Under nearest rounding alone, 3,650 of 3,653 SEEG pairs from 36 patients had no uniform 5 percent 70–100 Hz power guarantee in second 60 s ds004100 v1.1.3 excerpts

We tested a conditional power limit in public SEEG data. The source was OpenNeuro ds004100 v1.1.3. We used two 60 s excerpts from each of 36 patients and 3,653 matched adjacent contact pairs. The model assumed nearest rounding on each released EDF contact grid, with no other input constraint. The main test used the second excerpt and 70–100 Hz mean Welch power. In 3,650 pairs, two voltage traces gave exactly the same contact integers but had disjoint 5 percent relative power intervals. No estimate from those integers alone can meet a uniform 5 percent guarantee for both inputs. The mean patient fraction was 99.884 percent. Its descriptive 95 percent patient bootstrap interval was 99.653–100 percent. A half-size witness gave the same limit in 3,646 pairs. In 148 primary pairs, witness bipolar RMS was below 1 percent of the decoded trace RMS. These are lower bounds on ambiguity in the stated model. They do not estimate original measurement error or prove physiological plausibility. The package gives a witness tool, a conservative power bound, a proof, pinned retrieval instructions, raw examples, all feature arrays, and results. All 144 source ranges passed hash checks. A full raw-data repeat gave byte-identical feature files. An offline repeat gave identical results. The numerical criterion has no clinical validation here.

SEEG · quantization · rounding · band power · signal processing · precision guarantee · OpenNeuro