Two days after Anthropic published its engineering post on Scion, a version control system it says it built to replace Git internally, a former Meta engineer accused it of copying the design of Meta's Mononoke source control server without crediting it. An Anthropic author replied on Hacker News. We have already covered what Scion is, in our write-up of Anthropic replacing Git with Scion and the Scion vs Git comparison. This post is about the dispute, and it is a claimed-versus-verified account because the key allegations rest on private information.
TL;DR: who says what
| Point | Critic (former Mononoke engineer) | Anthropic author (on Hacker News) |
|---|---|---|
| Similarity | "Goes well beyond" similar insights: derived data, changesets, whole data model | Inspired by many things, including "the great ideas in Mononoke's Bonsai data model" |
| Credit | Sapling credited only for "hard-won lessons"; Mononoke not named | Credited Sapling because that is how Meta externally names the whole system; footnote to be added |
| Intent | Says they received private confirmation of intentional copying and personnel overlap | Says there was no intent to obfuscate; many ex-Meta employees at Anthropic liked using Mononoke |
| Originality | Calls the data model "wildly contrarian" and Scion's resemblance deep | Says coming from a large Git repo rather than Mercurial required many novelties; cites Kleppmann's essay for derived data |
| Code | No code-copying claim | No code-copying claim |
| Evidence published | Side-by-side screenshots of both descriptions | Explanation only |
What did the critic publish?
In the first post, dated October 14, 2026, the critic placed Anthropic's description of Scion's derived data next to their own 2022 description of Mononoke's derived data, written in candidate materials for another employer. He argued that the level of imitation "goes well beyond having similar insights," that crediting Sapling only for "hard-won lessons" while not mentioning Mononoke "borders on plagiarism," and that the post reads as if Claude designed everything. The critic added that they had no problem with copying a good design and noted that Mononoke is open source and in every language model's training set.
In a follow-up on October 15, the critic went further, writing that they had privately received confirmation that the resemblance was intentional and that there was overlap in personnel, said the similarity extends to the entire data model and several structures built around it, and compared how Anthropic describes Scion changesets with how Mononoke describes Bonsai changesets, calling them "exactly the same." The critic also pointed out that the name Scion reverses the syllables of Bonsai and drops the B, adding that "This might be a coincidence, or might not."
The critic framed the problem as attribution, not copying: "I am completely fine with good ideas being copied." They also argued the claim lowers the impressiveness of Anthropic's story, since rebuilding a ten-year-old system in five months is different from designing one.
What did Anthropic say?
A Hacker News user posting as the original author of Anthropic's engineering post replied in the discussion thread. The comment, as we read it:
- Sapling was credited because that is how Meta itself refers externally to the whole system, including the Mononoke server, and Anthropic wanted to follow that convention. A footnote will be added to clarify.
- Scion was inspired by many things, including Mononoke's Bonsai data model, and Anthropic did not intend to obfuscate that.
- The derived data concept is also described in Martin Kleppmann's essay on turning the database inside out, which Anthropic cites as inspiration.
- Anthropic did not simply tell Claude to build a version control system; it researched what would work, and starting from a large Git repository, not Mercurial, required many novelties.
- Many ex-Meta employees at Anthropic were inspired by how joyful using source control was because of Mononoke.
We could verify that the comment exists and what it says. We cannot verify that it was written by the person it says, since we cannot authenticate a Hacker News username. A commenter in the thread questioned the username format, and two follow-up questions asked how much of Scion is inspired by Mononoke and whether Anthropic audits for overlap. We saw no answer to those in the replies we read.
What do other commenters say?
The Hacker News thread had 63 points and 11 comments when we read it, so it is early. Paraphrasing, with caution:
- One commenter, gopalv, argued that copycats are a kind of validation, and that the people who built the original now have evidence their approach suits an AI-native future.
- Another, MeetingsBrowser, agreed with the critic's conclusion that Anthropic's public engineering presentations may be bad-faith ripoffs, and doubted that a side project was worth the company's attention.
