A trust graph for prediction reputation on Avalanche.

1. Abstract

The current Eva Protocol product is not a trading venue and not a marketing-only landing page. It is a prediction reputation layer with market pages, thesis pages, predictor profiles, and evidence tools backed by the same trust graph.

The system is Avalanche-first. `EvaTrustGraph` is the canonical source of registered predictor identity and trust. ERC-8004 registries remain part of the trust boundary for identity, validation receipts, and reputation events.

2. Product

• Predictors publish market theses through the website or explicit @evapredicts commands.
• External venues provide odds; Eva stores rationale, source links, and copy intent.
• Evidence tools verify supporting claims when a thesis needs a stronger proof trail.
• Registered Eva identities make predictor reputation graph-backed instead of purely social.

Prediction theses are the primary product record, while article verification remains a supporting evidence primitive.

3. Prediction Flow

1. User publishes or is credited with a market thesis.
2. Eva stores the market, selected outcome, odds snapshot, rationale, and evidence links.
3. Other users can preview copy intent, counter, or inspect the claim trail.
4. Unclaimed X profiles can accumulate offchain market records.
5. Registered identities can later receive reputation feedback after durable resolutions.
6. Verification reports remain available through `/api/verify` as evidence objects.

4. Architecture

Vercel is the canonical deployment target for the product surface. The frontend is a dynamic Next.js app, and the backend routes are exposed through the same domain so markets, theses, predictors, trust, and evidence views do not drift across environments.

5. Contract Boundary

• Source of truth: the deployed `EvaTrustGraph` contract on Avalanche.
• Generated interfaces: frontend and backend both consume ABI output generated from the Solidity artifact instead of maintaining hand-written partial ABIs.
• Identity model: the Eva oracle uses agent #1599. Graph-backed predictors register their own wallet and agent identity before resolved outcomes can affect canonical trust.

6. Public API

The public surface is intentionally narrow and explicit:

• GET /api/markets and GET /api/markets/:id return external market context.
• POST /api/theses and GET /api/theses/:id create and read prediction theses.
• GET /api/predictors returns product and graph-backed predictor records.
• POST /api/copy-preview records external-link-only copy intent.
• POST /api/verify accepts a source URL and returns a scored evidence report.
• GET /api/article and GET /api/article/:id remain available as the verified source archive for older links and evidence references.
• GET /api/curators and GET /api/curator/:id remain available for graph identity records that now appear in-product as predictor trust.
• GET /api/trust/:address reads trust against the same tag semantics the backend writes.

x402 and native trade execution remain out of scope for v1. Payment enforcement is not claimed as live unless request verification is actually implemented end-to-end.

7. Roadmap

8. Boundary

Eva does not execute trades in v1. It does not custody funds, place market orders, or operate a native prediction market. The live product records theses and reputation around external markets.

The trust graph, prediction-layer APIs, and evidence tools are the canonical product surface.