GOVERNATLAS / PRODUCT DOSSIER

SHAP

The SHAP project documents a game-theoretic approach to explaining machine-learning model outputs through feature attributions.

Explore the product ↗Add to comparison ↗

01 / CONTEXT & SCOPE

Understand the proposition.
Define the boundary.

Where it fits

Review the explained output, reference population and selected explainer before interpreting feature contributions.

The decision to make

State what the attribution is relative to. The selected explainer, background data, output scale and feature dependencies affect the interpretation. A feature contribution is not a demonstration that intervening on the real-world feature would cause the outcome to change, and a global ranking does not replace a case-level explanation.

What this record does not establish

Feature-attribution documentation does not establish causal effects, a correct reference population, fairness or an appropriate explanation for every reader. Pin the implementation and dependency versions; no model was run for this profile.

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

Keep a synthetic routing model and target record fixed while comparing two documented background cohorts. Retain the prediction and both attributions, repeat approximate calculations where appropriate and ask a reviewer to explain a changed ranking without calling it a changed model.

GovernAtlas editorial proposal. We have not performed this test.

Evidence to bring back

An explainer/background/output-scale/feature-schema record.
A reproducible case explanation with a reference-cohort sensitivity review.

THE OPERATING DETAIL

A contribution needs its reference and output scale.

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

Preserve the reference choice

Keep a synthetic target record fixed while comparing two documented background populations where the method uses one. Retain the explainer settings and resulting attributions. Ask what changed in the comparison before interpreting a different feature ranking as a change in the model itself.

Check the output being explained

Record the selected class and whether the explanation is in a raw-score, transformed or probability-related space. Reconcile the displayed base value, contributions and output using the selected method’s assumptions. A familiar-looking chart is insufficient if the reviewer cannot explain what its numbers represent.

Inspect beyond the aggregate

Compare an individual case with a beeswarm or bar summary over a named dataset. Include a sparse region or unusual case whose pattern differs from the average. Keep correlation, approximation limits and input preprocessing in the interpretation. Model attribution should not be presented as a causal account of the world.

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

SHAP documents beeswarm and bar plots for aggregate views across explanations; these are model-behavior summaries, not causal findings.

Source, dates & preserved provenance

SHAP — primary scope and current identity ↗

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

Artifact SHA-256: afb60626290a664183adec8dc333ef99b4eb6bf583b6934b2f984d582158c923

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

first party asserted

Its README includes tree ensemble, natural language and deep-learning explanation examples; compatibility and assumptions depend on the selected explainer.

Source, dates & preserved provenance

SHAP — primary scope and current identity ↗

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

Artifact SHA-256: afb60626290a664183adec8dc333ef99b4eb6bf583b6934b2f984d582158c923

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

first party asserted

SHAP documents feature attributions connecting a reference value to a model output, including waterfall and force visualizations.

Source, dates & preserved provenance

SHAP — primary scope and current identity ↗

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

Artifact SHA-256: afb60626290a664183adec8dc333ef99b4eb6bf583b6934b2f984d582158c923

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

first party asserted

The SHAP project documents a game-theoretic approach to explaining machine-learning model outputs through feature attributions. Scope: primary documentation only; no independent model test, fairness determination, causal finding or compliance approval.

Source, dates & preserved provenance

SHAP — primary scope and current identity ↗

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

Artifact SHA-256: afb60626290a664183adec8dc333ef99b4eb6bf583b6934b2f984d582158c923

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

Source register

SHAP — 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 an explainer/background/output-scale/feature-schema record for the selected configuration?

Can you provide a reproducible case explanation with a reference-cohort sensitivity review 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.