NEWS

ratingtables 0.2.2 (2026-09-05)

Substantially improved non-trace batch-rating performance for standard rating plans. Exact lookup structures and rating-step metadata are now precompiled when a rating plan is created, and rating-data representations used repeatedly by lookups are cached once per batch.
Reworked exact batch lookup to use keyed vector matching while preserving rate-set selection, multi-variable interactions, specificity fallback, and automatic date-based rate-set selection. Plans requiring unsupported vectorized behavior continue to use the general reference engine.
Improved entity-workflow performance. Common sum, mean, and count aggregations now use grouped vectorized passes, and simple unique-key entity joins use direct matching rather than a general merge.
score_entity_rows() continues to provide the optimized non-trace entity-scoring path; rate_entities() retains detailed trace output for audit and reconciliation workflows.
Internal compiled caches are rebuilt automatically when factor lookup structure or rating specifications change. Value-only term_value changes can reuse the compiled lookup structure.
These performance changes do not require changes to existing factor-table or rating-specification formats and add no new runtime dependencies.
Added regression tests comparing optimized batch results with the trace-producing reference engine and added a repository-only performance benchmark script.