Every pick on BetsPlug is the end of a long, carefully engineered pipeline. No hunches, no cherry-picking, no hidden rules. This is the exact process - from the moment a fixture is announced, to the moment a verified pick lands on your dashboard.
Each stage is built and monitored independently. Here's exactly what happens under the hood.
Garbage in, garbage out. So our pipeline begins with reliable data sources covering fixtures, results, standings and bookmaker odds.
Every 6 hours we sync fixtures, results and standings from API-Football and football-data.org for 6 European leagues. Bookmaker odds are fetched from The Odds API every 2 hours. During match hours (12:00-24:00 UTC), live scores update every 5 minutes so your dashboard stays current.
Every record is timestamped so we can replay any moment in history exactly as our models saw it - critical for honest backtesting.
API-Football, football-data.org and The Odds API - each covering different aspects of the match.
Fixtures and predictions refresh every 6 hours. Live scores update every 5 minutes during match hours.
Nothing is overwritten. Every change is a new row - history is permanent.
Raw sports data is famously messy. Team names differ between feeds, leagues rename themselves, markets get suspended mid-match. We fix all of it - in a deterministic, reproducible way.
Teams, leagues, players and market types are mapped to a single canonical schema through a manually audited lookup table. Suspended, voided or late-cancelled markets are flagged and excluded before any model ever touches them.
Outlier detection catches bad data the moment it lands. Any record that fails a sanity check (impossible scores, negative times, odds outside a 1.01–1000 range) gets pulled out of the pipeline and logged for manual review.
A prediction is only as sharp as the signals it's built on. This is where a match becomes a structured profile our models can reason about.
For every fixture we compute features across several families: Elo ratings, Poisson goal rates, league standings, recent form, head-to-head history and bookmaker odds. These feed into three independent models that each look at the data differently.
All features use only data available before kick-off. Our models never see the future when generating a prediction - a key safeguard against overfitting.
Dynamic team strength ratings updated after every match result.
Expected attacking and defensive rates for each team based on recent matches.
Historical meetings between the two teams, home and away.
Pre-match odds from major bookmakers via The Odds API.
Current league position, points, goal difference and recent form.
Home/away status, scheduling and fixture details.
A single model will always have blind spots. So we run three different statistical approaches and combine their predictions with weighted averaging.
Each model captures a different aspect of match dynamics. Their predictions are combined into a single set of win/draw/loss probabilities plus a confidence score for every upcoming match.
A dynamic team strength rating that updates after every result. Converts the rating gap between two teams into win probabilities.
Estimates expected attacking and defensive goal rates for each team and integrates over all possible scorelines to produce match outcome probabilities.
A statistical classifier that uses multiple match features - form, standings, home advantage - to predict match outcomes directly.
The three probabilities are combined through weighted averaging. The current weights (Elo 1.0, Poisson 1.5, Logistic 1.0) give slightly more influence to the Poisson model.
Being right isn't enough. A pick only becomes a pick when our probability suggests value compared to the available odds.
We compare our ensemble probabilities to bookmaker odds fetched from The Odds API. Our Strategy Lab filters predictions using rules validated through walk-forward backtesting - such as requiring high home-win probability or low draw probability.
Currently 3 strategies have passed statistical validation with positive backtested ROI over 90 days. But these are backtested results on a limited sample - real-world performance will vary. We are honest about that.
If this number isn't clearly above zero, the pick never leaves the lab.
This is where most prediction sites get vague. This is where we get loud.
The moment a pick is generated, it is timestamped, signed, and published to your dashboard and our public ledger at the same second. You and a regulator see the exact same record.
Once the match ends, every pick is graded automatically against the official result feed. Wins, losses and pushes are all recorded. Losing picks stay on the ledger forever - because a track record that only shows winners isn't a track record, it's marketing.
Subscribers and the public ledger are updated simultaneously - no back-dating possible.
Results are scored from the official feed, not by a human who might look the other way.
Losing picks are permanent. The ledger is append-only by design.
Predictions mean nothing if you don't check them. Our system automatically evaluates every prediction once the match is finished.
Every 6 hours, after results are synced, the system scores all finished predictions. Each pick is evaluated with Brier score and log-loss metrics, and marked as correct or incorrect. This lets us track real accuracy over time.
Current 3-way accuracy (home/draw/away) is around 50% - above the 33% random baseline, but not extraordinary. We publish all results, wins and losses alike, so you can verify the track record yourself.
Every choice in this pipeline exists for one reason: to give you an honest, data-driven probability - not a guess.
Models only see data available before kick-off. Predictions are generated and timestamped before the match starts.
Three independent models (Elo, Poisson, Logistic) cross-check each other. One blind spot can't poison the whole prediction.
Strategy Lab filters are validated with walk-forward backtesting and bootstrap confidence intervals before being published.
You can verify every prediction we've published. Wins and losses. Current 3-way accuracy is around 50%.
Every finished match is scored automatically. Brier score and log-loss track calibration over time.
A two-person team of sports fanatics with an ICT background. Every line of this pipeline is hand-rolled, not bolted together from plugins.
If you've tried a tipster site before, you've probably been burned. Here's exactly why BetsPlug is different.
Now you know exactly how the pipeline works. See the predictions it produces - live, timestamped and ready to verify.