Anthropic's Scion Accused of Copying Meta's Mononoke Design

Anthropic's Scion Draws Heavily From Meta's Mononoke, Engineer Alleges

Anthropic recently unveiled Scion, a new version control system designed to handle large-scale repositories with greater efficiency. But a former Meta engineer is raising concerns that Scion's core architecture closely mirrors Mononoke, the open-source source control server developed at Facebook between 2017 and 2018 — without proper attribution.

In a detailed blog post, the engineer, who helped build Mononoke, points out that Anthropic's description of Scion's "derived data" mechanism is nearly identical to the concept used in Mononoke. The post includes side-by-side comparisons of Anthropic's engineering blog and the engineer's own 2022 write-up, highlighting the striking similarities in both terminology and design philosophy.

The 'Derived Data' Concept: A Closer Look

At the heart of the dispute is the concept of derived data — precomputed indexes that accelerate operations like blame and file history. In both Scion and Mononoke, derived data is computed recursively from changesets, with each type defined in terms of its parents' values.

This approach offers several benefits, as Anthropic's engineering post explains: it allows for easy addition of new index types, enables eager or deferred computation, and keeps internal data structures hidden from clients. Mononoke's implementation, described by the engineer in 2022, follows the same pattern, using a separate Merkle DAG with a Git-like tree/blob model to handle diff operations efficiently.

One of the key advantages of this design is its impact on blame operations. In traditional Git, blame can be slow on large repositories, requiring a traversal of the entire commit history. Mononoke's derived data approach reduces this to O(number of commits that touch the file), a significant improvement. The system now supports eight separate indexes, each tailored to specific use cases like copy detection and path-based file search.

Crediting Prior Art: Sapling vs. Mononoke

The former engineer's main grievance is not that Anthropic adopted the design — Mononoke is open source, and the team has always encouraged the spread of good ideas. Rather, it's the lack of credit where credit is due.

Anthropic's blog post mentions Sapling, Meta's source control client, for "hard-won lessons," but makes no mention of Mononoke, the server that pioneered the derived data architecture. The engineer argues this omission "borders on plagiarism," especially given that both systems share not just similar insights but nearly identical terminology and implementation details.

"The fact that Sapling was only credited for 'hard-won lessons', and that Mononoke was not mentioned at all, borders on plagiarism," the engineer wrote. "The post comes off as Anthropic claiming that Claude designed all this, while the reality is that our team at Meta came up with many of the insights that Claude dutifully applied almost a decade later."

The Role of LLMs in System Design

This controversy highlights a growing concern in the tech industry: as large language models (LLMs) become more involved in designing new systems, they are likely to index heavily on existing designs found in their training data. Mononoke is open source and part of every major LLM's training set, making it a natural reference point for AI-assisted design.

Interestingly, when asked about prior art for Scion's architecture, both Claude Opus and GPT-6 immediately point to Mononoke as the inspiration. This suggests that the similarity is not just coincidental but deeply embedded in the training data that informs these models.

The engineer argues that giving credit to prior work is now easier — and more important — than ever. "There is no longer an excuse," they wrote, emphasizing that attribution should be a standard practice in AI-assisted development.

Industry Reactions and Comparisons

The tech community has responded with a mix of skepticism and support. Some developers have noted that while Scion and Mononoke share core concepts, there are key differences in implementation. For example, Cursor's Vicent Marti wrote a blog post about scaling Git that addresses similar challenges but with different design choices.

"I think the level of imitation goes well beyond having similar insights," the engineer said, contrasting Scion's approach with Marti's work, which offers an alternative solution to the same problem. This distinction is important for understanding where the line between inspiration and imitation lies.

At the same time, the engineer acknowledged that they do not have a problem with Anthropic copying Mononoke's design. "This is part of why we put out open source — to benefit humanity through the spread of good ideas," they wrote. The issue is purely about transparency and giving credit where it's due.

Why This Matters for the Future of VCS

This incident raises important questions about the future of version control systems and the role of AI in their development. As repositories grow larger and more complex, the need for efficient systems like Scion and Mononoke will only increase. But so too will the need for proper attribution and ethical AI practices.

For developers evaluating new version control architectures, this debate serves as a reminder to look beyond the surface and understand the lineage of the technologies they adopt. It also underscores the importance of open-source communities in driving innovation and the responsibility of companies to acknowledge their contributions.

As the tech industry continues to grapple with these issues, the Scion-Mononoke controversy may become a case study in how AI-assisted design should — and shouldn't — handle prior art. For now, the ball is in Anthropic's court to respond to these allegations and clarify the origins of Scion's design.