- A commenter, flakiness, agreed with the high-level points but pointed out that the critic's wording about "large AI companies" would land better with some self-awareness if the critic works at Meta (flakiness's assumption; we have not verified it).
- Some comments were sarcastic or ideological rather than informative, and we skip them.
Opinions differ, and none of these commenters have published a comparison of the two systems.
A short timeline
- October 13, 2026: Anthropic publishes its engineering post on Scion. Our launch coverage summarises its headline figures, including 30,000 merges a day.
- October 14: the critic publishes the first post with side-by-side excerpts, and posts on Bluesky and LinkedIn, according to the follow-up.
- October 15: the critic publishes the follow-up, the Hacker News discussion opens with 63 points, and an Anthropic author comments. At the time we checked, the critic's follow-up said the critic had not seen an update to the original Anthropic post.
That compressed sequence is typical of how attribution disputes now play out: a specific, checkable comparison from someone with first-hand knowledge, a fast response from the company in a community forum rather than in the original post, and a debate about intent that the public evidence cannot settle.
Why the word "Sapling" matters here
The disagreement turns partly on a naming question. Sapling is the name of Meta's source control ecosystem, which includes a client and, in the critic's account, server components such as Mononoke. Anthropic's author says Meta itself uses Sapling externally as the name for the whole system, including the Mononoke server, so crediting Sapling followed that convention. The critic replies that Sapling can mean either the client or the system as a whole, that anyone who worked on Meta's source control team knows the difference, and that a section on changesets should name Bonsai changesets specifically.
Both can be partly right. A credit to an umbrella project is common practice in engineering posts, and a more specific credit is more useful to readers who want to follow the lineage. The promised footnote is the sort of fix that resolves the naming point without settling the larger question of how closely the designs track each other.
What can be checked, and what cannot
Can be checked now:
- Anthropic's post credits Sapling and lists prior art such as Google's Piper; our earlier coverage of it records that.
- The critic's side-by-side excerpts exist on the critic's posts.
- Anthropic's author has publicly acknowledged Mononoke's Bonsai data model as an inspiration and promised a footnote.
- Mononoke is open source and publicly documented.
Cannot be checked yet:
- Intent. "Private confirmation" of deliberate copying is not public evidence.
- Personnel overlap. Anthropic's author says there are ex-Meta employees there; neither side says they worked on Scion's design, and we did not verify it.
- The depth of structural similarity. That would need a careful comparison of both data models by someone who knows each.
- Whether the footnote has been added. We did not see an updated post when we checked.
Why this matters beyond one blog post
The critic made a point worth keeping whatever the facts of this case: language models index heavily on existing designs, and open-source systems are in their training data. Ask a current model about prior art for a new system and it may point straight to the closest open design. That makes giving credit easier, and it makes failing to give credit easier to notice.
For anyone writing up AI-assisted engineering work, a short checklist helps:
- Ask a model, and a human, for prior art before you publish, and read what it finds.
- Name the specific system, not only its ecosystem. "Sapling" and "Mononoke" refer to different things, as the dispute shows.
- Say what is new and what is borrowed, even when you are proud of the result.
- Correct publicly and quickly if someone points out a gap.
- Do not let the story become "the AI invented it" when people did the design work first.
Our comparison of Scion with Git and Cursor's approach shows how many teams are converging on similar ideas for agent-era source control, which makes lineage even more important to record. If you are weighing whether to trust agent-written infrastructure at all, our piece on server-centric vs distributed version control for agents covers the technical trade-offs.
What to watch next
- Whether Anthropic's engineering post gains the promised footnote, and what it says.
- Whether Meta's source control team, or the Mononoke project, comments.
- Whether anyone publishes a rigorous comparison of the two data models.
- Whether other labs adopt similar attribution practices in engineering write-ups.
We will update this post as the facts develop.
Related reading
- Anthropic replaced Git with Scion
- Scion vs Git: how Anthropic's source control differs
- Scion vs Git vs Cursor Origin
- Server-centric vs distributed version control for AI agents
- AI agents are not going rogue: frontier labs skip security basics
Details are as published by the cited authors on October 14 and 15, 2026 and may change.