The short version: a model estimates the probability of a draw, compares it with the market price, and we only publish when the difference is worth betting. Everything that happens next — results, profit, losses — is settled by software and published automatically.
The draw is the least loved outcome in football betting. Casual money chases winners, which means the X is where prices are most often a little wrong — and a little wrong, at odds around 3.0–3.5, is enough. We specialise in exactly one market instead of pretending to be good at all of them.
Mostly competitions outside the spotlight — second divisions and leagues where bookmaker prices move slowly and draw rates are structurally high. Big-league prices are efficient; we go where the market pays less attention.
We publish the underlying numbers: draw rates for 12 leagues across 7,600 matches, with the market price of the draw alongside each one.
For every candidate match the engine estimates the draw probability from historical scoring patterns, and compares the price we can take with the wider market's price for the same draw. A pick is published when our price sits above the market consensus and the model finds no reason for the gap — the edge is in the price, not in predicting draws better than anyone else. Each pick shows that price premium alongside the model's own probability. If nothing clears the bar, we publish nothing that day. No pick is ever added to fill space.
Model probabilities are estimates, not certainties. We show them so you can judge our reasoning, not because they are guaranteed to be right.
Picks are staked in units, not euros, so the record says what the picks did on any bankroll and in any currency. Most get one unit. A pick priced between 2.6 and 3.2 and above the market consensus gets two. When the two teams are also evenly matched — the market can barely separate them — it gets three: over twelve seasons, requiring both signals to agree returned more than either alone. A pick whose price only matches the market gets half.
A unit, for us, is 2% of the bankroll, capped at €120, recalculated once a week (Monday) and rounded down to a clean step, starting from €1,000. That's where the € figures on the track record come from — and why the two returns differ slightly. The unit ROI measures the picks; the € bankroll measures the picks plus the effect of a stake that grows and shrinks with the bank. Both come from the same settled results.
Results are fetched and settled automatically by software within minutes of full-time — no human edits, no quietly deleted losses. Every pick stays on the track record forever, with a link to the match so you can verify the score yourself.
The current configuration was validated on 13,002 qualifying picks across nine seasons (2018–2026) of the target competitions: a 35.3% draw win rate at average odds around 3.1, and +10.4% ROI with the staking ladder above, out-of-sample. The twelve competitions were each selected by the same rule — the gap between how often they draw and how often the price says they should — and only leagues whose gap survives its own margin of error are included. The live record is young and will swing — at these odds, losing streaks of 5–8 picks are statistically normal. Judge us over months, not weeks. Past results don't guarantee future returns.
Not "sure wins", not fixed matches, not 100 predictions a day. Some days have zero picks — that's the discipline working, not the model failing. We publish a handful of value picks and track every one of them in public.