Anthropic's Scion is heavily inspired by Mononoke

This piece compares Anthropic’s description of Scion’s “derived data” mechanism with an earlier account of Mononoke’s design, arguing the similarities are too close to be coincidental. It reproduces Anthropic’s explanation that changesets are compact and that auxiliary “derived data” structures (file blame, path history, skip lists) are computed recursively from parents to speed queries, then juxtaposes a 2022 description of Mononoke that explains the same idea: using separate Merkle DAGs and indices built from Bonsai changesets, flipping pointers between commits and trees, and a blame-friendly index that makes blame O(number of commits that touch a file) instead of O(total commits). The Mononoke write-up also notes that multiple specialized indices (eight, over time) were cheap to add and that derived data can be computed asynchronously, matching many of Scion’s stated benefits.

The central argument is that Anthropic’s writeup echoes Mononoke’s insights and that Mononoke receives no direct credit while Sapling is only cited for “lessons,” a near-plagiarism concern according to the author. The author acknowledges being proud that Meta’s team’s 2017-18 design remains influential and says copying is acceptable given open source, but stresses that large language models will now surface prior art easily (examples: Opus 5.5 and GPT-6 Astra immediately spot Mononoke) and therefore giving explicit credit is both easier and ethically required. A contrasting example with different design trade-offs is pointed to in Vicent Marti’s Cursor write-up.