Selected requests used fewer bytes than full return and kept fixed SEEG decisions in eight sessions with reliable replies

A receiver requested selected original values when four-bit feature codes could not establish a fixed reader decision. The primary method requested up to eight values in each round. In 67,309 later continuous windows from eight public SEEG sessions, it changed no decisions. The mean saving across people was 55.9 percent of message bytes against the prior full-vector return rule. The descriptive 95 percent bootstrap interval across people was 53.5 to 58.5 percent. It also saved 42.6 percent against full return with the same improved outer-bin bounds. The method needed 0.875 extra round trips per window, compared with 0.717 for the prior rule. With 32 extra bytes per message, it cost more than one float32 frame on average. The result assumes finite float32 features, a fixed affine reader, shared parameters, reliable ordered replies, and retained sender values. The package includes an interval proof, source recipes, features, all request masks, figures, code, and independent replay checks. It does not measure radio power, clinical accuracy, stimulation, or network delay.

brain-computer interface · SEEG · feature quantization · selective requests · communication cost · fixed decision · interval arithmetic · reliable transport