How we assess form, availability and prediction strength
Prognostix editorial team ยท Published
We use research assisted by latest-generation AI models and bookmaker data to prepare editorial assessments. Experimental numerical estimates supplement these assessments in the Extras. The two processes are distinct: the main 1X2 pick and exact score are not outputs of the Poisson calculation used for count ranges.
Identify the right fixture before analysing it
We check the requested dates, season, competition, venue and kickoff time. Online research helps locate information, while official calendars and announcements provide reference points for confirmation. A link between two teams can resolve to a previous meeting, so matching the names alone is not enough.
Recent form needs context
For clubs, league form is distinguished from cup matches; for national teams, we identify which matches have been verified. Results against different opponents are not a perfect measure of strength. When fewer than five results are available, we disclose that limitation rather than inventing entries to complete a sequence.
Injuries, squad selection and rotation are different questions
A confirmed injury, a suspension and an availability doubt do not mean the same thing. A squad call-up does not guarantee a starting place. We assess available reports and flag provisional information; an empty absence list does not establish a fully fit team. Assessments may change as kickoff approaches.
Odds and 1X2 strength are not model win probabilities
Odds are market prices that can change. Low, Moderate and Strong summarise editorial support for the selected 1X2 outcome only. Strong needs converging evidence: a short price or agreement between two AI systems does not automatically earn that label. Our scale is not calibrated against results and does not establish a betting edge.
Specific sources and visible limitations
We retain consulted page URLs, sample periods and research times internally. Reading a price now does not establish when the bookmaker last updated it. The match page shows the observed reference price and bookmaker with โMarket snapshot, not live oddsโ. This is not a live feed or proof of market consensus. If a source cannot be accessed, we record the limitation rather than claim a check that did not happen.
Which forecasts use Poisson
Corners, shots on target, yellow cards, offsides and goals in each half use regularized historical means and experimental Poisson ranges. Full-match goal ranges derive from the two half-time goal means. These are model quartiles, not validated empirical coverage: they assume a simple count distribution and do not incorporate every tactical dependency. Scorers remain conditional editorial candidates; double chance covers the main pick. Extras do not inherit 1X2 strength.
Reviewing the record after matches
We keep dated publication snapshots. Once results are verified, 1X2, exact scores, Over/Under and BTTS can be evaluated separately; postponed and unfinished matches remain pending. Revisions have separate versions, rather than selecting the best forecast after the result. We will publish performance reviews here only when verified results are available. Snapshots also retain Extras, head-to-head records and scorer candidates with separate hashes. The current main-market audit does not automatically assess Extras: actual counts and a dedicated evaluation are needed before publishing their performance.
External links are provided for information. Prognostix is not affiliated with the linked websites or brands; mentioning them does not imply a partnership or endorsement. External websites have their own terms and privacy policies.