Git platforms (Tangled) and preprint services (Chive) are being built on ATProto. Interesting experiments. But something doesn't sit right. Does ATProto actually fit repository-based services?
Short answer: ATProto is a protocol for declarative acts, not institutional trust. Repository-based services—especially academic ones—run on institutional trust. The friction is structural, not incidental.
How Tangled Gets It Right
Tangled's architecture reveals a deliberate compromise.
The PDS stores only issue records and repository pointers. Actual Git repositories live on separate infrastructure called Knot servers. Inside a Knot, you find ordinary bare git repositories organized under DID directories. No attempt to reimplement Git within ATProto.
ATProto handles three things: identity (DIDs), social metadata (issues, stars, collaboration permissions), and discoverability (the AppView as a unified interface). Code itself never becomes an ATProto record.
This is the correct call. ATProto records operate on CRUD semantics. Git's diff-chain model is a fundamentally different beast. The Merkle Search Tree manages content hashes, but that is not version control. Tangled uses ATProto as an identity and discovery layer while keeping code on infrastructure suited for code.
Chive's Ambition, and the Problem It Surfaces
Chive pushes further. Threaded reviews, formal endorsements, community-curated field taxonomy—all stored as ATProto records owned by individual users.
This is where structural friction emerges. Academic data operates under entirely different premises than blogs or social posts.
Three Points of Structural Friction
1. The "View" Is the Value
ATProto's design philosophy: data lives in personal repos, apps provide views. Viewing a Bluesky post through different clients is just a display difference. Blog services like WhiteWind and Leaflet fit this model naturally.
Academic services don't. Semantic Scholar's value isn't in displaying metadata—it's in constructing citation graphs, computing impact metrics, and generating recommendations. Value emerges only from cross-network processing. In ATProto terms, intellectual value concentrates in the AppView, hollowing out the meaning of "data ownership."
2. External Verification Creates Value
A social post is complete when written. A paper becomes academically valuable only after peer review and citation. Institutions guarantee this verification process. Writing "peer-reviewed" in a personal PDS doesn't answer the question: who vouches for that review?
Under ATProto's model, data belongs to individuals. Academic value does not. Research gains value through communal verification and accumulation. This asymmetry cannot be resolved by design.
3. Permanence and Access Control
DOIs exist because academic records require institutional permanence. ATProto supports both account deletion and record deletion by design. If a researcher migrates their PDS or hosting disappears, what happens to the scholarly record? The permanence guarantees provided by CrossRef and DataCite cannot be replaced by DID resolution alone.
Academic data also requires complex access control: embargo periods, provisional preprint access, restricted sharing among co-authors. ATProto records are designed as public data. The mismatch is significant.
The Watershed Between Blogs and Repositories
Blog services fit ATProto because blogs match ATProto's design assumptions. Single author. Public by default. Complete upon writing. Permanence is desirable but not mandatory.
Academic data violates every one of these assumptions. Co-authorship is the norm. Access control is essential. External verification is the source of value. Permanence is institutionally required.
Ignoring this watershed and attempting a full-stack academic platform on ATProto exceeds the protocol's design scope.
Where ATProto's Scope Ends
If ATProto can serve academic use cases at all, the scope is limited. Preprint sharing and open peer commentary—everything public, no institutional verification required—can work. A partial alternative to arXiv. That's a coherent scope.
But the functions currently served by Elsevier, Springer Nature, or national aggregators like Japan's NII (National Institute of Informatics)—trust certification, permanence guarantees, access control, cross-network analysis—these conflict with ATProto's design philosophy at the foundational level.
ATProto is a protocol for declarations, not institutions. Academia runs on institutions. Start from that recognition.
Not Replacement, but Layer Separation
This doesn't make ATProto useless for academia. It means we need to see the current system's problems clearly.
The current system's strength is institutional trust and permanence. DOI + CrossRef has been running for 30 years. Peer review credibility is backed by institutions. This is outside ATProto's scope, and there's no need to replace it.
The current system's weakness is data ownership and portability. Researchers deposit papers behind Elsevier's paywall, register with national aggregators, re-enter profile data across ORCID and Scopus, and watch their personal libraries in tools like Paperpile remain disconnected from every platform. This isn't "working well"—it's failing predictably.
These are different layers. The institutional trust layer requires existing infrastructure. The data ownership and portability layer is exactly where ATProto has room to improve things.
The future of academic infrastructure is not replacement but layer separation. DOIs and CrossRef stay. Institutional peer review stays. On top of that, a protocol like ATProto can handle personal data ownership, discoverability, and open commentary. Trying to solve everything with a single protocol exceeds any protocol's design scope. Separate the layers. Use the right mechanism for each. That's where the path is.