Correction, 30 September 2026. Technical wording rewritten in shorter sentences under the readability rule; the meaning is unchanged.
Correction, 1 October 2026. The outcome label now follows the study's value record, which counts this study as having changed a rule or a plan.
Correction, 19 October 2026. This text was reviewed after the study finished, in a catch-up of the earliest studies; the result, outcome and dates are unchanged.
The question
In study three, a test ranking each session among the company's ordinary sessions raised too many false alarms with ordinary sessions in place of 128 splits that added shares. We compare its textbook threshold with one from random sign flips by company, on those data, on made-up data of the same layout and on reverse splits cut to 94 companies.
Rules fixed before the result
The rules for each possible answer were written down and fixed before any result was seen.
Data and method
Each split session was replaced by an ordinary session of the same company, so there was no effect to find, and false alarms were counted over 40,000 repetitions. We did this on the splits that added shares, on reverse splits cut to 94 companies and on made-up bell-curve values of the same layout, judging the textbook and the sign-flip thresholds against 2.5% on each side. A planted fall of 1.75 percentage points measured how often each detected an effect.
What remains uncertain
Only the five-session window was studied, and the ordinary sessions came from one fixed draw made for an earlier study, so every rate depends on it. Stand-ins are drawn independently, so the test cannot show whether real split sessions of different companies move together on the same dates. The sign-flip threshold is exact only when each company's total is symmetric around zero, and it rests on 999 random flips.