How our predictions work

MetricKick is a transparent, model-driven engine — not tipster opinion. Here is exactly how a probability becomes a published value pick, end to end.

Team strength (Elo)

Every team carries an Elo rating updated after each result, adjusted for home advantage and margin of victory. Elo gives a stable baseline for who is stronger, independent of the betting market.

Goal model (Poisson + xG)

Team goal-rates — sharpened by expected goals (xG) as the sample grows and dampened for confirmed absences — feed an independent-Poisson scoreline matrix. From that single matrix we derive 1X2, Over/Under, BTTS, correct score, double chance, clean sheet and win-to-nil.

De-vig the market

Bookmaker odds include a margin (the "vig"). We strip it out across books to recover the market’s true implied probability — the honest benchmark every model number is measured against.

Blend & calibrate

The goal model only earns independence from the market once it has provably out-predicted the closing line on a real sample. Probabilities are then calibrated against graded history so a stated 60% really means 60%.

Value & confidence

Expected value (EV) = model probability × best available odds − 1. We publish a market only when the model probability beats the no-vig price, and score each pick’s confidence from model agreement, market consensus and lineup certainty.

Grade & prove it

Every published pick is graded against the real result and the closing line (CLV). The full, dated performance record is public — no cherry-picking.

See the receipts

We publish the model’s graded record — win rate, ROI and closing-line value — across every market, updated continuously.

View model performance

Predictions are informational and never guaranteed. Positive EV means the model probability exceeds the market price — not a certain outcome. 18+ · Bet responsibly.