Further Reflections on Anthropic's Scion and the Mononoke Plagiarism

Update: Nick G. from Anthropic has since replied on Hacker News, and the Anthropic blog post now includes the necessary attribution.

Following my previous piece regarding how Anthropic's Scion appears to be a derivative of Mononoke, I've encountered various inquiries and concerns. In my initial post, I attempted to remain charitable, suggesting that the similarities might have been accidental or the result of a failure in due diligence.

However, since then, I have received private confirmation that the appropriation of Mononoke was intentional and blatant, specifically noting an overlap in the personnel involved. While I suspected this previously, I didn't feel comfortable stating it publicly until I had confirmation.

Beyond "Derived Data"

Some critics argued that "derived data" is simply a common indexing concept. To be clear: derived data was merely the most obvious red flag. The similarities run much deeper, encompassing the entire data model and the surrounding architectural structures.

Comparative Analysis of Features

The naming convention is particularly telling. If you look at the transformation: $\text{Bonsai} \xrightarrow{\text{Syllable Swap + Drop 'B'}} \text{Scion}$

The Technical Pedigree

The resemblance extends to the server-side API used for version control mutations and queries. While I won't claim this specific API is entirely novel—as other systems likely utilized similar patterns—it is exactly what you would get if you fed Mononoke's documentation into Claude and asked it to replicate the system.

The Novelty of the Model

In 2018, the industry standard was (and largely remains) the Git-style data model. Mononoke's approach was a radical departure:

Building a write-ahead log-style Merkle DAG required solving formidable challenges, such as novel algorithms for server-side rebases. I remember having to defend this architecture to multiple levels of management at Meta; it is rare for "skip-level" executives to dive that deep into a data model. That system has now functioned for nearly a decade, and its creation remains a highlight of my professional life.

The Ethics of Attribution

Given the personnel overlap, it is impossible that the Scion team was unaware of Mononoke.

The core issue here is not the act of copying, but the lack of attribution.

I have no problem with great ideas being adopted. I do, however, have a problem when the "serial numbers" are stripped off—especially when done intentionally.

Clarifying the Ecosystem

Some have conflated Mononoke with Sapling. It is important to distinguish between:

• sl: The VCS client (originally a fork of hg).
• Mononoke: The underlying system.

Anyone from the Meta source control team knows the difference. While the Lore VCS claims inspiration from Mononoke, it maintains distinct differences. Scion, conversely, is a near-mirror. Simply crediting "Sapling" for "hard-won lessons" is a dishonest simplification.

The "LLM Narrative" vs. Reality

Anthropic's marketing of this achievement is significantly diminished by the truth. Consider the two different narratives:

• Narrative A: Our LLM designed a complex distributed system, productionized it, and solved all the edge cases in just 5 months.
• Narrative B: The Meta team spent 10 years building and productionizing a system; we pointed our LLM at it and rebuilt it in 5 months.

While Narrative B is still an impressive feat of engineering speed, it is fundamentally different from claiming original invention.

Summary of "Borrowed" Elements

• Write-ahead log Merkle DAG
• Bonsai $\rightarrow$ Scion Changeset logic
• Derived data indexing patterns
• Server-side mutation API

Final Thoughts

Engineering is largely about adapting existing work to new contexts. When you acknowledge your predecessors, you ensure that you will be acknowledged by those who follow.

On a personal level, while I am no longer on the Mononoke team, there are people still carrying that torch. It is disheartening to see their work plagiarized and presented as a fresh innovation. Imagine the frustration of the current maintainers seeing their life's work wholesale-copied and rebranded.

This concludes my thoughts on the matter, two days after my initial post.