On Tuesday 7 October 2025, the Swedish Embassy in France hosted a seminar on Glimt, the crowd-forecasting platform that FOI, the Swedish Defence Research Agency, runs for Ukraine. The invitation billed it as an example of Sweden–France cooperation. Ivar Ekman of FOI’s Defence Analysis Department would explain how a Swedish innovation supports Ukraine’s strategic decisions, and Hypermind, the French company that supplies the technology, would show how the system works behind the scenes.
Three days later, Émile Servan-Schreiber, Hypermind’s CEO, posted the deck he and Ekman had presented, titled “Once upon a Glimt…”. He thanked Colonel Malin Persson for organising the event. He also credited the team: Jonas Clausen Mork, Ida Selbing and Simon Hellman at FOI, and Maurice Balick and Florian Jacques at Hypermind. His post framed Glimt as a way for citizens in Europe and elsewhere to contribute directly to Ukraine’s defence.
What the deck says
The slides describe FOI as a research agency under Sweden’s Ministry of Defence, with about 1,600 employees, roughly a third of them PhDs. According to the deck, Glimt was part of Sweden’s 16th support package to Ukraine in June 2024, with a mission to give Ukraine methodological support in forecasting. The platform went live on 20 January 2025.
One slide explains why crowds help. Forecasting is two steps: gathering information, then guessing. An individual is limited by time and resources at the first step, and by cognitive biases and personal noise at the second. A crowd gathers more, and in more depth, like Wikipedia. Its guesses also improve because individual errors cancel out while information adds up. This is the “wisdom of crowds”.
- About 20,000 registered forecasters and about 300 forecasts a day.
- Three languages, forecast questions supplied by Ukraine, and a large communications campaign.
- Consolidated forecasts delivered to the Ukrainian government.
- In the pipeline: group-versus-group forecasting and hybrid forecasting that combines AI with human forecasters.
Interim accuracy: 71% and well calibrated
The results slide reports that 71% of outcomes were correctly forecast across 31 resolved questions. Questions about the war were forecast best. Political questions were the hardest.
A second chart tests calibration. A well-calibrated forecaster’s 70% predictions come true about 70% of the time, its 20% predictions about 20% of the time, and so on. The deck plots the crowd’s forecast probabilities against how often those outcomes actually happened, across 31 questions and 116 possible outcomes. It reports a 98% correlation between the two.
Le Figaro: “an initial operational level”
On 22 October, Le Figaro paired Glimt with Sweden’s offer of Gripen fighter jets to Ukraine as two forms of Swedish support. Nicolas Barotte wrote that the platform had just reached an initial operational level. He reported that its questions come from Ukrainian intelligence, that each attracts hundreds of anonymous participants, and that FOI had recruited with a campaign in the Stockholm metro.
The article quoted several live forecasts. About 90% of participants expected no Zelensky–Putin meeting in 2025, and about 90% expected no robust US security guarantees for Ukraine before the end of the year. A narrower 56% expected no new US sanctions. Servan-Schreiber gave the paper a breakdown by topic: “87% of forecasts on the war are correct, 70% on the economy, 44% on politics” (our translation). He described the best forecasters as rather modest people with analytical minds, who are able to change their opinions. Ekman’s closing line was simpler: the future will remain uncertain.
France 24, and a missing French connection
France 24 followed on 23 October, with an English version the next day. Grégoire Sauvage linked Glimt to the US intelligence community’s ACE forecasting tournaments of 2011–2015, which involved some 10,000 forecasters. He reported about 20,000 registered users and around 500 forecasters per question, with questions written by Ukrainian intelligence services. At the time of writing, the crowd gave a Zelensky–Putin meeting in 2025 about a 1-in-10 chance. It gave 25% to Tomahawk missiles being delivered to Ukraine before 1 February 2026, and 91% to hostilities not ending before 2026. Ekman told the channel that questions usually run from a few weeks to a year, so the data is meant to be useful to Ukraine day to day. He added that FOI does not know how the Ukrainians make their decisions, which is a sensitive part of their government’s work.
Servan-Schreiber welcomed the piece but regretted that it did not mention the platform’s technology came from a French company. The Swedish Embassy’s repost made the point itself, saying Hypermind’s French expertise is behind Glimt. That fits the theme of the 7 October evening, and of the platform’s French-language version.
