Acknowledgements

The 36 projects every number is measured against

Every figure on the benchmarks page comes from the real history of these repositories, across the twelve supported languages. The benchmark would not exist without them, and we are grateful to their maintainers and contributors.

Python

fastapi, rich, faker, saleor, wagtail, scrapy

TypeScript

hono, ink, faker-js, excalidraw, outline

JavaScript

express, commander, eslint

Go

gh-cli, hugo

Rust

ripgrep, bat

Java

guava, junit5

C#

powershell, jellyfin

C

redis, curl

C++

rocksdb, fmt

Ruby

homebrew, rubocop

PHP

laravel, composer

Object Pascal

castle-engine, mORMot2, uos, ideU, MSEide/MSEgui

Multi-language

dagster

argot vendors and redistributes none of this code.The harness clones each repository at a pinned commit, reads its history locally, and ships nothing from it. Each project remains under its own license, held by its own authors. Corpora are chosen for history depth and idiom variety — never for how well argot scores on them, and one is never dropped because a fixture fails to fire.

The one thing argot does redistribute

The code-embedding model behind the redundant andmisplaced rules is a 15.6 MB static table argot distilled from jina-embeddings-v2-base-code (Jina AI, Apache-2.0), using the model2vec technique (MinishLab, MIT). Its weights are compiled into the binary and redistributed under Apache-2.0. argot's tokenisation also embeds the tokenizer of microsoft/unixcoder-base (Apache-2.0). Full terms: NOTICE. argot is not affiliated with, endorsed by, or sponsored by any project named here.

And the tools underneath

argot parses twelve languages with tree-sitter and reads history through libgit2. The full dependency graph and its licenses live in the repository. argot itself is MIT-licensed.