GOVERNATLAS / PRODUCT DOSSIER

Aequitas

Aequitas documents an open-source toolkit for bias auditing and fairness-aware machine learning.

Explore the product ↗Add to comparison ↗

01 / CONTEXT & SCOPE

Understand the proposition.
Define the boundary.

Where it fits

Inspect the label, score and group definitions before interpreting an audit or mitigation experiment.

The decision to make

Check the prediction and label semantics before interpreting a disparity plot. A categorical sensitive-feature column and a binary outcome encode a particular assessment question. Decide whether the task is auditing fixed predictions, applying a post-processing change or training a new pipeline; those actions require different evidence and responsibilities.

What this record does not establish

The toolkit description is not proof that a model is unbiased or that changing thresholds is suitable for a real decision. No sensitive-person eligibility determination or actual mitigation was performed here.

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

Prepare fictional binary routing predictions with a deliberate missing label and uneven cohort sizes. Inspect a selected audit metric, then consider a mitigation only on a separate held-out workflow. Preserve the original predictions and document which cases were excluded rather than presenting an automatically corrected result.

GovernAtlas editorial proposal. We have not performed this test.

Evidence to bring back

A prediction/label/group schema and audit denominator record.
An unchanged audit baseline and a separately reviewed mitigation experiment.

THE OPERATING DETAIL

Make the input table part of the review evidence.

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

Check the meaning of the score

Use a synthetic binary-classification table with documented labels and group categories. Record whether the score represents a thresholded decision or a probability at each stage. Compare the resulting confusion-matrix measures with a manually inspected small subset before interpreting a plot.

Preserve the experiment split

Keep training, validation and test roles separate when comparing methods or adjusting group thresholds. Retain the experiment settings, seeds and selected criterion. An alternative selected on the same cases used for final reporting should be identified as such rather than presented as unseen-case evidence.

Challenge the operating constraint

Review whether a proposed intervention changes the dataset, estimator or postprocessing rule, and who would be allowed to apply it. Document the assumptions behind group-specific behavior. A toolkit demonstration should not become an instruction to change real consequential decisions without the responsible process and review.

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 methods span preprocessing, in-processing and postprocessing, including group-specific threshold adjustments; no mitigation is established as appropriate for a real decision here.

Source, dates & preserved provenance

Aequitas — primary scope and current identity ↗

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

Artifact SHA-256: 96e0dff3e5c8a44767a6727d24ee90052380d67a5c67018ec98ca86087f71336

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

first party asserted

The README describes Aequitas Flow experiments with configurable methods, runs and dataset splitting.

Source, dates & preserved provenance

Aequitas — primary scope and current identity ↗

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

Artifact SHA-256: 96e0dff3e5c8a44767a6727d24ee90052380d67a5c67018ec98ca86087f71336

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

first party asserted

Aequitas documents auditing model predictions from a pandas DataFrame containing labels, model outputs and categorical sensitive attributes.

Source, dates & preserved provenance

Aequitas — primary scope and current identity ↗

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

Artifact SHA-256: 96e0dff3e5c8a44767a6727d24ee90052380d67a5c67018ec98ca86087f71336

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

first party asserted

Aequitas documents an open-source toolkit for bias auditing and fairness-aware machine learning. Scope: primary documentation only; no independent model test, fairness determination, causal finding or compliance approval.

Source, dates & preserved provenance

Aequitas — primary scope and current identity ↗

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

Artifact SHA-256: 96e0dff3e5c8a44767a6727d24ee90052380d67a5c67018ec98ca86087f71336

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

Source register

Aequitas — 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 prediction/label/group schema and audit denominator record for the selected configuration?

Can you provide an unchanged audit baseline and a separately reviewed mitigation experiment 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.