Calibration that omitted the test movement code had higher later fNIRS prediction error in all 12 tested adults with equal sample counts

The study tested whether an early fNIRS calibration fit could transfer to a movement code that it did not contain. We used all 12 adults in OpenNeuro ds007420 version 1.0.5. Each fit used ten early events with equal sample counts within an adult. One fit omitted the test code. Five separate fits retained it and each omitted another code. We compared their errors on the same later events. The primary score combined 84 target channels and six codes. The geometric absent/present error ratio was 1.228, with a 95% adult bootstrap interval of 1.131 to 1.398. All 12 adult scores favored code inclusion. Eight planned checks had the same direction. Removing the adult with the largest effect gave a ratio of 1.157. All saved arrays matched extraction from the original source files. The full numerical analysis repeated exactly. Source duration fields differ between BIDS and SNIRF; both are preserved. The fixed event order ties code coverage to calibration timing. The result concerns optical prediction under this replacement rule. It does not show recovery of brain activity. The package preserves data, source hashes, code, predictions, and reproduction commands.

functional near-infrared spectroscopy · calibration transfer · prediction error · movement codes · optical signals