How Guaranteed Correct Generates Predictions
No black-box secrets. Our probabilities come strictly from a deterministic Expected Goals (xG) Poisson statistical engine. The LLM's only job is writing readable tactical previews.
Data Ingestion & Team Stats
Nightly background jobs fetch fixtures and recent scoring form — goals scored and conceded per game for both sides, split home and away. Those figures are the model's entire input.
Head-to-head notes and team news are shown only for fixtures where we hold them; our fixture feed does not supply injury data, so they are added by hand and are often absent. When present they inform the written preview only — never the probabilities.
Poisson Expected Goals (xG) Engine
Expresses each side's attack and defence relative to its league's scoring baseline, giving expected goals for home and away. A Poisson score matrix over 0–10 goals per side, with the Dixon-Coles correction applied to the low-scoring cells, converts those into win/draw/loss, GG and Over 2.5 probabilities.
Gemini LLM Preview Generation
The already-computed probabilities, plus whatever team news we hold, are passed to Gemini to write a two-paragraph tactical preview. The model writes prose only — it never invents or alters a probability, and where team news is missing it is told so rather than left to guess.
Top 10 & AI Top 5 Ranking
Top 10 lists per market are ranked by confidence. The flagship "AI's Top 5" selects the top 5 highest-conviction cross-market picks across all fixtures.
Core Architectural Non-Negotiables
- ✓ Probabilities before Prose: Numbers are generated by math before any LLM prompt is executed.
- ✓ Source Traceability: AI picks carry an electric blue AI badge (`#38BDF8`). Expert picks carry an amber Expert badge (`#F5A623`) with analyst bio. Sources are never merged into an unlabeled list.
- ✓ Ad-Free Subscribers: Active paid accounts experience zero advertisements across every view.