Benchmarks

The results behind each suggestion, by bookmaker and league.

Public beta18+ · Virtual bankrolls · Independent research

Independence, method & limitations

We benchmark model probabilities against bookmakers and tune for simulated returns with virtual stakes and bankrolls. No real money is wagered or won. Simulated profits do not promise real returns.

An ongoing public-beta experiment comparing model probabilities with bookmaker-implied probabilities for football statistical outcomes. We tune and evaluate our models to improve simulated returns using virtual stakes and bankrolls. Recorded profit and loss are hypothetical; no real money is wagered or won here.

StatsGem is independent. We are not a bookmaker, gambling operator, betting broker or bookmaker affiliate. We accept no bets or gambling deposits, hold no player funds and offer no cash prizes or withdrawals. Subscriptions pay for research access. Bookmaker names and logos identify data sources; they do not imply a partnership or endorsement.

Simulated performance does not predict real-world profit. Models can be wrong, odds can change, and historical results can overstate an apparent advantage. Gambling can cause financial harm and addiction. Get support

PredictionsPredictions MonitoringMonitoring Player propsProps FantasyFantasy BenchmarksBenchmarks

Historical direction benchmarks

Benchmarks from 15 August 2026 · half-point lines only. Compare normal and reversed one-unit returns for each bookmaker, market and league.

Historical evidence

4Cells with a direction

3Cells still gathering evidence

5Earlier matches required

7Cells in this selection

Bookmaker

League

StatisticOriginal model directionScopeCurrently producingPick stageShow cellsOrder benchmark rows

Reset

Rows with the most historical matches appear first. This only changes the table order; it does not assign Free or VIP.

Normal follows the original model direction. Reversed takes the opposite side. The higher historical profit selects a direction to evaluate, including the smaller loss when both lost. A live pick still needs positive estimated return at its current odds.

1–4 of 4 cellsOne unit per earlier signal · returns rounded to two decimals

Bookmaker / league	Original model signal	Earlier matches	Normal	Reversed	Selected direction	Current picks
KambiCzech First League	Goals · UnderAway team Evidence 7 reconstructed · 0 recorded 1001159633 · scraped	7	+9.52u136.0% ROI	−4.47u-63.9% ROI	Normal → UnderHigher earlier return	No current Free picks
KambiCzech First League	Goals · UnderHome team Evidence 7 reconstructed · 0 recorded 1001159967 · scraped	7	−1.55u-22.1% ROI	−1.00u-14.3% ROI	Reverse → OverHigher earlier return	No current Free picks
KambiCzech First League	Goals · UnderMatch total Evidence 6 reconstructed · 0 recorded 1001159926 · scraped	6	+12.00u200.0% ROI	−3.53u-58.8% ROI	Normal → UnderHigher earlier return	No current Free picks
KambiCzech First League	Goals · OverMatch total Evidence 5 reconstructed · 0 recorded 1001159926 · scraped	5	+0.30u6.0% ROI	−0.68u-13.6% ROI	Normal → OverHigher earlier return	No current Free picks

How to read this evidence

A cell needs 5 distinct earlier matches. Higher return selects normal or reversed, including the smaller loss. This qualifies a direction for monitoring; lineup, contract and current-edge checks follow.

Reconstructed history is labelled separately from genuinely recorded pre-match signals. Published prediction returns have their own record.

Each benchmark row is historical evidence for a bookmaker, league and market. It has no fixed Free or VIP tier. Tiers are assigned to live picks using the estimated return at the available odds. Top 50% of approved live market groups, ranked by estimated return.

Market groups are ranked together at each evaluation. The VIP share is rounded up to a whole group, so a small approved pool may leave no Free groups. Monitoring alerts use a provisional ranking that can change at confirmation.

Release 20261019-live-v4 · measured 20/10/2026, 12:55:09

Benchmarks

Historical direction benchmarks

On this page

This category

StatsGem