v0.1.1

Documentation only; no change to the code or the results. The README now renders the same on PyPI as on GitHub: the "How it works" diagram is an image, and every link is absolute. Zenodo DOI, project URLs and contact details added.

glossdex 0.1.0

First public release.

• Indexing: local describers (Ollama vision and text models, or text-only), phrase parsing, whole-gloss and per-phrase embeddings (Ollama or sentence-transformers), and an incremental SQLite store.
• Search: blended multi-vector scoring with a rarity-weighted keyword-coverage boost, and the evidence-gated result set: noise floor, evidence gate, floor-anchored band, coverage-gap tightening and conditional padding. One strictness control.
• Threshold profiles for EmbeddingGemma-300M, one per kind of content: embeddinggemma-300m for glosses written by a model, and embeddinggemma-300m/passages for raw text indexed as written (band 0.90, evidence gate 0.52, minimum 1 result), selected automatically with the text-only describer.
• Glossdex.context() for retrieval-augmented generation that can return "not found".
• CLI (index, search, explain, serve, doctor) and a local web UI with a step-by-step explanation and a fixed top-k comparison.
• Benchmark on three public BEIR collections (SciFact, NFCorpus, FiQA) against dense, BM25 and hybrid search at fixed cutoffs, a similarity threshold and Adaptive-k, with scripts and results in benchmarks/beir/. Every setting is chosen on development collections and frozen.

The result-set determination is the subject of U.S. Provisional Patent Application No. 64/172,444 (patent pending); see PATENTS.md.