48 trading ideas, one pass, and no strategy proven yet

15 Minute Markets writes down every trading idea before it tests it, and it publishes the result whether the idea worked or not. Its study “48 ideas, one bar” read the crypto registry on 2026-10-10. The registry held 48 trials on Kalshi’s 15-minute crypto contracts. Thirty had reached a verdict: 29 failed and one passed.

That one pass has not yet become a proven strategy. The site recommends a strategy only after it passes E1, its live gate on real fills. No strategy has, and the recommendation page says so.

How a trial is registered

A trial is one record in a registry file. It names the idea, the measure it is judged on (net per contract, after fees), the threshold it must beat, and how many 15-minute windows must close before the answer counts. Nobody can edit the record afterward. A person who changes their mind registers a new trial, and both count.

The scorer reads only windows that closed after the registration time, so the data an idea was found on can never confirm it. Until the registered window count arrives, the trial reads “undecided”, however good its numbers look.

Every trial also raises the bar for the rest. The bar is a Bonferroni correction over everything the registry has ever held, abandoned trials included. The first trial had to clear 1.96 standard errors; by the 48th the bar stood at 3.279.

A later rule refuses any trial too small to detect the effect it claims, at 80% power. The study says this was the rule added last and needed first.

What the failures looked like

The failures were not near misses. Of the 21 failed trials judged on dollars a contract, 17 landed within one standard error of zero. Eleven had a positive mean, about what coin flips would give. The closest any failure came was 1.68 standard errors short of its bar.

Most of these trials could not have passed even if the idea was sound. The median failed trial was sized to detect an edge of about 9.7 cents a contract, and the study puts the edges these contracts leave at a cent or two. Halving the edge a test can see takes four times the windows. The three forward trials registered under the power rule ask for 20,600 to 250,400 windows, and at their recorded rates their verdicts arrive between mid-2027 and late 2028.

The fee accounts for much of the rest. On the study’s explore days, favourites won about 78% of the time against an average ask of about 78 cents. The price was close to fair, and the taker fee, about a cent a contract, was the loss.

The study does not show that no edge exists on these contracts. It shows that none of the 48 registered ideas found one large enough to see at its bar and its sample sizes.

The one pass

Trial 11, final-seconds-conditional, asked whether a model that already knows most of the 60-second settlement average beats the book in the final seconds of a window still near even money. It cleared its bar.

The study puts three cautions beside it. A model scored it on recorded order books after the fact, not on orders that filled. Its first scorer had a look-ahead error, and a corrected rescore gave the same mean. Its forward copy, which places paper orders in real time under the registered rule, had 3 of the 30 windows it needs when the registry was read. The study’s own advice is to read the pass “as a reason to run the forward copy, not as an edge.”

Why this matters if you want an edge

Anyone selling a strategy for short event contracts can show you a backtest that wins. The backtest alone does not say how many ideas were tried to find it. Ask how many ideas were tried before this one, and whether the test was written down before the data existed. A result found after 48 tries has to clear a far higher bar than the same result found on the first.

Then ask whether the figures come from fills or from a model. 15 Minute Markets measured that gap in its own record. For its live strategy, paper results overstated live results by a wide margin. A paper fill assumes the order filled in full at the price it asked for. A live order can miss.

A registry that publishes its failures gives you a count to read every result against. The full study stamps each figure to the query, the data files and the code commit that produced it. The summary is free, and the full report with its audit appendix is on the Research plan.

15 Minute Markets is independent and is not affiliated with, endorsed by, or sponsored by Kalshi or Polymarket. This post is education, not investment advice. It reports research results and does not suggest that any strategy will make money.