Publisher and purpose
Server Khalilov operates Exact Predictors. Legal publisher and contact information is available in the Impressum. The site exists to make football forecasts inspectable: visitors can see the model inputs and probabilities, read the method, and compare stored forecasts with settled results.
The service publishes estimates for information and entertainment. It does not guarantee an outcome and does not present a forecast as professional or financial advice.
What is calculated automatically
Expected goals, score probabilities, outcome probabilities, adjustments, confidence, and result evaluation are produced by code from structured inputs. Forecast sweeps refresh fixtures when their inputs change. The system validates numeric ranges, identifiers, season provenance, and required fields before publication.
Individual calculations are not manually selected to make the record look stronger. Failed or incomplete inputs can result in no forecast, and published results retain the pre-kickoff record rather than replacing it with a later calculation.
How language models are used
Some match pages include an AI-generated scenario. The language model receives structured match statistics and model output and must return a fixed schema. Automated validation checks structure and consistency; scenarios are not manually edited or approved one by one before they appear. The AI can choose a different plausible score, but it uses the same underlying match evidence and is not an independent source.
Site methodology, policies, and legal pages are curated publisher content rather than automatically created match scenarios.
Updates and material changes
Fixture data and forecasts can update automatically when new results, form, or availability inputs arrive. Material calculation changes are reviewed and documented internally so historical evaluation remains interpretable. Refactors and presentation changes are kept separate from changes that alter forecast output.
Corrections
To report an incorrect result, team mapping, source attribution, model explanation, or legal statement, email contact@exactpredictors.eu. Include the page URL, the incorrect statement or record, the expected correction, and a supporting source when possible.
Reproducible source-mapping and result errors are corrected in the underlying data or code so every affected surface receives the same fix. Material methodology or policy corrections are updated on the relevant public page. Historical forecasts are not rewritten merely because their predicted outcome was wrong.