A burst adjustment reduced score variance in 183 ds000030 BART records at a cap of 12 and kept the expected mean under hidden independent uniform burst points from 1 to 12

We tested whether a burst adjustment could reduce the variance of a capped mean score in the Balloon Analogue Risk Task (BART). We used the first ten complete reward trials from 263 OpenNeuro ds000030 version 1.0.0 records. We fitted one coefficient on 80 people and tested it on 183 other people. The main pump cap was 12. The adjusted to weighted sample variance ratio was 0.71115. Its paired 95% interval from 20,000 person bootstrap samples was 0.59685 to 0.86124. The test mean standard error fell from 0.13424 to 0.11320 pumps. The mean changed from 4.60080 to 4.57553 pumps. A proof shows that the control score has zero expected value when burst points are hidden, independent, and uniform on integers 1 through 12. A coefficient fitted on independent people then keeps the expected capped intention mean under actual earlier histories. All five planned checks had variance ratio intervals below one. Exact fraction checks, a separate parser and score calculation, fresh downloads, and a full result replay passed. Variance increased for a fixed-four-pump simulated policy. A fixed-one-pump exact case also proves that no nonzero coefficient gives a uniform variance reduction. This result concerns a group mean under stated task assumptions. It does not establish individual intention accuracy or a neural effect. The package preserves source bytes, hashes, the local plan, proof, code, scores, saved resamples, simulations, figures, and instructions.

Balloon Analogue Risk Task · BART · control variates · burst adjustment · censored observations · variance reduction · risk measurement