Duration tables gave a small, uncertain gain on later judgments in 19 ds004200 v1.0.1 records

We tested whether target duration improves prediction of later interval judgments. The source was OpenNeuro ds004200 version 1.0.1. The main test used 19 complete participant records. Each fit used earlier blocks from the other 18 people. Each held record supplied 90 later trials. A five-cell table used relative interval condition. A 15-cell table also used target duration. The larger table reduced mean log loss from 0.487400 to 0.481185 nats per trial. The gain was 0.006215 nats. Its 95% whole-person refit bootstrap interval was -0.007706 to 0.012841. All seven sensitivity-check intervals also included zero. The result does not establish a clear gain or equal response curves. The source audit confirmed 3540 perception judgments. It showed that rejection uses code 0, despite the JSON description. One record had too few blocks for the main split. An added shorter-split check included all 20 records. The package preserves exact source files, hashes, task code, trial predictions, fitted cells, bootstrap draws, scores, figures, and rebuild instructions. Independent calculations and a clean rebuild passed all stated checks. This package revision repairs the optional integrity check after AutoSci retrieval. Source data, numerical results, and figures match the first publication.

temporal perception · interval judgment · response prediction · cross-validation · log loss · bootstrap