Glonce

Mononoke engineer says Anthropic's Scion copies its design without credit

toolsThe Glonce editorsHacker News🔥71Glonce rating71Impact50Novelty35Relevance75Surprise70Players80Credibility70

A blog post titled "Anthropic's Scion is heavily inspired by Mononoke" accuses Anthropic of borrowing the design of Mononoke, the source control server built at Meta, without saying so.

The author sets two passages side by side. In the first, Anthropic describes, the day before the post, how its new version control system, Scion, uses a concept called "derived data" to make operations like blames more efficient. In the second, the author describes, back in 2022 in his Oxide candidate materials, how the Mononoke source control server he helped build uses a concept called "derived data" to make operations like blames more efficient.

The author's view is that the level of imitation goes well beyond having similar insights. He says Anthropic credited Sapling only for "hard-won lessons" and did not mention Mononoke at all, which in his words borders on plagiarism. He adds that the Anthropic post comes off as claiming that Claude designed all this, while in his telling his "extraordinary" team at Meta came up with many of the insights that Claude applied almost a decade later. His team, he notes, also studied plenty of prior art for Mononoke and Sapling, from Google's SrcFS to write-ahead logs in databases.

He is not objecting to the copying itself. He says he is glad the vision his team had back in 2017-18 proved powerful and enduring, and that he has no problem with Anthropic copying Mononoke's design, since part of the reason for open sourcing it was to spread good ideas.

His larger point is about LLMs. As they assist in designing new systems, he writes, they are likely to index heavily on existing designs. Mononoke is open source and in every LLM's training set, and if you ask Opus 5.5 or GPT-6 Astra about prior art, both immediately point out the similarities to Mononoke. He thanks Matthew Fowle in connection with GPT-6 Astra. His conclusion: giving credit to those who came before is now easier and more important than ever, and there is no longer an excuse.

In a footnote he points to Vicent Marti's blog post for Cursor as an example of a system with similar insights but many key design differences.

Key facts

  • The author, who says he helped build the Mononoke source control server at Meta, claims Anthropic's Scion post imitates Mononoke's "derived data" design, which makes operations like blames more efficient.
  • He says Anthropic credited Sapling only for "hard-won lessons" and never mentioned Mononoke, which he says borders on plagiarism.
  • He says he has no problem with the copying itself; his objection is the missing credit and the impression that Claude designed it all.
  • He says Opus 5.5 and GPT-6 Astra both point out the similarities to Mononoke when asked about prior art, since Mononoke is open source and in LLM training sets.
  • He cites Vicent Marti's blog post for Cursor as a system with similar insights but many key design differences.

Why it matters

The dispute is about attribution when an AI company presents a system design. The author's argument is that open source ideas end up in every LLM's training set, so models will keep leaning on existing designs, and that makes naming prior art both easier and more necessary. He frames the omission as a question of credit, not of legality or permission: he says outright he has no problem with the design being copied.

Who it affects

Engineers and teams whose open source work feeds into AI-assisted design, including the Meta team behind Mononoke and Sapling. It also touches Anthropic and any company that publishes technical write-ups of systems designed with LLM help, since readers may take the model or company to be the originator of ideas that came from elsewhere.

How to use it

There is nothing to install or run here. The practical takeaway the author offers is a habit: when an LLM helps design something, ask it about prior art. He says Opus 5.5 and GPT-6 Astra both immediately point out the similarities to Mononoke when asked, and he argues that this leaves no excuse for skipping the credit.

How solid is it

This is one person's first-hand argument, and the claims of imitation and plagiarism are his own judgement. The source text shows only his paraphrase of the Anthropic post, not the post itself, and no response from Anthropic is given. He is a participant, since he says he helped build Mononoke. His statement about what the two models answer is his own report. The source does not describe how Scion works beyond the "derived data" concept and blames being made more efficient.

Risks and caveats

Without the full Anthropic text in the source, the strength of the overlap cannot be judged from this account alone. The author himself notes that other systems, such as the one in Vicent Marti's Cursor post, have similar insights but many key design differences, so similar ideas do not by themselves prove copying. The source does not say whether Anthropic has amended its post.

“I think the level of imitation goes well beyond having similar insights.”

— Author of the post, who says he helped build Mononoke