Release Notes: v0.1.3¶

Summary¶

This release improves query performance, chunk-aware indexing internals, and cache correctness for repeated query workflows.

Highlights¶

Query and Index Performance¶

• Added persistent Tantivy lexical index reuse across CLI runs.
• Added binary core snapshot cache for fast in-memory index restoration.
• Added dense doc_u32 score accumulators and segment-tree top-k selection.
• Added trie-backed term/entity lookup with conservative prefix fallback.

Chunking and Retrieval Structure¶

• Added section chunk metadata with line ranges (start_line, end_line).
• Added hybrid chunking strategy (line-stable + token-aware splitting).
• Added configurable chunk parameters:
• --chunk-strategy heading|line|hybrid
• --chunk-lines
• --chunk-overlap
• --chunk-target-tokens
• --chunk-max-tokens

Incremental/Change Awareness¶

• Added corpus fingerprint validation for cache reuse.
• Fingerprint now includes content hashing for robust change detection.
• Cache reload now rebuilds only when source corpus/settings fingerprints differ.

Query Output and UX¶

• Query output now includes elapsed time metadata:
• elapsed_ms
• result_count
• results
• Deduplicated matched_entities and matched_terms in query results.

Internal Data Structures¶

• Added chunk-level postings and forward indexes.
• Added doc↔chunk adjacency mappings.
• Added interval-tree support for changed line-range to chunk impact mapping.

Compatibility¶

• Existing CLI commands remain compatible.
• Query command remains:
• --query "..."
• Index command remains:
• --index
• JSON cache remains available for readability; binary cache is additive.