01
/seer/dataset/1.0.0
Dataset
Content-addressed AI or research data with a declared schema and license.
- type
- root: CID
- schema
- license
RFC · SDEP-0001 · Draft
An open base layer for AI data, collective evaluation and agents.
Seer Data & Evaluation Protocol (SDEP) turns datasets, rubrics, evaluations and signals into content-addressed, versioned, composable artifacts — carried by libp2p, IPFS, Multiformats and IPLD.
Content-addressed
CID-native artifacts
Verifiable
Provenance graphs
Composable
Off-frontend usage
Peer-to-peer
libp2p transport
Devcon 8 · Community Hub
Ethereum DPG Study Group with libp2p, IPFS and SEER
We are applying to host a Community Hub at Devcon 8 in India. Community Hubs are physical spaces inside Devcon where communities host their own programming, workshops, discussions and research sessions.
This year’s Hubs are going back to the roots: Ethereum’s core principles and the communities strengthening them. Our proposal connects those principles to public goods funding, contribution attribution, and open infrastructure — bringing together builders, researchers, grant operators, and community friends to explore better public goods funding with open AI models.
Learn about Devcon →Devcon 8 · India
Road to Devcon · Workshop · Aug 3–5
Data curation and annotation with libp2p, IPFS and py-libp2p
A three-day Road to Devcon workshop held August 3–5, focused on hands-on data curation and data annotation using libp2p, IPFS, Python libraries and py-libp2p — the same primitives SDEP builds on.
Organized by Johanna Moran and Manu Sheel Gupta in collaboration with the India AI Mission under NIELIT and MEITY, Government of India.
Foundation
py-libp2p — the flexible networking stack
Inspection
Wireshark dissectors — X-ray vision for P2P traffic
Telemetry
Luminar prototype — real-time P2P monitoring
§ 01 · Thesis
Seer is not an app. It's a protocol.
The existing Seer experience already gestures at something larger. Deep Seer lets users publish predictions and run strategies. AI Prediction Markets aggregate collective intelligence around hard questions. The Ethereum evaluation work asks contributors to score open-source projects across structured dimensions.
Underneath, the same primitive keeps appearing: a mechanism for collective intelligence around complex evaluation. SDEP makes that primitive protocol-native — so contributors, agents and dapps can build on the same substrate.
§ 02 · Core objects
Six primitives. One shared graph.
Every SDEP object is content-addressed and linked. Consume any of them from an application, an agent, or a dapp — no Seer API required.
02
/seer/rubric/1.0.0
Rubric
Explicit, immutable evaluation criteria and weights. A new CID for every change.
- criteria[]
- weight
- version
- immutable
03
/seer/evaluation/1.0.0
Evaluation
The result of applying a rubric to a subject and dataset, with evidence.
- dataset
- rubric
- subject
- score
- evidence[]
04
/seer/signal/1.0.0
Signal
A single contribution: human vote, market price, model output, agent score.
- subject
- dimension
- value
- source
05
/seer/provenance/1.0.0
Provenance
The verifiable graph linking inputs and outputs across every operation.
- actor
- operation
- inputs[]
- outputs[]
06
/seer/agent/1.0.0
Agent
A network peer with declared capabilities and supported schemas.
- peerId
- capabilities[]
- schemas[]
§ 03 · Pipeline
From raw data to a verifiable result — as a single content-addressed graph.
- 01Contributorpublishes datasetbafy…d3f2
- 02Authordefines rubricbafy…9ab1
- 03Agentruns evaluation12D3Koo…
- 04Communitycontributes signalsn × Signal
- 05Aggregatorcomputes weighted score0.82 ± 0.04
- 06Dapp / Agentconsumes resultIPLD link
§ 04 · Architecture
A protocol layer, not another blockchain.
SDEP composes proven open infrastructure. Seer defines the objects, the events and the agent contracts — the network moves the bits.
