v0.302026-06-06trustWe dropped the "+EV" claim — and here is exactly why
We tested publishing +EV betting picks. The data falsified it: for our users the playable market is ~2 GGL-licensed German sportsbooks, which is structurally too thin to beat the closing line — "best price across 40+ books" was never actually bettable for you. So we dropped the profit claim and repositioned as a decision-support analytics tool. We kept every past result visible, including the finding that killed the old approach — deleting losers is what tipsters do.
- Retired the +EV / beat-the-closing-line promise. For a German user the playable market is ~2 GGL-licensed books — you can't price-shop two books, so a betting edge isn't reachable on this data. We say so plainly instead of burying it.
- Methodology fix (we were wrong): we used to count each bookmaker's price for the same selection as a separate bet. That is statistically wrong — the same decision across 10 books is ONE decision, not ten. Headline counts are now decision-level (~288 decisions, not 891 book-rows).
- /track-record now leads with model CALIBRATION — are our probabilities accurate? — instead of betting ROI: predicted vs actual win rate, Brier score vs the base-rate baseline, with a statistical-significance gate on the 'beats baseline' badge. Accuracy, not profit.
- The old +EV pick ledger stays as a clearly-labeled "research-phase archive (retired)", including the finding that ended it. Nothing backdated, nothing smoothed — the earlier changelog entries below are the real sequence.
- Removed /sure-bets: our feed surfaces ~0 real arbitrage, and a phantom arb page would be a false promise.
- Reframed the site to the tool's real value — de-vigged fair value, a sharp-consensus reference line, and odds comparison across the books you can actually bet — research in seconds, no profit guarantee.