Add explainers/lime.md.
SHAP Values already explains one model-explainability technique; LIME (Local Interpretable Model-agnostic Explanations) is the other major one, and currently only gets a single comparison-table row inside that explainer (| LIME | Local only | ✗ Approximate | ✓ Yes | Slow |) with no explanation of what it actually is. Cover what LIME does (perturbs an input, fits a simple local surrogate model around one prediction, reads the surrogate's coefficients as the explanation), how that differs from SHAP's game-theoretic (Shapley value) approach, and why 'local only' and 'approximate' are real trade-offs a fairness audit needs to know about before trusting either tool's explanation of a flagged decision.
Suggested structure (match shap-values.md): one-sentence definition -> why it matters for fairness -> core concept -> concrete example tied to a repo audit -> runnable detection code (using the lime package) -> limitations -> related concepts/projects/further reading.
Related: shap-values, confounding-variable.
See CONTRIBUTING.md: add explainers/<slug>.md plus a one-line entry in assets/explainers-data.json; the build script generates the page.
Freeze-safe: explainers are fully open for contribution during the paper freeze. If you quote any Fair Code benchmark result, use the frozen numbers in paper/results-frozen/ (never re-run your own) - see CLAUDE.md.
Add
explainers/lime.md.SHAP Values already explains one model-explainability technique; LIME (Local Interpretable Model-agnostic Explanations) is the other major one, and currently only gets a single comparison-table row inside that explainer (
| LIME | Local only | ✗ Approximate | ✓ Yes | Slow |) with no explanation of what it actually is. Cover what LIME does (perturbs an input, fits a simple local surrogate model around one prediction, reads the surrogate's coefficients as the explanation), how that differs from SHAP's game-theoretic (Shapley value) approach, and why 'local only' and 'approximate' are real trade-offs a fairness audit needs to know about before trusting either tool's explanation of a flagged decision.Suggested structure (match shap-values.md): one-sentence definition -> why it matters for fairness -> core concept -> concrete example tied to a repo audit -> runnable detection code (using the
limepackage) -> limitations -> related concepts/projects/further reading.Related: shap-values, confounding-variable.
See CONTRIBUTING.md: add
explainers/<slug>.mdplus a one-line entry inassets/explainers-data.json; the build script generates the page.