GOVERNATLAS / PRODUCT DOSSIER

DiCE

DiCE documents methods for generating diverse counterfactual explanations of machine-learning predictions.

Explore the product ↗Add to comparison ↗

01 / CONTEXT & SCOPE

Understand the proposition.
Define the boundary.

Where it fits

Evaluate whether a generated alternative respects domain constraints and can be interpreted as more than a model-input change.

The decision to make

Specify what may change before asking for a different prediction. A counterfactual can be close to an input mathematically while proposing an impossible historical change or an inconsistent feature combination. Keep model-output alternatives separate from feasible actions and from causal claims about the wider business process.

What this record does not establish

Diverse counterfactual examples are not real-world recourse, causal evidence or advice for a real eligibility decision. Feasibility constraints, preprocessing and backend support require case-specific review.

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 a fictional queue-routing case with one editable field and one immutable historical field. Generate proposed alternatives only in an authorized synthetic trial, reject an infeasible one and retain the constraints and rejection rationale. Verify the model output separately from any downstream routing rule.

GovernAtlas editorial proposal. We have not performed this test.

Evidence to bring back

A model/backend/preprocessing and permitted-feature-change manifest.
A proposed and a rejected alternative with feasibility rationale and output evidence.

THE OPERATING DETAIL

A counterfactual is a candidate, not a promise of recourse.

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

Constrain the hypothetical change

Use a fictional routing model and identify immutable fields, permitted ranges and dependent features. Preserve the constraints passed to the generator. Inspect an alternative that changes the model output but violates a real operating dependency. A mathematically valid input can still describe an impossible situation.

Compare diversity and proximity

Generate a small set of alternatives under two documented settings while keeping the model and target fixed. Review which features changed and why the alternatives differ. A nearby point is not necessarily easier to act on, and diverse outputs are not necessarily representative of the feasible action space.

Keep the decision authority clear

Retain the selected generation method, model-access assumptions and any use of training data or metadata. Ask a reviewer to document what could actually be changed and who controls that change. An outcome-flipping input does not establish a causal intervention, a guaranteed future result or a right to a particular decision.

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

The project documents use with pretrained models, with or without the original training data; generated counterfactuals do not establish feasible real-world recourse.

Source, dates & preserved provenance

DiCE — primary scope and current identity ↗

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

Artifact SHA-256: 72cd9127c8aff282bec8578c6bba6dd79d27db487a0bfa40f68f741bd5b45372

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

first party asserted

Its model-agnostic generation options include random sampling, KD-tree search and genetic algorithms; gradient-based options require a differentiable model.

Source, dates & preserved provenance

DiCE — primary scope and current identity ↗

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

Artifact SHA-256: 72cd9127c8aff282bec8578c6bba6dd79d27db487a0bfa40f68f741bd5b45372

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

first party asserted

DiCE documents diverse counterfactual generation with tunable diversity and proximity, plus constraints on changeable features and permitted ranges.

Source, dates & preserved provenance

DiCE — primary scope and current identity ↗

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

Artifact SHA-256: 72cd9127c8aff282bec8578c6bba6dd79d27db487a0bfa40f68f741bd5b45372

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

first party asserted

DiCE documents methods for generating diverse counterfactual explanations of machine-learning predictions. Scope: primary documentation only; no independent model test, fairness determination, causal finding or compliance approval.

Source, dates & preserved provenance

DiCE — primary scope and current identity ↗

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

Artifact SHA-256: 72cd9127c8aff282bec8578c6bba6dd79d27db487a0bfa40f68f741bd5b45372

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

Source register

DiCE — 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 model/backend/preprocessing and permitted-feature-change manifest for the selected configuration?

Can you provide a proposed and a rejected alternative with feasibility rationale and output evidence 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.