The markets we switched off

There's one decision almost nobody publishes: removing something that half worked.

In 2026 we pulled three markets out of Delta's recommendations. None of them left for being wrong. They left for being right at a price that didn't pay, and that's the most expensive mistake in this whole business.

The trap: winning cheap

Picture a market you get right six times out of ten. Sounds good. Now add the price: it pays 1.40.

Across ten attempts, six return 1.40 and four take everything. You put in ten, you get back 8.40. You were right 60 % of the time and you lost 16 % of your money.

That's exactly the profile of a short-priced market: high hit rate, negative result. It's comfortable to sell because the headline sounds great, and it bleeds you slowly.

What we switched off

• Double chance (August 2026). It pays little by definition: you're covering two of three outcomes. Our model overestimated it at the root, and no amount of recalibration fixed that.
• Both teams to score (August 2026). The painful one, because it's popular. Measured across every match — not only the ones that reached a recommendation — the real frequency matched what the market paid almost exactly. Translated: there was no edge there to find. Not ours, not anyone's.
• Corners (August 2026). Same pattern: hit rate comfortably above half, negative return, prices too short to sustain it.

What we still calculate

The corners projection didn't disappear from the app. It's still calculated and still shown on the match page, because it's useful information about how a match is likely to be played.

What changed is that it no longer produces picks. Showing a number and betting on it are two different things, and keeping them apart is what lets us be honest about the second one.

What's left running

Today the model recommends on: match winner, over/under goals across three lines, goals by half and shots on target. Fewer markets than a year ago, deliberately.

Why we're telling you

We could skip it. Nobody audits anybody in this industry, and a quietly retired market is a question no user ever asks.

We're telling you for three reasons:

• Because a model that only adds things isn't learning anything. Removing is as much a part of the method as adding.
• Because it explains why there are sometimes fewer picks than there used to be. The model isn't weaker: it stopped betting where it wasn't winning.
• Because it changes how you read any hit-rate claim, ours and everyone else's. A win percentage without the price beside it means nothing.

The right question is never "how often are you right?". It's "at what price?".