The AI tipster standings moved. Nobody changed a thing.

On the morning of 28 July, Gemini's return in the Silicon Tipster League sat below the break-even line. By that evening it was comfortably above it. Nothing about the model changed in between. An evening of racing settled, a few of its picks came in at long prices, and a number near the top of our table moved a long way in about ten hours.

That is the honest state of the league after 25 days and 6,820 settled picks. Five AI chatbots, our own racing model entered twice and a back-the-favourite baseline each back one horse in every race we publish a card for, logged before the off and settled at the industry Starting Price, wins and losses alike. The table updates as results come in, and it does not sit still.

We could write you a post every week about whoever happens to be top. It would waste your time. Over the ten days to 30 July, Claude's combined return crossed the break-even line four separate times while its method stayed exactly the same. Our own Edge entry was the highest name on the board on 21 July and sits at the bottom of it now.

So this one is about the movement rather than the standings. Once you know what makes a young record swing about like that, you will read every tipster's record differently, ours included.

Two things make these numbers jumpy, and neither one is a fault in the tipsters.

The first is that a return is carried by a handful of results. In AI Finds the Value Bets, where each model sees the market's own view and is asked to find where it is wrong, Gemini has 45 winners from 413 settled picks, a strike rate of 11%. Its return in that arm is +16.2%. Whatever that figure reads when you land on this page, it rests on those few dozen results, most of them at long prices. Two horses landing or failing to land on a Saturday move the whole line.

The second is that one model can look like two different tipsters at the same moment. Gemini's return with the odds hidden is -3.7%; with the market in front of it, +16.2%. Claude's two arms currently sit at +2.9% and -8.6%, and a week ago they were the other way round. That is one model reading the same races over the same fortnight. If one tipster can hold two records that far apart at once, no single one of them is telling you much yet.

What to do with a record this young

Whoever leads today leads partly on luck, and that is not modesty. Over a sample this size the distance between the top and bottom names on the board is the sort of distance that ordinary variation produces on its own. Treat the order as provisional and expect it to reshuffle again, because it has reshuffled every week so far.

Our own two entries sit at the bottom of it. Stablebet Edge is returning -39.4% and the Stablebet Model -31.7%, and we publish those for the same reason we publish everything else. A record you can only see when it flatters us is not a record. On 21 July the Edge entry led the whole table, which tells you as much about it then as its position tells you now.

Underneath the churn, the thing that has not moved is the margin built into every price. It applies to all eight entrants equally, no betting system in the Lab has got past it, and a fortnight of one model running warm is not evidence that anything has.

Follow it live on the Silicon Tipster League, see what changes when an AI sees the odds, or read the standings roundup. Nothing on that board is a tip, and none of it is betting advice. 18+, please gamble responsibly.