It sits small beneath the text and promises: the source stands behind the sentence. An AI assistant that comes up empty signs it with a reference that does not exist. NotaVera collects on the promise: the real passage beside your sentence, judged blind, no author, no title, no year. The literature cites on credit.
Sources from OpenAlex, resolved through Crossref. No model invents a locator here.You wrote a hundred pages. They need one passage that does not hold. One. After that the rest stops counting.
In a dossier every statement carries a reference. The agency reads every one of them, and not with goodwill.
A number without backing does not survive the first challenge. That challenge is rarely private.
Product liability, environment, medical devices. Where the study is the evidence, its reach decides the case.
Before the committee it does not count how much you read, only whether every footnote holds. It reads the one that does not.
A claim in print is not withdrawn. The fact-check arrives after publication, and it arrives in public.
One paragraph. NotaVera immediately sees what most authors miss.
Supported, contested, boundary case, or open. Most authors learn the difference from the review report. You know it beforehand.
The source must carry the sentence. Existing is not enough. No match means: you know it now, your examiner finds out later.
Counter-evidence stands beside the sentence. You can work with it or ignore it. The reviewer finds it either way. The question is who had it first.
NotaVera finds the source. Turnitin finds you.
Read months ago, half forgotten, now in the text. Plagiarism software judges after submission. By then it's too late. NotaVera shows the passage as you write. Everyone else submits and prays.
See one complete example, with the located passage →A percentage is a suspicion. A passage is a work order: source, excerpt, revision.
Citation markers, work identity, page context, and source quality are assessed together. Clean citation needs no defence.
NotaVera shows every finding to you first, in private, with no verdict attached. Nothing is rewritten without you.
Two real claims, two real sources, both open to inspection. No sample text, no invented DOI. What NotaVera shows here is the finding the editor produces.
A statistically significant result is necessarily scientifically important.
contestedThe source does not carry the word “necessarily”. A significant result may still be trivial, biased, or substantively unimportant.
What this does not prove: This static demonstration shows one bounded source–claim comparison. It is not a review of a manuscript and does not guarantee truth, originality, or publication readiness.
Reviewed 2026-07-15 · v1.0.0
Nonsense used to at least sound like nonsense. Now it reads like peer review. A generative model is trained to agree with you; the expert on the other side is trained to refute you. NotaVera plays the second role while you can still change something.
Your client reads with goodwill, your team reads along. The other side is searching, and it is paid to.
NotaVera does not believe you either. But NotaVera works for you.
You write the sentence; NotaVera looks for the source that actually supports it. No keyword search, no scrolling through results that turn out to be about something else.
Every DOI is resolved against Crossref before it is called verified. What cannot be resolved stays marked as a candidate instead of quietly passing as evidence.
Citation style is a setting, not a rewrite. The same manuscript exports to Markdown, DOCX and PDF with the footnotes already in shape.
After a check, NotaVera can turn the finding into an Evidence Receipt: one page carrying the sentence that was checked, the sources separated by role, and a signature. You decide whether it becomes public, and you can revoke the link at any time.
The page names the checked sentence, every source with its DOI, and whether that source supports, contradicts or merely borders the claim. Nothing has to be taken on trust.
Publication is a separate, explicit step, an expiry date is optional, and one click revokes the link. Nothing is published by writing.
The visible excerpt must occur in the sentence that was actually examined, and that binding is part of what the signature protects.
It records evidence fit for a scoped claim. It is not a truth, plagiarism, or academic-approval certificate.
Your client reads with goodwill. Your team reads along. The other side reads to find something. NotaVera reads like the other side, only earlier.
The smoothest sentences are often the emptiest. NotaVera detects claims that need evidence as you write; otherwise it shows what is missing.
You can ignore it. The other side cannot. It stands in the margin until you face it.
When a sentence claims more than the study supports, NotaVera proposes a version that holds.
It proves something exists. Not that it carries your sentence. What fails that gets a warning instead of a footnote.
NotaVera distinguishes the dominant position, credible minority evidence, and speculation. A majority is not a mechanism.
Text in the center, footnotes below, evidence in the margin. Whoever hides where things stand has a reason.
You do not. Whoever is paid for it reads slower.
Write about dry fasting and you cannot get around Russian research. Most tools get around it and call that the state of the field. NotaVera ranks by method. Whether a country of origin is welcome is not a scientific category.
Important thoughts rarely announce themselves. What looks like a fragment today can be the core of a thesis years later. Or it disappears into a folder nobody opens again.
Thoughts sit by kinship, not by date. What belongs together ends up side by side, even with years in between.
The library shows your work as a sphere or as terrain. Chapters, claims, footnotes and sources are points of their own; fresh work glows, cold corners call you back.
On a click, the space replays how your thinking grew: which thought arrived when, and what became of the fragments.
Obsidian stores what you wrote. NotaVera shows you which of it carries evidence.
Full editor, counter-evidence in the margin, Evidence Weave, DOCX and PDF typesetting. Plus 560 processing units per calendar month. You are billed for real work, not for clicks: a paragraph costs a fraction of a unit, a chapter costs several.
Checking first costs nothing: 42 units a month, no card. Take a sentence whose answer you already know. There is no other way to judge a checking tool.
That the source exists and is peer-reviewed was never the hard question. The hard one is whether it carries your sentence at exactly that reach. NotaVera marks the places where the claim goes further than the finding: a “necessarily”, an “all”, an effect transferred to a different population. That is where an opposing expert starts, and it almost never shows up in the bibliography.
NotaVera searches OpenAlex and resolves through Crossref. That is scholarly literature. Case law, technical standards, agency files and your internal data are not there, and NotaVera will not claim otherwise. For sources you supply yourself it still checks reach, boundary cases and counter-evidence.
Zero Data Retention applies at the models: no training, no storage at the provider. Built and operated in the EU. You can export everything at any time, including after cancellation, and a plan change never removes access to what you have already written.
That is what the Evidence Receipt is for: one page carrying the checked sentence, every source with its DOI and its role — supports, contradicts, or draws the boundary — plus a signature. You decide whether it becomes public, and you can revoke the link. It records evidence fit for a bounded claim and is explicitly not a certificate of truth, originality, or academic approval.
Then the error is visible, with its source and locator beside it, and you refute it in a minute. Every finding names what it rests on. A tool whose statements cannot be recomputed would be the exact problem this one was built against.
No. NotaVera checks what is in OpenAlex and Crossref, and it checks the fit between your sentence and the source. It proves no claim, it guarantees no completeness, and at no point does it take the professional judgement off your hands. Anyone promising you that is selling you the same complaisance this tool was built against. What you get is a finding with a locator that you and a third party can recompute. One guarantee we do give: the other side will try.
What the database has not heard of is called unstudied two pages later. A limit of reach becomes a statement of fact, and nobody lied.
→July 16, 2026AI PsychosisThe machine does not hallucinate confidence out of nothing. It inherits the user's, amplifies it, and hands it back dressed as an independent verdict.
→15. Juli 2026Die Illusion des »Company Brains«Mehr Daten erzeugen kein Wissen. Entscheidend ist die nachvollziehbare Kette von Quelle, Begrenzung und menschlicher Entscheidung.
→July 14, 2026Geography Is Not CryptographyA German postcode is not a security architecture. When a vendor sells data residency as data sovereignty, it is offering reassurance in place of control.
→The opposing expert is no better than you. They just have more time.
Check a document