We reconstruct the origin, document the chain of outlets that carried it, and attach a significance test to every figure. Built to be handed to a board, a regulator or opposing counsel.
From €2,500 per case. No platform access required.
Scope
Newjee documents what happened. Where a claim first appeared, which outlets carried it and when, and whether the framing differences between them are real.
We do not remove content, and we do not detect bot networks, fake accounts, deepfakes or coordinated inauthentic behaviour. Those are different products built on different evidence, and they are not on our roadmap.
Why monitoring is not enough
Each outlet cites the one before it. A list of links is not an origin.
A measurable question about how the framing moved, and which outlets moved it.
A number with no test attached will not survive being asked how it was calculated.
Why it holds
A permutation p-value, a confidence interval, and the variation inside the blocs themselves. When a gap falls inside the ordinary spread, the document records no finding.
The common alternative
A percentage change against last week, with no test attached.
Where the coverage does not support a figure, nothing is returned and the reason is recorded. Dropped sources are counted and listed, so the denominator stays visible.
The common alternative
A score is produced regardless, with no visible denominator.
Two measures on two tracks. A credible outlet can carry an unverified claim. Merging the two is the hardest position to hold once counsel has noticed.
A formal review channel, open to people who are not our clients, including an outlet named in your own dossier. A press officer is one of the roles it accepts.
Before you commission anything
You will be asked to rely on this method in front of people who will test it.
Bloc-Conditional Event States, a KVAi Lab preprint, sets out the density-matrix representation, trace distance, and three inferential layers. Two worked case studies: Hormuz 2026 across 16 outlets, Navalny 2024 across 14.
Read itA Multi-Layer AI Framework for Information Disorder, at the LREC 2026 Workshop on Information Disorder. Beyond Sentiment, at the LREC 2026 Political NLP workshop.
Venues and abstractsFormulas, weights and decay constants, with the protocol for breaking changes. Applying the published method to your own corpus should reach comparable results.
MethodologyTo be supplied
One completed reconstruction with the client and outlet names removed, showing the chain established and the figures reported.
Requires a completed engagement and written client consent. Nothing to show until the first dossier is delivered.
To be supplied
The named analysts who would reconstruct your case, with their background.
Requires sign-off on which names and credentials to publish. Credentials are not ours to assert on somebody else’s behalf.
How it works
The claim, the coverage, the window, and who you answer to. A named executive works as well as a company.
What we expect to establish and what we do not, with the turnaround and the fee. If it cannot be done with our methods, we say so here rather than on delivery.
Origin and chain first, then framing divergence with its significance test, then propagation. Thirteen languages, including Arabic, Persian, Chinese and Russian. Audio and video from a URL are transcribed.
With a walkthrough of the method for your counsel or your board, if that is useful.
Turnaround
Ten to fifteen working days
Agreed in writing before we start. Tell us if your deadline is shorter and we will say at once whether it is achievable.
What moves it
Questions
No. Takedowns, legal notices, platform escalation and litigation are all outside what we do. We produce the document that the people doing that work need. In Italy, Reputation Manager is the most established firm for it, and they will often ask you for exactly this kind of reconstruction.
No, and not coordinated inauthentic behaviour either. Those methods examine account behaviour, network structure and forensic signals inside a media file. We examine published text, which outlet published it, and how the framing moved. For that work there is Cyabra and Blackbird.AI, and IdentifAI on synthetic media. Some engagements need both.
They count mentions and classify tone, across more sources than we cover. If you need to know how much you are being discussed, they are the right tool and they cost less. We answer where a claim first appeared, and whether a difference between outlets is real or noise.
We record it and do not produce the number. Sources excluded from a calculation are counted and listed with the grounds. Spread on closed channels is not observable to us, so we do not estimate it. A document with stated gaps can be defended. One with hidden gaps cannot.
From licensed editorial data suppliers and from open sources. Those licences prohibit redistributing article text, so the dossier carries analysis, metrics, timestamps and links. Our own analysis is separate: the right to republish it commercially, in a market communication or a defence filing, is part of the institutional contract and is agreed in writing.
Neither yet. The platform measures how framing has moved between two points in time and tests it at 2 sigma. Per-organisation watchlists, alerting and a composite risk indicator are in development. When the indicator ships it will be presented as a measure of risk, never as a judgement on whether a claim is well founded.
Get in touch
A person reads this. We reply within two business days, and if the case is not one for us we say so at once.
