llamastash — third-party attributions
====================================

llamastash itself is MIT-licensed (see `LICENSE`). This NOTICE file lists
third-party sources whose code or curation feeds the benchmark snapshot
regeneration flow (Unit 7 of the v2 plan). The Rust binary does not link
or embed any of the sources below; they run only in the CI workflow
that produces `data/benchmark-snapshot.json`.

When a source is vendored, the upstream commit hash and license are
recorded next to the corresponding adapter under
`scripts/benchmark_sources/`.

Vendored sources
----------------

1. whichllm (https://github.com/Andyyyy64/whichllm)
   License: MIT
   Usage: benchmark-source adapter shape (Open LLM Leaderboard parquet
   + rows-API fallback; Aider polyglot YAML scrape) and the
   AIDER_NAME_TO_HF_IDS curated map.
   Vendored upstream commit: 73cd92f9a35a1c3f02e01ec3bbf09fb135a1df26
   Vendored on: 2026-05-19
   See scripts/benchmark_sources/whichllm.py for the attribution shim.

2. Open LLM Leaderboard (https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard)
   License: Apache-2.0 for the leaderboard code; CC-BY for the data.
   Usage: general-purpose benchmark scores in the snapshot's
   `benchmark_score` field. Accessed via the public
   datasets-server.huggingface.co rows API in
   scripts/benchmark_sources/open_llm_leaderboard.py.

3. Aider polyglot benchmark (https://aider.chat/docs/leaderboards/)
   License: Apache-2.0.
   Usage: code-lane benchmark scores in the snapshot's `benchmark_score`
   field when `task_hints` includes `"code"`. Source data fetched from
   the polyglot_leaderboard.yml in Aider's GitHub repo by
   scripts/benchmark_sources/aider.py.

If you ship a binary build of llamastash, you do not need to redistribute
this NOTICE — none of the above is in the compiled artefact. The file
is part of the source release for transparency about where the snapshot
data came from.
