GOVERNATLAS / PRODUCT DOSSIER

AI Fairness 360

The AI Fairness 360 project documents an open-source toolkit with fairness metrics and bias-mitigation algorithms for datasets and models.

Explore the product ↗Add to comparison ↗

01 / CONTEXT & SCOPE

Understand the proposition.
Define the boundary.

Where it fits

Choose the relevant harm, metric and intervention stage before comparing mitigation results.

The decision to make

Choose a harm question and metric before selecting a mitigation algorithm. Dataset metrics, prediction metrics and modifications to training or outputs have different implications. Pin the labels, group definitions and model versions, and ask which relevant aspects of the operating situation cannot be represented by the selected quantitative comparison.

What this record does not establish

The research toolkit does not certify fairness, determine discrimination or establish legal suitability. Available algorithms may need separate dependencies; source availability is not managed-service support or proof of complete harm coverage.

1 current source · 4 published assertions
Source coverage is a research boundary, not a quality score.

02 / YOUR EVALUATION

Turn interest into a useful trial.

Open the evaluation playbook ↗

A test worth running

Use entirely fictional routing data with documented cohorts and known labels. Compute the planned comparison in a separate authorized trial, retain a sparse or missing-label cohort and examine one proposed mitigation on a held-out split. Record the performance trade-off and the reason for rejecting an unsuitable metric.

GovernAtlas editorial proposal. We have not performed this test.

Evidence to bring back

A harm/metric/population/label-definition and package-version manifest.
A held-out comparison preserving sparse groups, exclusions and mitigation trade-offs.

THE OPERATING DETAIL

Treat a mitigation as a hypothesis to examine.

GovernAtlas analysis: practical review considerations, not findings from a product test.

Define the comparison population

For a synthetic routing task, preserve the label, group definition, sample sizes and chosen metrics. Explain why those metrics address the concern. Inspect one small or missing group explicitly. A numerical comparison should retain its denominator and uncertainty instead of presenting an unexplained fairness status.

Track what the method changes

Choose one preprocessing or postprocessing method and record the difference between the original data, trained model and final decision rule. Use a fixed held-out split for comparison. Preserve both group outcomes and task performance, including any result that worsens after the intervention.

Keep the judgment outside the metric

Ask a reviewer to explain what the result cannot answer about the actual operating process. Record accepted trade-offs and required follow-up separately from the library output. Reproducing a documented algorithm is a technical achievement; it does not establish that a real system is fair, lawful or suitable for deployment.

03 / DOCUMENTED EVIDENCE

Read the claim. Inspect the basis.

Each statement retains its classification and preserved source. A vendor assertion is not an independent test result.

first party asserted

Its documented metrics include selection-rate and error-rate group measures, sample distortion measures and subgroup analysis; those metrics do not constitute a fairness determination.

Source, dates & preserved provenance

AI Fairness 360 — primary scope and current identity ↗

Authority: first party
Observed 2026-10-16 · Reviewed 2026-10-16 · current until 2026-11-15

Artifact SHA-256: 398cc10cf97c54654ce1fa1383b5315ac1041aa2d4019edad648f2b4a347e977

The hash identifies the preserved capture. The live source may have changed.

first party asserted

Listed mitigation techniques include reweighing, adversarial debiasing and equalized-odds postprocessing.

Source, dates & preserved provenance

AI Fairness 360 — primary scope and current identity ↗

Authority: first party
Observed 2026-10-16 · Reviewed 2026-10-16 · current until 2026-11-15

Artifact SHA-256: 398cc10cf97c54654ce1fa1383b5315ac1041aa2d4019edad648f2b4a347e977

The hash identifies the preserved capture. The live source may have changed.

first party asserted

The AIF360 README describes Python and R packages with dataset and model fairness metrics, metric explanations and bias-mitigation algorithms.

Source, dates & preserved provenance

AI Fairness 360 — primary scope and current identity ↗

Authority: first party
Observed 2026-10-16 · Reviewed 2026-10-16 · current until 2026-11-15

Artifact SHA-256: 398cc10cf97c54654ce1fa1383b5315ac1041aa2d4019edad648f2b4a347e977

The hash identifies the preserved capture. The live source may have changed.

first party asserted

The AI Fairness 360 project documents an open-source toolkit with fairness metrics and bias-mitigation algorithms for datasets and models. Scope: primary documentation only; no independent model test, fairness determination, causal finding or compliance approval.

Source, dates & preserved provenance

AI Fairness 360 — primary scope and current identity ↗

Authority: first party
Observed 2026-10-16 · Reviewed 2026-10-16 · current until 2026-11-15

Artifact SHA-256: 398cc10cf97c54654ce1fa1383b5315ac1041aa2d4019edad648f2b4a347e977

The hash identifies the preserved capture. The live source may have changed.

Source register

AI Fairness 360 — primary scope and current identity ↗first party · captured 2026-10-16

Brand asset provenance

Organization identity asset from its official website; inclusion does not imply endorsement.

Official website ↗ · Original asset ↗

04 / BEFORE YOU COMMIT

Better questions.
A clearer decision.

Resolve these with the provider and your own evaluation team.

Can you provide a harm/metric/population/label-definition and package-version manifest for the selected configuration?

Can you provide a held-out comparison preserving sparse groups, exclusions and mitigation trade-offs for the selected configuration?

Which exact version, entitlements, data paths and responsible roles apply?

Regulatory applicability requires a separate, current determination. Inclusion is not certification.

COMPLIANCE / CLAIMS, SCOPE & GAPS

Start with the requirement.
Inspect the support.

Prepare your vendor review ↗

These labels describe this dossier’s current evidence coverage. A published mention may be a framework mapping, a qualified assertion or a limitation; it is not a compliance finding.