Alternating two stable timing maps made one fixed SEEG state reader unstable with 100 ms simulated gaps

One fixed learned SEEG state reader had two stable cycle maps with spectral radii 0.978644 and 0.834426. Their alternating two-cycle product had spectral radius 2.377684. The maps used 100 ms simulated gaps and requested observed intervals of 28 and 62 ms. A search of 36 fixed readers found 99 such duration pairs in two readers. We added a gap-end operator norm guard that does not need the next observed interval. Under fixed weights, known gap length, and shared state-independent voltage history, its bound proves contraction of initial state differences for any observed interval. In nine reused public rest records, the guard reduced equal record mean phase error by 10.07 degrees against held voltage but increased it by 1.23 degrees against the original learned rule. Large phase errors remained. All 7.2 million selected voltage values matched fresh public source extracts. A separate full run reproduced all 14 numerical result files exactly. The package includes data, weights, code, proof, figures, source receipts, and instructions. The results concern a numerical filter state. They do not establish biological phase accuracy, actual artifact recovery, hardware performance, or stimulation benefit.

SEEG · state estimation · switching stability · contraction · missing data · phase error