← back to the WC pass The Receipts

127 bets. 55.1% strike. +16.5% ROI.

No vertical equity curves. No 96% accuracy claims. Just the actual numbers, with the actual definitions.

Audited bets (since 2026-03-12) 127

Every staked bet logged automatically by the model runner, then settled match-by-match. Real money, not paper.

Strike rate 55.1%

Bets won ÷ bets settled. For comparison, a coin-flip strategy run at ~2.0 odds breaks even at ~50%; we need ~52-55%+ to clear typical bookmaker margin depending on average odds taken.

ROI on stakes +16.5%

Return on every stake across the full settled sample. We keep the focus on the rate, not the pennies — a young track record is best judged on ROI, strike rate, and closing-line value, not the size of the bankroll behind it. Settled results only; open futures positions (outrights still running) sit outside these numbers until they settle.

Closing-line value (CLV) under review

CLV measures whether our entry odds beat the final market price — the metric tipsters can't fake, which is exactly why we won't publish a soft one. We're re-verifying our closing-odds data for quality right now, so we're holding the CLV figure until it's clean. We'd rather show “under review” than a number we can't stand behind.

Settled — World Cup (model bets) 11W – 7L

🇲🇽 Mexico Under 2.5 ✅ · 🇨🇦 Canada BTTS Yes ✅ · 🇺🇸 USA −0.5 ✅ · 🇶🇦 Qatar +1.5 ✅ · 🇧🇷 Brazil BTTS No ✅ · 🇪🇸 Spain BTTS No ✅ · 🇨🇮 Ivory Coast BTTS No ✅ · 🇫🇷 France −0.5 ✅ · 🏴󠁧󠁢󠁥󠁮󠁧󠁿 England Win-to-Nil ✅ · 🇿🇦 SA-Canada BTTS No ✅ · 🇫🇷 France Over 2.5 (Sweden)
🇳🇱 Netherlands Under 2.5 ❌ · 🇬🇭 Ghana/Panama +0.5 ❌ · 🇫🇷 France Over 3.5 (Iraq) ❌ · 🏴󠁧󠁢󠁥󠁮󠁧󠁿 Kane 2+ SOT ❌ · 🇩🇪 Germany Win-to-Nil ❌ · 🇩🇪 Germany BTTS No ❌ · 🇫🇷 France Over 3.5 (Sweden) ❌ — we post the losses the same as the wins.
Eighteen model bets settled, eleven landed. England's win-to-nil over Panama — built on Panama failing to score a single goal all tournament — and France's −0.5 win over Norway kept the run going; the misses stay public too — France's 3-0 over Iraq fell a goal short of our Over line (the exact risk we flagged), and a Kane shots-on-target call lost. The Round of 32 then brought a double clean-sheet loss on Germany v Paraguay, where Paraguay scored and went on to knock Germany out, a lesson on filtering clean-sheet bets to genuinely toothless opponents. Then France's 3-0 win over Sweden split our two overs: the Over 2.5 landed comfortably while the Over 3.5 juice fell a goal short, a clean safe-won, juice-missed night that nudged the ROI down honestly. Non-model and for-fun bets are logged separately and stay out of the model's column, win or lose.

See all 127 bets — full ledger →

Performance by market

MarketVerdict
Over/Under goalsStrongest edge — best-performing market across the track
BTTS (both teams to score)Mid — selectively profitable, regional bias
1X2 (match result)Worst — efficient market, hardest to beat
Cards & cornersBacktested vs AFCON/CONCACAF — no per-match edge found, so we don't stake them

Posting the worst market alongside the best is intentional. If a tipster only shows you their winners, they're showing you a survivor bias, not a strategy.

How the model works

We use a multi-layer ensemble model. Layers include Dixon-Coles match probability fitting, Elo ratings, expected-goals divergence (rolling xG vs actual finishing), weather effects on goal totals, confirmed lineup/absence gating, referee discipline profiles, head-to-head venue patterns, and others.

We don't reveal the exact layer count, weights, thresholds, or layer names — that's the only piece of IP we have, and giving it away would kill the edge. What we do reveal: every bet, every edge %, the odds taken, every win and loss, in real time.

The bets are public. The results are settled in real time. The method stays private.

External verification

Methodology limitations (be honest about it)

127 bets is suggestive, not statistically significant at the p<0.05 level. With this sample size, the true edge could plausibly sit anywhere from +5% (real but smaller than headline) to +25% (lucky sample). We'd need 500+ settled bets to make a definitive statistical claim. We're publishing the track record as it builds — you're watching the experiment run, not buying a finished proof.

Other caveats worth knowing: this is still a young sample, so variance is high in percentage terms. Our closing-line-value data is being re-verified for quality before we publish a figure — we won't put a CLV number up until we trust it. The Cards market is still in micro-validation and shouldn't be treated as a proven layer yet. If those caveats kill your interest, that's the correct response — better now than after the £25 pass.
Back to the WC pass →