How NET produces independent forecasts and identifies betting value — without exposing the proprietary edge.
NET's core forecasting engine is a Poisson / Dixon-Coles score model built on persistent team-strength ratings (Elo) and attack/defence profiles. It estimates the expected goals for each side and generates a full score probability matrix from 0-0 through 6-6.
The critical principle: bookmaker odds are never an input to our probability model. NET computes its own fundamental probabilities first, then compares them against the market to find genuine mispricing. This is the only way to identify value rather than mirror consensus.
We never pretend coverage is equal across leagues. Every forecast carries a grade reflecting data confidence:
A+ / A rich data · B solid team data · C forecast only · D experimental
Premium recommendations are only made from high-confidence, well-covered fixtures. We sell honest probability, never false precision.
For each match we remove the bookmaker overround to estimate market probability, then compare against NET's own probability. A genuine value bet must clear both an edge threshold and an expected-value threshold. Every recommendation publishes the fair odds, the minimum acceptable price, and the expected value — so subscribers can verify whether their available price still represents value.
Every recommendation is archived with its price snapshot, model version and feature set at the moment it was published — fully reproducible. Results settle automatically against verified sources. Losing tips are never deleted, and the full record is public.
NET's ratings update after every settled match. Model changes are tested walk-forward and promoted only when they measurably improve calibration on a meaningful sample — never on a lucky run. Bad changes are rolled back automatically.