Methodology & data provenance

Where every label and score comes from — and how it's computed.

Replication Radar · human-readable here, machine-readable as methodology.json

← Back to the Radar

Replication Radar architecture: a researcher (on the static web app) and an AI agent (on the replication-radar MCP, run beside the OpenAIRE MCP) both read the OpenAIRE Graph, the Science Live nanopublication network, and GitHub / Software Heritage / Zenodo into a ranked queue with verdict overlays and software-reuse signals. Proving a claim publishes a signed constellation back to the nanopublication network, closing the loop.
How it fits together. Two entry points — a researcher on the web app, an AI agent on the replication-radar MCP run beside the OpenAIRE MCP — read the same grounded sources: the OpenAIRE Graph, the Science Live nanopublications, and GitHub / Software Heritage / Zenodo. Proving a claim publishes a signed constellation back to the network, so the next search finds it already proven.