Changelog

All notable changes to Fuzler are documented in this file. The project follows Semantic Versioning.

0.1.3 - 2026-08-29

Add configurable similarity_score/3, detailed compare/2, native batch scoring, stable top-k ranking, and explicit algorithm versioning.
Add NFC/NFKC normalisation, canonical Unicode equivalence, optional diacritic removal, and grapheme-aware edit distance.
Add per-input, token, batch-size, and total-byte resource limits.
Add a reviewed golden dataset with score intervals, precision/recall, and ranking expectations.
Add StreamData and Proptest properties, a libFuzzer target and workflow, deterministic benchmarks with regression budgets, and an isolated Hex package smoke test.

Normalise case, punctuation, and repeated whitespace consistently before scoring.
Run the native comparison on a dirty CPU scheduler.
Keep token matching symmetric and account for repeated tokens as a multiset.
Penalise partial matches according to their coverage of the longer input.
Move native loading behind a private module while keeping the public Fuzler.similarity_score/2 API unchanged.
Expand HexDocs and package metadata, and add reproducible CI and release workflows.
Require Rust 1.82 for source builds and verify it in CI.

Do not use Hamming distance for unequal-length inputs; values such as "a" and "ab" can no longer be reported as identical.
Do not combine weak matches found in unrelated parts of a target.
Keep UTF-8 token-window slicing on valid byte boundaries.
Catch unexpected Rust panics without building the release NIF with panic = "abort".
Make a missing NIF fail the test suite instead of silently bypassing all assertions.

Add ExUnit doctests, a 150-example behavioural corpus, generated properties, invariant and concurrency coverage, malformed-input cases, and native Rust tests.
Enforce more than 90% Elixir coverage and warnings-as-errors in CI.
Run strict Credo, Dialyzer, Hex dependency audit, HexDocs verification, package validation, Rust formatting, tests, and Clippy in CI.