0003 — Code provenance and license

Status: accepted · Date: 2026-08-01

Context

The project will be published on GitHub — a distribution under the meaning of free licenses. The obligations apply fully from the first commit (§10).

Decision

• Project license: Apache-2.0. As permissive as MIT, plus an explicit patent grant and a contribution clause — the usual default for a library aimed at enterprise adoption. LICENSE file present from initialization; PackageLicenseExpression=Apache-2.0 in the metadata.
• Provenance rule: no transcription of copyleft code. Translating from one language to another is a derivative work. python-Levenshtein (GPL) is therefore excluded — neither transcribed nor even used to generate test data (for hygiene, though the GPL claims nothing over a program's outputs).
• Allowed sources, in order of preference: published papers/pseudo-code (algorithms are not protectable); textbooks and documentation; permissively licensed implementations as a behavior reference only — rapidfuzz (MIT), jellyfish (MIT), textdistance (MIT), scikit-learn (BSD-3). We reproduce inputs/outputs and analogous naming, never the source.
• Test-data generation. rapidfuzz/jellyfish are run by tools/generate_oracles.py to produce the oracle JSON. They are development dependencies, never runtime ones; they are not redistributed.

Consequences

• Each implementation whose inspiration source is worth tracing gets a note in this decisions/ folder.
• NOTICE and THIRD-PARTY-NOTICES.md record attributions; they are updated at the same time as any third-party dependency or resource is added.
• ONNX model weights (Lot 3) are not redistributed in the repository: downloaded at runtime + license documented case by case.