@cortensor

🛠️ DevLog – Fact-Check API Foundation (WIP, Built on Oracle/Claim-Check Design) We’ve started implementing the fact-check path described in the recent AI Oracle & Claim-Check design - this is work in progress, but the core shape is now in place. 🔹 What Exists Today (WIP Foundation) - Single structured endpoint: POST /api/v1/factcheck (alias: /api/v1/fact-check). - Supports two modes: standard and realtime, each resolving to a dedicated fact-check session via config/env. - Runs redundant checks across multiple miners (minimum redundancy = 3) and uses Cortensor’s core protocol to aggregate them. - Returns a stable, structured response with: - top-level verdict + confidence - per-run / per-miner evidence blocks so agents can inspect how consensus was formed. - realtime mode already has hooks for web/news grounding and source ingestion, but uses the same schema as standard. 🔹 What’s Still TODO (Next Iterations) - This is deliberately a foundation layer: contract + consensus semantics first, richer evidence later. - Upcoming work (not implemented yet, but designed to fit behind the same API): - pluggable evidence providers (news, image/OCR, specialized web) - source trust scoring and weighting - stricter citation validation - domain-specific policies (finance, sports, hoaxes, etc.) Because the API is shaped around mode + policy + consensus + evidence, we can keep iterating internally (more providers, better scoring, tuned policies) without breaking clients that integrate with /api/v1/factcheck today.

0 reply

0 recast

0 reaction