In the monthly NARINT issues we assign events a number from 0 to 10 and call it the Narrative Divergence Index. This text explains what stands behind the number, how it is calculated and, just as importantly, what it does not mean.
The problem we measure
Imagine five people watching the same film. All agree on what happened on screen: who went where, who said what. But one leaves the cinema saying “this is a film about betrayal”, another is certain he watched a love story, and a third saw political satire. The facts are shared; the meaning belongs to each viewer.
The same happens with news, except the viewers here are not random individuals but entire information environments: Belarusian state media, independent Belarusian outlets, the media of Lithuania, Poland and Ukraine, the state agencies of Russia and China. They describe the same event, say, a prisoner exchange or military exercises near the border, with the same basic facts but with meanings so different that their audiences end up living in different versions of reality.
Usually people describe this in words: “propaganda spins it differently”, “the independents write it another way”. True, but imprecise. How differently? On which exact question do the versions split? Has the gap grown since last month? To answer such questions you need a ruler. NDI is that ruler.
How it is calculated: four steps
Step one: separate facts from interpretations. For every event we first establish what all environments agree on. In June’s case of Lithuanian blogger Kandrotas nobody disputed the facts: he left for Belarus, a Lithuanian appellate court sentenced him to 3.5 years, the police are preparing an international search. That is the foundation. The divergence begins above it.
Step two: find the axis of the dispute. Every event has a central question on which the versions split. For the Kandrotas case it is the question “who is he”: a criminal who fled his sentence and works for propaganda, or an honest activist persecuted for his views. We call this the axis of divergence. One end is marked 0, the other 10. Importantly, each event has its own axis. For the story of the German brigade in Lithuania the axis was “is this a successful strengthening of defence or a failure exposing NATO’s weakness”. For the prisoner exchange, “whose homecoming counts as the event’s main content”.
Step three: place the environments on the axis. Next we look at what each environment actually wrote and determine its place on the scale. Lithuanian media in the Kandrotas case stand at mark 1: a convicted violator, part of an influence operation. BELTA stands at 9: an honest entrepreneur hounded for the truth. Sputnik at 10: a victim of political repression. The position reflects not the outlet’s general political colouring but its interpretation of this specific event.
Step four: measure the spread. An event’s NDI equals the distance between the leftmost and the rightmost position. In the Kandrotas case the extremes were taken by Lithuanian media (1) and Sputnik (10), so the NDI is 9. Had all environments crowded around the middle of the scale, say between 4 and 6, the index would be 2: the event is described similarly, with no serious divergence.
How to read the values
NDI below 4: working disagreements. The environments place their accents differently, but their readers, if they met, could discuss the event and understand each other. This is what coverage of most protocol and technical news looks like.
NDI from 4 to 6: notable divergence. The versions are already arguing with each other. Usually this means some environments are pushing a favourable frame while others contest or ignore it. June’s example: the return of Belarusian athletes to international competition, which state media called a full restoration of rights while independent outlets pointed to the neutral status and the reason for the sanctions.
NDI 7 and above: incompatible realities. A BELTA reader and a reader of a Lithuanian portal, discussing the Kandrotas case, will be talking about two different people. Shared facts are no longer enough to agree on meaning: the very question “what really happened” receives opposite answers. These are the events we place at the top of every issue.
What NDI does not mean
NDI does not say who is right. The index measures the distance between versions, not their truthfulness. A high NDI does not by itself mean one side is lying: in theory two honest interpretations can also diverge. In practice, in the Belarusian information space a high NDI most often appears where state media build a frame incompatible with verifiable context, but that is a conclusion of fact-checking, not of the index itself. NDI shows where to look; fact-checking answers what was found there.
A position on the axis is not a rating of the outlet. A mark of 9 or 10 does not mean “a bad media outlet” but a specific place in a specific dispute. In some events Russian state agencies end up at the edge of the scale, in others, protocol stories for example, in the middle, alongside everyone else.
A low NDI is not a guarantee of truth. If an event was covered only by environments with similar positions, the index will be low regardless of the quality of coverage. That is why we calculate NDI only for events covered by at least four different environments, and separately track situations where an environment stays silent: silence is a signal too, but we have another instrument for it, the silence map.
Who does all this: the machine and the human
Our sources publish more than 25,000 items a month in five languages. No human can read them all, so the primary work is done by the FORESIGHT platform: it aligns texts into a common language field, automatically groups publications into events and finds those covered by several environments at once. A language model then reads the material of all environments for each selected event, formulates the axis and proposes positions. The final check is human: an analyst tests the axis against common sense and verifies contested positions against the original texts.
The method has honest limitations worth knowing about. A position on the scale is the model’s expert judgement, not the result of physical measurement: the difference between 8 and 9 is less reliable than the difference between 2 and 9. If an environment published little about an event, its position is estimated from incomplete data, and we flag such cases. Finally, the axis itself is a simplification: a real dispute is always richer than a single scale. We accept this simplification deliberately, because it delivers the essential thing: the ability to compare events with each other and month with month.
Why this matters to the reader
NDI does not replace reading the news and does not hand out ready-made conclusions. It makes visible what usually goes unnoticed: the very fact that diverging versions have been built around an event, and the precise question on which they diverge. A reader who knows the axis of the dispute is better protected than any reader who has seen only one version, even a truthful one. That is why every NARINT issue begins not with an answer, but with measuring the distance between answers.
This publication was prepared by FactCheck.LT . NARINT (Narrative Intelligence) is a monthly series based on the FORESIGHT platform. Methodology questions: info@factcheck.lt.