Every pick BetsPlug publishes is timestamped, probability-weighted and logged to a public ledger - win or lose. This is how the numbers are built, and how real users put them to work.
A rolling, honest snapshot of how our models performed over the last year of live publishing.
Hit rate on value picks
Only picks flagged with +EV ≥ 3%
ROI on flat 1u stake
After closing-line slippage
Graded predictions
Across 9 sports and 70+ leagues
Brier score (lower = better)
Calibrated across win/draw/loss
No black boxes. Here is the exact pipeline every pick runs through before it ever reaches your dashboard.
We pull match, lineup, injury and odds data from 14 audited providers every 30 seconds. Each record is versioned so we can replay any moment in history.
Team names, leagues and market types are mapped to a single canonical schema. Outliers, late-cancelled games and suspended markets are flagged and removed before modelling.
For every fixture we compute 1,200+ features - Elo, xG trends, rest days, travel, head-to-head, referee bias, market movement - then store them in a point-in-time feature store so training never leaks future information.
Four independent models vote on the outcome: a Poisson goals model, a gradient-boosted classifier, an Elo-based baseline and a market-implied calibrator. Their probabilities are blended with stacked regression.
We compare the ensemble probability to the best available market odds. Only picks with a statistically significant edge - after commission and expected closing-line slippage - are marked as value.
Picks are timestamped the moment they go live. Once the match ends, results are graded automatically and written to the public track-record ledger - win, loss or push.
Every number on this page survives these guardrails. If a result can't, it doesn't get counted.
Models are trained on data that was actually available at kick-off - no hindsight, no silently re-graded matches.
ROI figures subtract realistic slippage between our publish time and the closing line, so you see returns you could actually execute.
Losing picks stay on the ledger forever. A track record that only shows winners isn't a track record - it's marketing.
Every pick includes the match ID, market, odds and timestamp so any user or auditor can verify the outcome independently.
The numbers only matter if they change decisions. Here's what that looks like for three typical BetsPlug members.
Weekend football specialist
“I filter the track record by Serie A and over/under markets only. In 2025 that slice hit 61% on 412 picks - so I size up confidently on Sunday mornings instead of second-guessing myself.”
Uses the ledger to pick a market where the model has a proven edge, then ignores the rest. Result: fewer bets, higher conviction, better ROI.
Data-driven NBA fan
“I care about calibration more than hit rate. Seeing that BetsPlug's NBA moneyline Brier score sits under 0.07 tells me the probabilities are honest - not just lucky.”
Trusts the probabilities to build her own parlays and spread bets. The public ledger is her proof the models haven't drifted.
Full-time value bettor
“I plug the track record CSV into my own bankroll tool every Monday. Because every pick has a timestamp and closing price, I can verify the edge is real before I risk a cent.”
Uses the historical ledger as an independent backtest. If the closing-line value holds up there, he trusts it for live staking.
Every graded prediction we've ever published is searchable by match, date, market and model. No filters hide the losers.