When ten bets are one bet.

The Kelly fraction derived in bet sizing answers one question: given a bankroll and one bet, how much maximises long-run growth? Its derivation contains no other positions. Applied to a portfolio, it silently assumes each bet resolves on its own coin.

Stake 10% on each of five genuinely independent contracts and the arithmetic behaves roughly as intended — some win, some lose, the bankroll compounds. Stake 10% on each of five contracts that all resolve on one central bank decision and you have not made five bets at 10%. You have made one bet at 50%, on a proposition you sized as though it were worth 10%.

The same event, sliced. Exchanges list many contracts per event — thresholds, ranges, margins. They are constructed from one underlying and resolve together by design. This is the easy case and still catches people, because the tickets have different names and different prices.

The same driver, different subjects. Harder, and more common. Several markets that each depend on one inflation print, one election result, or one weather system are correlated without sharing a title. Nothing on the screen groups them.

The same source of edge. Subtlest of all. If every position came from one model, one signal, or one person's read, then the thing being bet is the reliability of that source. The contracts may resolve on unrelated facts and still fail together, because what fails is the method.

This applies squarely to what this site does, so it is worth stating plainly. Alerts drawn from a small roster of wallets are not independent draws. Several may enter the same market, or several markets driven by one event, within the same hour — and every one of them shares a single dependency: whether those wallets' edge is real and still there.

That is a structural property of following anybody's flow, not a defect particular to this roster. It is why the record publishes each wallet's sample size next to its win rate rather than a headline number: a strategy whose positions share a dependency should be judged on how much evidence stands behind that dependency.

The honest answer is that correlation is measured badly by everyone, including institutions with dedicated risk teams, because the correlations that matter are the ones that appear during the event you did not model. What is tractable is cruder and more useful: group positions by what would resolve them, and size the group rather than the tickets.

If five contracts all pay out or all expire worthless on one announcement, the exposure is that announcement, and the arithmetic in the sizing guide takes the group as its input. That reframing costs nothing and removes the most common way a book turns out to have been one bet — which, again, is a description of how the mathematics behaves and not a recommendation about your money.