More on Anthropic's Scion ripping off Mononoke

What it claims

The author, who led the 2018 design of Meta's Mononoke data model, escalates the earlier post. The original post said the similarity was probably unintentional and the missing credit was an oversight. This one says it was deliberate. The author reports private confirmation, after the first post went up, that Scion "intentionally and blatantly" copied Mononoke and that the two teams shared personnel. The author says the overlap goes beyond derived-data indexing and covers:

• The entire changeset data model. Scion changesets are described as "exactly the same" as Bonsai changesets. The author also notes that "Scion" reverses the syllables of "Bonsai" without the "B", and says this may be a coincidence.
• The derived indexes built around it, such as blame, along with the same terminology and framing.
• A server-side API for version-control queries and mutations. The author does not claim this is novel, since earlier systems had it, but says pointing an LLM at Mononoke would likely produce something very similar.

The author still calls a 5-month rebuild "quite stunning." Their objection is that the result was presented as original work rather than as a rebuild of Mononoke.

How it works (the lineage being claimed)

• Bonsai is unusual. It is a write-ahead-log-style Merkle DAG. In 2018 the prevailing approach was Git-like data models.
• It needed new algorithms. The author names server-side rebases as one of several novel algorithms in Mononoke, and says the model has since been stretched "far outside its initial design parameters" and still holds up.
• Derived-data indexes are a consequence of the model. Systems built on Bonsai need indexes like blame, so the author's complaint is about matching terminology, not about indexes existing.
• "Sapling" is ambiguous. It can mean the sl client, which was forked from hg, or the whole system. Mononoke is the server side. Crediting "Sapling" for "hard-won lessons" therefore does not name the component that was copied.
• Contrast with Lore VCS. The author says Lore can fairly claim inspiration from Mononoke because it differs in several ways. Scion, in their view, does not differ enough.

For an engineer evaluating Scion, the practical point is that Mononoke and its Bonsai changesets are the prior art to study for design rationale and known limits.

Limits

• The page is one side of a dispute. The intent and personnel claims rest on unnamed private sources that the reader cannot verify.
• The side-by-side comparison is missing from the fetched text. The page quotes Anthropic's description of Scion changesets next to the author's 2022 Oxide description of Bonsai, but neither quote came through. The "exactly the same" claim cannot be checked from this page.
• Anthropic's side is not described. The top-of-page update says Anthropic responded on Hacker News and added "sufficient attribution" to its blog post, but gives no content from that response. The author says they "stand by" both posts as of when they were written.
• The body predates the update. Its closing statements, that no update had appeared and that the author now assumes Anthropic's engineering posts are "largely bad-faith ripoffs," were written before the attribution was added.