Anthropic's Scion is heavily inspired by Mononoke

What it claims

The author helped build Mononoke, Meta's source control server behind Sapling, and says Anthropic's write-up of Scion reuses Mononoke's "derived data" approach. The overlap, the author argues, "goes well beyond having similar insights." They put two descriptions side by side: Anthropic's post on Scion using derived data to make blame and similar operations efficient, and their own 2022 description of Mononoke doing the same.

The summary's "doesn't object" is only half the position. The author is fine with the design being reused, because it is open source. The objection is to the attribution. Crediting Sapling only for "hard-won lessons" and leaving Mononoke out "borders on plagiarism," in the author's words. They also say the post reads as if Claude designed the system, when a Meta team had worked out many of the same ideas in 2017-18, almost a decade earlier.

Useful pointers for an engineer evaluating Scion

• Mononoke is the open-source predecessor. It is the place to study the derived-data design and how it held up in production. The author says the design "has stood the test of time."
• Mononoke's own lineage. The team drew on Google's SrcFS and on write-ahead logs in databases. Neither is the origin of every idea here.
• A contrasting design. A footnote cites Vicent Marti's blog post for Cursor as a system with "similar insights but many key design differences." The author offers it as an example of a related system that is not a copy. It is worth reading next to Scion if you are comparing approaches.

Limits

• This is an opinion post with no technical depth. The side-by-side comparison is presented as two quoted excerpts, and their text is not in the fetched page. The specific overlapping passages cannot be checked from this material.
• The page offers no technical comparison of Scion and Mononoke beyond the derived-data and blame use case.
• The claim that two LLMs point to Mononoke as prior art is a one-line anecdote, with the GPT-6 Astra case credited to Matthew Fowle. No prompts or outputs are shown.
• The author is a former Mononoke engineer, so this is a first-party account and not a neutral one.