How to tell if a picks service is real

Every paid tipster advertises a win rate. Almost none of them publish the three things that would make it mean something. Here is what to ask for, in the order that kills the most bad services fastest.

1. At what price?

A win rate on its own is not a claim. It is half of one. Break-even depends entirely on the odds:

A service betting heavy favourites can advertise a spectacular hit rate while losing money every month, and nothing in the number is a lie. If the average price is not published alongside the win rate, the win rate is decoration. (How to convert any price into its break-even rate.)

2. Over how many bets — and how many events?

Fifty-five percent over 40 bets is not evidence of anything. The honest way to express a record is with an interval around it, and at small samples that interval is enormous:

Every one of those intervals contains 50%. Each of those records is consistent with having no edge at all — and the first one, the best-looking of the three, is the weakest evidence of the three.

3. Was the record written down in advance?

A published record is only worth reading if the losers could not be removed afterwards. Ask two questions: were the picks timestamped before the event, and is the denominator fixed — that is, does every graded call appear, or only the ones someone chose to grade?

A service that decides after the fact which bets counted as "official" is not reporting a record. It is curating one.

4. Do they beat the closing line?

This is the question professionals actually use, and it is the one almost no consumer service will answer.

Closing line value compares the price you got against the price the market settled at just before the event. If you consistently bought at 58c on markets that closed at 63c, you were repeatedly getting a better price than the final consensus — and that is very hard to do by luck.

It matters because it needs far fewer samples than a win rate. A win rate has to wait for outcomes, which are noisy; CLV is measurable on every single bet immediately, whether it won or lost. A service with 300 bets of positive CLV has shown you something real. A service with 300 bets and a 55% hit rate has shown you a coin that landed heads a bit more often.

Applying it to us

It would be dishonest to publish that test and exempt ourselves from it. So: Sharply grades its own calls automatically against the venue's settled outcome, counts each market once rather than each analysis, and shows the running total publicly rather than a selected one.

At the sample size that record currently stands at, the honest statement is that it is not yet enough to claim an edge from — the interval is still wide enough to contain a coin flip, exactly like the first row of the table above. We would rather say that than advertise a number the arithmetic does not support. When the sample is large enough to mean something, it will be published with its interval attached, and you will be able to apply all four questions to it yourself.

In the meantime what the product does is checkable on every single report: run any market through the analyser and it shows the price it read, the estimate it formed without seeing that price, the de-vigged comparison, and the sources it used. You do not have to trust the track record to audit an individual call — which is the point.