morai

Review a fitted mortality model (GLM, GAM, CatBoost, or Neural) focused on actual-vs-expected (A/E) analysis. Use when evaluating model fit, segment-level A/E, rate comparisons vs VBT, partial dependence, and overfitting checks. Model-family-agnostic; branches into family-specific diagnostics in the addendum section.

Build a mortality rate table from a fitted model. Use when generating a 1-D or 2-D rate table, applying graduation, adding an ultimate period, building select/ultimate structure, or producing an output table for downstream use. Distinct from model-review — this skill produces a deliverable table, not diagnostics.

Neural-network-specific diagnostics for a fitted mortality model. Use AFTER running the model-review skill to cover the items unique to neural nets — SHAP via KernelExplainer, categorical embedding analysis (cosine similarity, PCA), and training/validation loss inspection.