Fairshift Lab v1.3.0 — Robustness Lab

Robustness Lab: preregistered synthetic specification-stress study with six controlled stressors, a second inspectable model family, a frozen 60-cell registry, and an accessible interactive lab. Full details in CHANGELOG.md.

Fairshift Lab v1.2.0 — External Evidence

Fairshift Lab v1.2.0 adds governed historical observational evidence without blurring it into the synthetic laboratory.

Highlights:

• pre-result dataset admission gate with documented acceptance, deferral, and rejection decisions
• checksum-pinned UCI Adult acquisition with no raw-data redistribution
• 48-cell sensitivity registry across five split seeds, two cohorts, two missingness rules, three error-cost declarations, and four policy families
• accessible External Evidence interface with limitations and uncertainty before rankings
• provenance manifest, dataset card, preregistered protocol, bounded report, and updated continuation brief
• 100% Python statement and branch coverage; web, accessibility, reproducibility, dependency, and CodeQL gates pass

Scientific boundary: UCI Adult is a 1994 Census-derived historical reference table. Its label is not merit, qualification, need, or ground truth; its binary provider sex field is not identity truth. This release is not external peer review, current-population validation, a policy recommendation, legal advice, fairness certification, or deployment approval.

Fairshift Lab v1.1.0 — Policy Studio

Fairshift Policy Studio makes decision assumptions explicit.

Highlights:

• 20-seed, 72-cell benchmark across three distribution shifts, three error-cost declarations, and eight policy families
• Accessible fairness–utility Pareto explorer with uncertainty ranges and a complete data table
• Deterministic scenario JSON export with provenance and limitations
• Preregistered protocol and report plus a frozen, byte-reproducible result registry
• 100% Python statement and branch coverage; all web, accessibility, dependency, and code-scanning checks pass

Scientific boundary: synthetic, descriptive, internally verified evidence. This release is not externally peer reviewed, a policy recommendation, legal advice, fairness certification, or deployment approval.

Fairshift Lab v1.0.1

Accessibility and stability patch for the verified v1 research platform. All normal-text palette combinations now meet the WCAG 2.2 AA 4.5:1 contrast threshold, the threshold-chart SVG title hydrates as one stable text node, and automated release tests cover both conditions. The scientific registry and bounded v1 report are unchanged.

Fairshift Lab v1.0.0

Stable research platform

Fairshift Lab v1.0.0 turns the exploratory laboratory into a reproducible public research release.

Research evidence

• Frozen registry of 300 complete experiments across three interventions, five magnitudes, and 20 independent seeds
• Internally verified research report with explicit hypothesis dispositions, negative control, validity threats, and interpretation boundaries
• Byte-stable registry regeneration across supported platforms

Accessibility and resilience

• WCAG 2.2 AA-oriented keyboard navigation, visible focus, reflow, reduced motion, high-contrast, and forced-color support
• Equivalent descriptions and inspectable data tables for every scientific chart
• Accessible error recovery and missing-page routes
• Public accessibility statement and dedicated automated accessibility contract

Quality gates

• 45 deterministic Python tests with 100% statement and branch coverage
• Strict typing, linting, package builds, web rendering tests, production dependency audit, and CodeQL
• Dedicated CI verification that regenerates the complete research registry

This release uses synthetic data. It is not external peer review, legal certification, or evidence that a real-world decision system is fair.

Fairshift Lab v0.3.1

Fairshift Lab v0.3.1 is a focused rendering-stability patch for the interactive calibration and threshold-sensitivity release.

Fixed

• Deterministically round generated chart coordinates before server rendering.
• Prevent React hydration warnings caused by one-unit floating-point serialization differences between server and browser.
• Add a regression assertion and rerun the complete Python, web, security, and CodeQL gates.

All v0.3.0 scientific features and interpretation boundaries remain unchanged.

Fairshift Lab v0.3.0

Fairshift Lab v0.3.0 makes probability calibration and decision-threshold dependence visible under controlled distribution shift.

Highlights

• Independent training, source-calibration, source-evaluation, and target-evaluation populations
• Source-only temperature scaling selected by held-out negative log likelihood
• Raw and calibrated Brier score, expected calibration error, and reliability bins
• Formal 19-point source and target threshold sensitivity analysis
• Interactive reliability diagram and linked fairness/performance decision curves
• 39 deterministic Python tests with 100% statement and branch coverage
• Patched React, Vite, Vinext, and Cloudflare build chain with a production dependency audit

Calibration can fail after distribution shift, ECE depends on binning, and threshold sweeps do not choose a real-world decision policy. This synthetic laboratory is research and educational infrastructure—not a fairness certificate or compliance assessment.

Fairshift Lab v0.2.0

Fairshift Lab v0.2.0 turns distribution-shift auditing into an uncertainty-aware, interactive research laboratory.

Highlights

• Group-stratified percentile-bootstrap intervals
• Interactive covariate, concept, and prevalence-shift experiments
• Accessible source-to-target visual comparisons
• Explicit interpretation limits and a primary-source research trail
• 29 deterministic Python tests with 100% statement and branch coverage
• Production web build tests and CodeQL for Python and JavaScript/TypeScript

The synthetic laboratory is educational and methodological infrastructure, not a fairness certificate or evidence about a real population.

Fairshift Lab v0.1.0

Reproducible baseline

This initial research release establishes a transparent laboratory for evaluating model performance and group-fairness measurements under controlled covariate, concept, and protected-group prevalence shifts.

Included

• documented structural synthetic generator
• inspectable logistic-regression baseline
• accuracy, AUROC, demographic parity, equal opportunity, and equalized odds metrics
• deterministic JSON experiment CLI
• research protocol, methodology, Data Card, Model Card, ADR, and citation metadata
• 24 tests with 100% statement and branch coverage
• linting, strict typing, package build, CI, CodeQL, and Dependabot

Responsible-use boundary

This release is for research scaffolding and education. It is not suitable for decisions about people, legal-compliance certification, or claims about real demographic groups.