Litmuz
Litmuz is a claim-level verification layer for life-sciences research agents.
Core trick
Splits a memo into atomic claims, then checks genomic claims deterministically against reference datasets, returning contradicted or cannot-confirm instead of guessing.
Description
Litmuz is a claim-level verification layer for life-sciences research agents. An AI agent's memo is fluent and cited, but some claims are unsupported or flatly contradict the primary data, and nothing checks them one by one before they become results. Litmuz breaks a memo into atomic claims and returns an honest, per-claim verdict (grounded, needs review, or flagged), routes anything uncertain to a human, and never auto-passes a safety-critical claim.
For this hackathon I added a genomic mode that verifies claims directly against Gladstone Institutes datasets: the Pollard lab's Human Accelerated Regions and the Zoonomia mammalian-constraint work. The check is deterministic, with no language model in the trust-critical path. A claim such as "HAR1 shows no human-specific change" is flagged as contradicted, with the exact evidence cited, and anything outside the curated reference returns an honest "cannot confirm." It is the verification layer autonomous genomics research is missing.
litmuz.co
README excerpt
# Litmuz **A claim-level verification layer for life-sciences research agents.** Paste an AI agent's research memo. Litmuz breaks it into atomic claims, checks each claim's citations against the primary literature, retrieves evidence, judges entailment with Claude, applies a safety gate, and returns an auditable, per-claim traffic-light verdict: - 🟢 **green — grounded**: supported by evidence, with a clean, resolvable citation. - 🟡 **yellow — needs review**: unverifiable, unsupported, or safety-critical. Never a pass. - 🔴 **red — flagged**: contradicted by the evidence, or the citation is fabricated. Litmuz is built on one principle — **honest negatives**. It never dresses an unsupported claim as a pass, and safety-critical claims (a dose, an indication, a molecular target) never auto-pass, no matter how confident the model is. --- ## The problem AI research agents are fluent, fast, and confidently wrong in ways that are expensive in the life sciences. They cite papers that don't exist, cite real papers that don't actually support the claim, and state doses or indications with the same certainty as a well-grounded fact. A reader cannot tell a grounded claim from a fabricated one by looking at the prose. In a domain where an unsupported dosing claim can cause real harm, "sounds right" is not good enough. Litmuz sits between the agent and the reader as a verification layer: it re-derives, claim by claim, whether what the agent wrote is actually supported by the literatu