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.
Rules fixed before the result
The rules for each possible answer were written down and fixed before any of the new cases were computed. The lopsided rates themselves had already been seen in earlier studies, and one check reproducing an earlier result ran before the rules were fixed.
What remains uncertain
All cases used one set of ordinary sessions, so chance in how that set was drawn cannot be seen, and stand-ins were drawn independently, so companies moving together on the same dates were not tested. The simulated thresholds were set on this sample, and their separate check guards against simulation error, not against use on another sample. Detection was measured at one planted size, 1.75 percentage points, and one concentration, 12 of 94 companies.