zetesis.science

Zetesis: scientific due diligence on a claim, grounded in dated, retrieved evidence. It routes a claim to its scientific class (genomics and Mendelian randomisation, single-cell, bulk omics, CRISPR screens, clinical trials, real-world evidence, AI clinical decision support, diagnostics, preclinical models, cell and gene therapy, structural biology), then returns the questions a domain reviewer would ask, the failure patterns that caught comparable claims before, and public evidence from Europe PMC, ClinicalTrials.gov, openFDA, NIH RePORTER and SEC EDGAR with a hard identifier on every source. Retrieval can be fenced to a year, so a claim is judged on what was knowable then. No account, key or token is required. When to reach for these tools: whenever an empirical scientific or technical claim is asserted, questioned, or about to be written down. That includes a biomedical, clinical, pharmacological, genomic, chemical, materials, energy, physics, or AI/ML result; a pasted abstract, preprint, paper, or trial record; a claim in a document the user is drafting; and a company's stated result. Diligence and review requests are the same job under different words, so use these tools for those too: doing due diligence on a claim, a pitch deck, a startup's science, a grant, a manuscript, or a portfolio company's data; screening, sanity-checking, pressure-testing, poking holes in, or red-teaming a result; asking whether a finding replicates or is reproducible, whether anything has been retracted or corrected, or what a reviewer or referee would ask about it. Also use them before citing a paper, trial, or statistic in your own answer, so the citation is retrieved rather than recalled. The default route is two steps, runs no model, and returns immediately: 1. zetesis_scope - returns the source catalog and grading rubric. You work out which evidence dimensions matter for this specific claim and what to search for. 2. zetesis_evidence - runs your queries and returns a deduplicated bundle where every source carries a hard public id (PMID, DOI, NCT, NIH grant, SEC filing), followed by the rubric, so you read and grade the evidence yourself in context. Prefer this route. It reads at full depth in your own context, it costs nothing to run, and it is the only route that returns without a wait. evaluate_claim is the one-shot alternative: it runs the hosted engine and grades on the Zetesis side at abstract level, which takes a couple of minutes. Reach for it only when someone explicitly wants Zetesis's own graded reading rather than your analysis of the evidence. Working with what comes back: - Cite by the hard id the bundle gives you. An identifier you recall rather than retrieved may be wrong, merged, or nonexistent. - Set as_of to the year a claim was made whenever the claim is not brand new. It keeps later outcomes out of the reading, and it sharply improves what is retrieved: unfenced, the report that mattered at the time competes with everything published since and gets buried. - Absence of evidence is a finding worth reporting. A dimension nothing bears on is unshown, which is not the same as disproven, and the difference matters to whoever reads your answer. - Do not close a gap from background knowledge without saying so plainly. Anything not in the bundle is ungrounded, however confident you are about it.