For one faulty shaft, the median of shaft means had a 5.1% higher mean normalized bound in four SEEG records

We compared contact medians with medians of shaft means in public SEEG data from four patients. We allowed one fixed shaft to change after a causal 70 to 110 Hz filter. All other shafts stayed fixed. We derived exact finite bounds for the reference error RMS. We tested 48 fixed pre-pulse epochs and all 54 shafts, for 648 cases. The mean of four patient median normalized bounds was 0.13390 for the contact median and 0.14074 for the median of shaft means. The second value was 5.1% higher. Three patient medians were higher and one was lower with the shaft method. Removing the pulse shaft changed the direction in two patients. Both mean references had unbounded errors under the same unlimited fault amplitude model. Direct finite replacements checked all primary and sensitivity bounds. Fresh source extraction and repeat analysis matched the archived inputs and numerical results. The package includes a reusable bound tool, proof, exact filtered grids, raw check contacts, all case tables, four figures, source hashes, and repeat instructions. The result measures reference stability. It does not establish decoding or stimulation benefit.

SEEG · common reference · median reference · shaft mean · fault tolerance · signal processing · brain-computer interface