v0.3.3 — paper abstracts in the MCP (extract the claim)

This release surfaces the paper's abstract through the verified-knowledge MCP. The replication_status and radar tools now return the paper's title and abstract (from the OpenAIRE Graph, with JATS/HTML markup stripped) alongside the replication verdicts.

An agent can read the abstract and extract the paper's atomic claim — an AIDA statement — to reason about it, with no language-model key needed on the server side. The verified-knowledge layer now carries the text that turns a paper into a structured claim, pairing the OpenAIRE Graph's metadata with claim-level reasoning and verification.

v0.3.2 — retraction- and supersession-aware verdicts

Replication Radar turns the OpenAIRE Graph into a ranked replication queue and overlays Science Live's verified-knowledge layer: whether a high-impact claim has been independently replicated, and with what verdict, drawn live from the nanopublication network (any signer).

This release makes the verdict overlay aware of a nanopublication's lifecycle. Replication outcomes that have been retracted, invalidated, or superseded by their original author are no longer counted or shown, so a verdict on screen always reflects the current, authoritative version. A third party's retraction cannot suppress someone else's outcome, and disapproval — disagreement rather than retraction — never hides a verdict. The validity check runs as a lightweight, best-effort query, so it can never take down live verdicts.

v0.3.1 — verdict over-attribution fix

Fix verdict over-attribution: a paper linked from a replication Outcome only via a provenance CiTO relation (usesMethodIn / usesDataFrom / credits …) no longer inherits that Outcome's verdict. A study that reuses Phillips et al. 2009's method and was itself Contradicted does not contradict Phillips. Verified live — Phillips et al. 2009 (10.1890/07-2153.1) now shows Validated only (previously "Validated, Contradicted"); the live verified_claims() corpus is correspondingly cleaner.

Source code under MIT; story and verdict index under CC-BY 4.0.

v0.3.0 — live verified-knowledge layer

Replication verdicts are now pulled live from the nanopub network, author-agnostic: every FORRT replication outcome and citation published by anyone, joined and overlaid on the OpenAIRE Graph by DOI — no longer a bundled snapshot (the bundle remains an offline fallback). A new tool, verified_claims, exposes the whole verified-knowledge corpus: every claim the network holds an independent replication verdict for.

This turns Replication Radar into the verified-knowledge layer to pair with the OpenAIRE MCP — one gives an agent the structural research graph (citations, authors, funders), the other answers whether a claim has been independently checked and how it held.

Source code under MIT; story and verdict index under CC-BY 4.0.

v0.2.0 — readiness scoring + richer verdict overlay

Replication targets are now ranked by a transparent replication-readiness score that combines citation impact, the availability of reusable software independent of the original authors, and the presence of reference data — so the most impactful and feasible-to-replicate work rises to the top. Query handling is more robust (de-duplication and term expansion), and the verification overlay is now built reproducibly from the published Science Live FORRT chains, covering seven source works across species-distribution, biodiversity, coastal-ocean, and marine-telemetry research.

Source code under MIT; story and verdict index under CC-BY 4.0.

v0.1.0 — Replication Radar

Replication Radar turns the OpenAIRE Graph into a ranked replication queue. It surfaces high-impact research worth replicating, finds reusable method software that is independent of the original authors (the distinction between a replication and a reproduction, made computable as author-disjointness), and overlays whether a claim has already been independently checked and how it held — the reliability signal that citation-based metrics cannot express.

Distributed as an MCP server exposing three tools — radar, find_independent_software, and replication_status — that any agent can use alongside the OpenAIRE MCP. It queries the public OpenAIRE Graph API and is endpoint-agnostic. Built for the OpenAIRE AI Hackathon (Theme B) and reusable through the FORRT replication template.

Source code under MIT; the accompanying story and verdict index under CC-BY 4.0.