← Insights Responsible betting

Why we publish every losing selection

The prevailing convention in this industry is to present a curated record. Successful positions are screenshotted and circulated; unsuccessful ones are quietly removed. The resulting impression is of a service that rarely gets anything wrong.

We have taken the opposite approach. Every selection we publish is graded won, lost or void after full time and retained permanently on our public record. Our reasoning is set out below.

An unverifiable strike rate is not information

Where a service claims a high success rate but publishes only the positions it chooses to show, that figure carries no weight. Survivorship bias accounts for the entire effect: issue ten different selections across ten different channels, remove the nine that lose, and the remaining one appears to demonstrate exceptional judgement.

The only meaningful strike rate is one calculated across every position taken, including those that failed.

Losing positions carry analytical value

When a selection loses, the outcome is fed back into our model. Markets in which we underperform receive tighter filtering criteria; assumptions that proved unreliable are revisited. A complete record of losses is a prerequisite for improving the process — concealing them would remove the basis for any correction.

It informs how members stake

Where members can see that a well-constructed process still loses roughly three positions in ten, individual selections are treated proportionately rather than as certainties. Staking becomes more measured, loss-chasing becomes less likely, and performance is assessed over a reasonable period rather than a single evening. That shift protects members more effectively than any individual selection.

What to require of any analysis service

Where a service cannot provide these, the reason for the omission is worth considering carefully. Where it can, hold it to that standard consistently. We built ours to be assessed on exactly these terms.

View the public record