March 2026 monthly summary for apache/lucene development: Delivered probabilistic hybrid search capabilities with BayesianScoreQuery and LogOddsFusionQuery; replaced per-term BayesianBM25Similarity with a query-level BayesianScoreQuery to preserve BM25 ranking while generating probability scores; added extensive tests for hybrid search across text and vector fields; improved numeric stability and formatting; expanded test coverage for vector+text Boolean combinations; aligned with performance-review expectations. This work improves search relevance, ranking reliability, and modularity for scalable hybrid search in production.

February 2026 monthly summary for embeddings-benchmark/mteb: Delivered Bayesian BM25 (bb25) probabilistic scoring to enable hybrid search fusion. The bb25 baseline converts BM25 scores into calibrated probabilities in [0,1] while preserving rankings with the default prior_weight=0.0. The release includes backend integration using bm25s, focused refactoring for consistency, and comprehensive testing. Documentation updates clarify calibration and prior context; code style improvements (ruff) and naming consistency (encode->_encode) were applied to improve maintainability and future extensibility.