▸ libp2p
▸ IPFS
▸ Multiformats
▸ IPLD
Applications
Deep Seer · AI Markets · dapps · agents
L7
Seer SDK / API
Ergonomic developer interface
L6
SDEP Protocol
Dataset · Rubric · Evaluation · Signal · Provenance
L5
IPLD
Linked, content-addressed data graphs
L4
IPFS
Content-addressed storage & distribution
L3
Multiformats
CIDs, multihash, multicodec — future-proof identity
L2
libp2p
Peer identity, DHT, GossipSub, req/resp
L1
§ 05 · Example
A rubric is a protocol object.
Once published, a rubric is immutable. Any change produces a new CID. An evaluation isn't just "Project X scored 72." It's "Project X scored 72 using v3 of this rubric, against v5 of this dataset, from these signals."
That reproducibility is the whole point.
rubric · repository-originality · v1.0bafy…9ab1
{
"type": "seer/rubric",
"version": "1.0",
"name": "repository-originality",
"criteria": [
{ "id": "code-originality", "weight": 0.30 },
{ "id": "historical-development", "weight": 0.25 },
{ "id": "novel-contribution", "weight": 0.45 }
]
}evaluation · go-ethereum · appliedbafy…7c02
{
"type": "seer/evaluation",
"dataset": "bafy…d3f2",
"rubric": "bafy…9ab1",
"subject": "github:ethereum/go-ethereum",
"evaluator":"12D3KooW…",
"score": 0.82,
"confidence": 0.76,
"evidence": ["bafy…", "bafy…"],
"provenance":"bafy…"
}§ 06 · Initial vertical
Ethereum open-source, evaluated as a public artifact.
Seer already touches a large surface of Ethereum repositories and evaluation questions. SDEP makes each category — execution clients, consensus clients, dev tools, security, infra, libraries — an addressable evaluation subject with its own rubric.
github:ethereum/go-ethereum
weighted · rubric v3 · dataset v5
Security
0.91
Reliability
0.84
Maintenance
0.78
Adoption
0.88
Originality
0.72
Ecosystem impact
0.86
aggregate0.82
§ 07 · Sources
Two documents. One direction.
The strategic proposal frames the opportunity; the technical RFC describes how it's built. A third document tracks the collaboration with Public AI and the Berkman Center at Harvard.
Strategic proposal
Seer as an open AI data, evaluation and agent layer
The full case for turning Seer into an open protocol layer at the intersection of AI data, agents, evaluation and prediction markets.
→ tinyurl.com/SEER-libp2p-Open-AI-Data-Eval
Technical RFC
SDEP architecture on libp2p, IPFS, Multiformats & IPLD
The proposed data model, protocol IDs and network topology — built on Ethereum's open infrastructure libraries.
→ tinyurl.com/seer-libp2p-technical-rfc
Collaboration
Whitepaper with Public AI & the Berkman Center
Discussion thread on the pldg-dashboard repository, plus the licensing note co-developed with Public AI and libp2p.
→ github.com/seetadev/pldg-dashboard/discussions/104
→ tinyurl.com/License-PublicAI-SEER-libp2p
§ 08 · Milestones
A staged path from spec to a live agent network.
M1
Protocol definition
SDEP object types, IPLD schemas, CID conventions, protocol IDs, dataset licensing metadata.
M2
Reference node
seer-node with libp2p, IPFS integration, dataset publish + retrieval, DHT discovery.
M3
Evaluation layer
Rubrics, evaluations, signals and pluggable weighted aggregation.
M4
Deep Seer integration
Convert prediction and evaluation artifacts into SDEP objects, backward compatible.
M5
Agent network
Agent capability discovery, dataset discovery, evaluation execution, publication.
Apache-2.0 reference implementationCC-BY-4.0 specificationMonorepo · seer-protocol/
Call for contribution
Seer owns the product. The ecosystem can build the protocol.
The specification is a draft. Contributors, agent authors and dapp builders are invited to shape SDEP alongside the reference node.