How we measure VIDI.
Every number on this site about EE Partner's VIDI comes from tests on decisions already closed, with the real outcome. Here is where each one comes from.
Top 50 in the world at forecasting
The right choice on European public tenders.
VIDI against the AI model Claude Opus 5.5, used on its own, on the same decisions.
On European public tenders, VIDI by EE Partner points to the right choice 79.9% of the time. Claude Opus 5.5, used on its own, 66.4% of the time.
- The decisions
- 1,605, all arriving from July 2026, after Opus finished training: no one could have known the outcome.
- Which ones
- 22 types of decision on European tenders, chosen in a first test where VIDI did better, then measured again on new decisions.
- The check
- On every tender in the first test, not just the chosen types: VIDI 79.7%, Claude Opus 5.5 71.0%, on 2,187 decisions.
- The comparison
- Claude Opus 5.5 is the more powerful of the two Anthropic models we tested. It received the same question, without VIDI's data and calculations. VIDI works with Claude too.
- The source
- EE Partner internal tests, 20 October 2026.
How far to trust an answer.
Every VIDI answer says how likely it is. We counted how often those probabilities came true.
95%
When VIDI by EE Partner is 90% sure or more, it's right 95 times out of 100.
Of 14,498 decisions already closed, mostly on European public tenders, this happens in 3,535: 3,364 of those answers are right (95.2%). EE Partner internal tests, 20 October 2026.
Band by band: what VIDI says and what happens
- 90% or more3,535 decisions95.2%, VIDI says on average 95.5%
- 80 to 90%2,766 decisions87.4%, VIDI says on average 85.2%
- 70 to 80%2,451 decisions76.3%, VIDI says on average 75.1%
- 60 to 70%2,429 decisions64.5%, VIDI says on average 65.0%
- under 60%3,317 decisions51.9%, VIDI says on average 52.3%
The bar almost always reaches the mark: VIDI's percentages mean what they say. Same 14,498 decisions, scale from 0 to 100.
ForecastBench
Top 50 in the world at forecasting.
ForecastBench is the world ranking of forecasting systems. We measured VIDI with its scoring, in a simulation on questions from rounds already closed: 2.6 points behind first place.
- 1stGoogle DeepMindFirst in the ranking69.1
- 42ndVIDIWith the data series from the 7 June 2026 round66.5
The higher the index, the better the forecasts. The top 50 holds with the series from the 7 June round.
EE Partner simulation with the ForecastBench scoring, report of 11 October 2026, against the ranking of 3 October 2026 (377 systems).
The data VIDI reads.
Public registers and archives, always cited with their code and date in the documents you receive.
3 billion
Over 3 billion data points read from public registers.
- What data
- Companies, financial statements, tenders, state aid and public projects.
- How we count
- Every value VIDI reads, with no empty or technical fields. A register counts each of its dated versions: that is how we see what changes over time.
- Among the sources
- OpenCoesione projects, the National State Aid Register, ANAC and TED tenders, Eurostat, and the company registers of the United Kingdom, France, Norway, Latvia and Estonia.
- The source
- EE Partner count, 21 October 2026.
The rules, set before the results.
They apply to every number on this page.
Decisions already closed
Every test uses past decisions, rebuilt from public archives, with the real outcome. None comes from a client.
Rules written first
Questions, comparisons and measures are set before the results are known. Each test is looked at only once.
No glimpse of the future
On past decisions, VIDI uses nothing that would reveal the outcome, such as today's web search.
Negative results stay
What doesn't work stays on record, and a number that changes carries its date.
Let's talk.
Want to know how far you can trust the answers to your questions? Tell us what you need to decide.
With VIDI, your first question is free.