Truth Desk — Scientific Claims Verification Platform
Truth Desk is an autonomous multi-vertical platform that verifies scientific claims
against authoritative databases. It ingests peer-reviewed literature from PMC Open
Access, extracts molecular and biological claims, and cross-references each claim
against the Protein Data Bank (PDB), PubChem, PubMed, and UniProt.
Every verdict is traceable to a primary evidence source. LLMs are used only for
claim extraction — never for verdict generation. The result is a machine-readable
audit trail that researchers, investors, and AI agents can query directly.
Research Verticals
Truth Desk operates across multiple scientific domains. Each vertical uses
domain-specific MeSH queries, curated evidence sources, and a two-pass quality
pipeline to ensure high-fidelity claim verification.
-
Structural Biology — Verifies protein structure claims against
PDB deposited crystal structures. Covers fold topology, binding site geometry,
resolution thresholds, and crystallographic evidence.
-
Salmon Biotech — Verifies aquaculture and marine biotech claims
against PubChem compound data and PubMed literature. Covers feed additives,
growth factors, and disease resistance compounds.
-
Drug Discovery — Coming soon. Will cover small molecule
candidates, target binding affinity, and ADMET property claims.
-
Clinical Genomics — Coming soon. Will cover variant
pathogenicity, gene expression, and clinical association claims.
-
Cancer Biology — Coming soon. Will cover oncogene activity,
tumour suppressor function, and therapeutic target claims.
View all verticals and live statistics
Claims Registry and Knowledge Graph
All verified claims are published in a machine-readable registry accessible at
/api/public/claims.json. The registry
includes claim text, verdict (supported, refuted, or inconclusive), confidence
score, evidence source, and a link to the full audit report.
The interactive knowledge graph at /graph visualises
relationships between scientific documents, extracted claims, and evidence nodes.
Document nodes are colour-coded by vertical domain. Evidence nodes are
colour-coded by verdict. The graph supports filtering by domain, verdict, and
label search, and can be embedded in external sites via iframe.
AI agents can query the platform directly via the MCP tool card at
/.well-known/mcp.json, the markdown summary
at /api/md, and the agent-callable verification endpoint
at /api/public/verify-claim.
Browse the public claims registry