Automatic calibration of semantic search thresholds per programming language, new CLI commands, and codebase refactoring.
FonctionnalitéAdaptive ML thresholds - automatic calibration of semantic search thresholds per programming language
FonctionnalitéBootstrap mode - collects 100% of distance observations until 500 samples
FonctionnalitéProduction mode - switches to 10% sampling after bootstrap completes
FonctionnalitéAuto-calibration - triggers automatically when bootstrap completes
FonctionnalitéPercentile-based thresholds - uses p90 for max_distance, p10 for min_similarity
FonctionnalitéPer-language calibration with fallback cascade (language → global → defaults)
FonctionnalitéNew `calibrate` CLI command - manual threshold calibration with `--dry-run` option
FonctionnalitéML stats in `stats` command - shows bootstrap progress, observations per language, calibrated thresholds
FonctionnalitéNew database tables - `distance_observations` and `threshold_calibration` for ML data
FonctionnalitéCI workflows for `dev` branch - tests, Clippy, format checks, and multi-platform builds
AméliorationRefactored `store.rs` (2108 lines → 8 modules) - better code organization with dedicated modules for files, symbols, chunks, dependencies, observations, calibrations, and tests
AméliorationRefactored `engine.rs` (1049 lines → 5 modules) - cleaner architecture with dedicated modules for search, threshold, analysis, and tests
BreakingSchema version bumped to 3 - triggers automatic reindex on upgrade