The Consensus Model
Prognostix combines AI-assisted analysis and match context to build a reasoned consensus on possible game scenarios. The result is an editorial synthesis of available evidence, rather than a simple vote between models.
How It Works
1.Match context
We consider recent form, opposition quality, previous meetings, availability and possible tactical choices. These factors are assessed together: a winning run or one absence does not automatically determine the pick.
2.Scenario analysis
Latest-generation AI models help compare how the game could develop, including each team’s strengths and vulnerabilities. We also examine arguments against the favoured scenario and distinguish verified information from assumptions.
3.Synthesis and strength
We publish a 1X2 pick and a possible score alongside goal markets. Low, Moderate or Strong describes support for the 1X2 pick only; it is not the probability of every selection on the page succeeding.
4.Publication and review
Dated forecast versions are retained for comparison with verified results. Updates remain separate from previous versions; the original analysis is not rewritten after the match to fit its outcome.
AI Models & Bookmaker Data
Consensus means a coherent synthesis of the evidence supporting a forecast and the arguments challenging it. Latest-generation AI assists the analysis, while bookmaker data provide a market reference. The final assessment remains editorial: agreement between models alone does not establish accuracy.
The match card answers several questions. The 1X2 pick concerns a winner or draw; the exact score describes one scenario. Over/Under and BTTS concern goals. Double chance broadens coverage of the selected outcome, while the goal range describes total match goals. The displayed odds are an observed price snapshot, not a live feed or a price for every market shown.
Numerical Extras use an experimental Poisson model to turn expected means into count ranges. Possible scorers are separate editorial candidates, conditional on starting. These approaches do not form a single calibrated model: each prediction has its own limitations and no selection is presented as certain.
1X2 strength
More evidence converges on the selected 1X2 outcome, with less compelling objections. A different result remains possible: Strong is neither certainty nor a validated success percentage.
The 1X2 pick has reasonable support, but team balance, tactical alternatives or provisional information leave competing scenarios open.
The selected scenario has a limited advantage or substantial uncertainties remain. The prediction is more fragile and needs closer attention to context.
Extra
Extras are experimental count ranges, not recorded results or odds. Clubs use their last five league games and a league baseline; national teams use available data from the current edition. National-team half-goal estimates use a general timing split derived from club leagues, rather than team-specific half-time data. Cards mean yellow cards only; referee effects are not modelled. Ranges come from an unvalidated Poisson model and do not represent win probabilities or 1X2 strength. A dash means insufficient data. Shots on target are predicted on-target attempts per team, not all shots. Possible scorers are editorial candidates based on probable lineups, conditional on starting; they are not calibrated scoring probabilities. Without verified probable participation we show “To be confirmed”. Double chance covers the published 1X2 pick; a draw adds the side with more expected goals (home on a tie). Goal ranges are inclusive, for each team and the match, over 90 minutes plus stoppage time, derived from both half estimates. These experimental selections have no quoted odds or calibrated probabilities and may differ from bookmaker lines. Club offsides are an exception: they use rounded 2026/27 league-season means and the source’s available game count, regularized toward the league reference, not the last five games. This sample may be older than other statistics. Opponent offside-trap effects are not modelled.