation, packaging, release, or publication work must not change the underlying protocols or reinterpret classifications by altering the predeclared decision rules.