A literature review you can hand to a reviewer, because it shows exactly how it was made.
Verified Literature Review screens every candidate paper in your corpus against your review protocol, deep-reads the ones that make the cut from full text, and hands back a receipt (how many were screened, which were included or excluded and why). Not a confident paragraph you have to defend on faith, a review whose method is visible.
Get a verified review of your literature See a sample receipt
THE PROBLEM
A literature review that goes into a regulatory submission, an HTA dossier, or a grant is a decision artifact, and right now the fast way to make one is the untrustworthy way. Paste your question into an AI research tool and it hands you a fluent summary of findings, some of which the papers never made. In 2025 people logged over 146,000 hallucinated citations, and roughly half of retrieval-augmented answers turned out to be fabricated. The slow way (screen every paper by hand, record every inclusion and exclusion) is defensible but it takes weeks. So you are stuck choosing between a review that is fast and indefensible, or one that is defensible and too slow to matter.
HOW IT WORKS
We run the same two-pass discipline a good systematic reviewer uses, and we keep the count at every step. First a deterministic fetch pulls the candidate papers, then every one is triaged against the protocol you define (your inclusion and exclusion criteria, your scope, your threshold). Only the papers that clear the bar get the expensive second pass, a full-text deep read that extracts the actual findings, not a guess from the abstract. Everything is deduplicated against a seen-ledger. At the end the run reconciles its own counts (screened, included, excluded), and that reconciliation is the receipt, so a missing corpus or a broken fetch shows up as a gap you can see, never as a clean review that quietly dropped half its inputs.
WHAT YOU GET
One audited review artifact: - the included set, each paper scored against your protocol and deep-read from full text, with the findings that earned its place, - the excluded set, each with a reason, so a reviewer can see what was screened out and why (the part every honest systematic review has to show), - the receipt, a reconciled count of how many papers were screened, included, and excluded (the screened / included / excluded trail that makes the method auditable), - the deep-read analysis for the included papers, grounded in the real text.
WHO IT'S FOR
Anyone who has to produce a review whose method survives a second reader. Regulatory-affairs teams in med-device and pharma, HTA and systematic-review groups, grant and policy researchers, and technical R&D leads who need current evidence they can defend. If you have ever been asked "how did you decide which papers to include", and you needed a real answer, this is for you.
PRICING (the ladder)
- Free / dogfood: the verified review we run on our own research corpus, keyless beyond the model key. - Per-review: one systematic review over your corpus, with the screened / included / excluded receipt. - Team: recurring reviews on a living protocol, a review that re-runs as new papers land and shows you the diff. - Verified / platform tier: the span-grounded citation-audit upgrade, where every included claim is traced back to its source span with a receipt. This rides our deep-read and external-walk verifier, the same verification discipline that is the core of what we build, and it is licensable into a regulated buyer's evidence stack.
IP is licensed, never assigned. The engine stays ours, the review is yours.
THE PROOF (dogfood)
We run this on ourselves. The first verified review the company ships is a review of the literature the company depends on, our own current focus (multi-agent orchestration, verification, de-identification, context compression, text-to-SQL grounding, LLM security). We screen the papers our own radar has already triaged, deep-read the included set, and produce the same screened / included / excluded receipt we would hand you. We are not selling a review method we do not use.
HONEST NOTE
Straight facts about where this is, because a product about honesty should be honest about itself. The review engine is built and it passes our quality gate, and it reuses capabilities we already had (the deterministic feed fetch, the triage and deep-read spans, and the count reconciliation that becomes the receipt). We did not rebuild it for the launch, we reframed what was already running. Two things are on the ladder, not in the box yet. The knowledge graph (mapping the concepts and relations across the included papers) is a follow-on, not part of the shipped review, so we do not claim it as delivered. And the receipt today is the reconciled screened / included / excluded count, which is real and audit-shaped, not a rendered PRISMA-2020 flow diagram, that formatted artifact comes next. The span-grounded citation receipt (every claim traced to its exact source) is the verified tier on the ladder, not the base review. What is also true, plainly, is that a long public record of live audited runs needs a running mesh, a funded model key, and full-text access wired up for an outside user, and those are switches we have not thrown yet. We would rather tell you that than let the first review pretend to be the audited one.
Get a verified review of your literature
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*