A max-min phase rule had higher mean worst loss than an early selected forecast for two delay laws in seven SEEG rest records

We tested phase forecasts under two assumed command delay laws. The laws were a fixed 50 ms delay and a uniform 0–100 ms delay. Their mixture was unknown. The data were seven public OpenNeuro ds006065 v1.0.0 SEEG rest records. Each record supplied four early selected bipolar pairs. An exact max-min rule used two estimated complex phase vectors. Early data selected each model. Later data supplied the test targets. The rule had a mean worst-law cosine loss of 0.794012. An early selected endpoint forecast had a loss of 0.793370. The mean gain was -0.000641. The descriptive 95% person bootstrap interval was [-0.001798, 0.000673]. Three of seven people had positive gains. The exact rule guarantee applies to estimated vectors. The offline phase target uses future voltage. The study does not test live stimulation. The package preserves the rule proof, code, voltage, source records, fitted models, scores, three figures, and reproduction steps. All 3.36 million input values passed new public source checks. A full analysis replay reproduced all saved arrays exactly.

phase forecasting · intracranial EEG · SEEG · robust decision rule · delay uncertainty · cosine loss · brain-computer interface