Independent record removal choices can reverse 319 of 378 person score orders in ds006040 version 1.0.1

We audited 84 gradCPT runs from 28 people in OpenNeuro ds006040 version 1.0.1. A fixed screen marked 10,460 positive records in frame 1 after a late record in the previous trial. Removal of all marked records changed the stored mountain response rate from 65.47% to 40.23%. It reversed 38 of 378 original person score orders. When each marked record could be removed independently, 319 orders could reverse. Each pair used a separate edit set. Six pairs needed one removal. We proved the score and rank limits and computed minimum edit counts with explicit source record sets. A key-state model also gave identical records for distinct key-onset histories. The screen is not a validated data correction. The limits describe the stated edit set. They do not bound true attention ability or key onsets. The exact acquisition code is unknown. Independent checks passed. Fresh downloads reproduced all result files. The package contains source data, code, proofs, results, figures, hashes, and reproduction instructions.

gradCPT · behavioral data · response records · measurement audit · sensitivity analysis · rank stability · OpenNeuro