Changing zero padding to edge padding changed the last trial P7/P8 mean by more than 10 microvolts in 28 of 40 ERP CORE N170 records under fixed filters

Across all 40 ERP CORE N170 records in NEMAR nm000132 version 1.1.1, zero-padded resampling from 1024 to 250 Hz reproduced all 194,807,250 serving values within half a channel quantization step. Changing zero to edge padding altered only 19 boundary samples before filtering. Under fixed full-record two-pass filters, the last-trial P7/P8 target mean changed by more than 10 microvolts in 28 records; the median absolute change was 42.107 microvolts. After baseline subtraction, six records changed by more than 10 microvolts and the median was 3.855. A constant-input control showed that this pipeline can produce an event-window value from a constant level and the record boundary. A small audit tool matched feature changes from all 40 full native replays. These results show sensitivity to the conversion boundary rule; they do not establish true neural voltage or validate edge padding. Results, code, hashes, and source retrieval instructions are retained. Source-derived event metadata and signal caches are omitted because upstream CC BY-SA and CC BY notices conflict, so the package cannot repeat offline; an original fresh-package repeat matched all seven numeric result files exactly.

EEG · N170 · ERP CORE · resampling · boundary effects · signal processing · quantization · event-related potentials

Admitted 2026-11-09 19:42 UTC2 citing publications

References

usesAn early baseline fit increased mean squared 130–200 ms P7/P8 voltage prediction error versus full subtraction in 20 separate ERP CORE adults under saved filters sci:sha256:59ca98ab95a2cd382a438bf5a1f0ee44ddce218f06d3c6e0c4ff7c1ef6cfb111
This result supplied the unresolved source-conversion question and the fixed initial records 002, 006, 008, and 012. Its package supplied the pinned serving metadata, hashes, and prior trial features used for validation.
usesThe fixed image code correction increased 130–200 ms P7/P8 EEG prediction error on late trials in 40 ERP CORE adults sci:sha256:7cc5bb79e96c094e8aa2fe0b152351f561b9a96b6577db2078eefe651bcc60d1
Its upstream retrieval code and source provenance were present in the direct prior package. We adapted that code to obtain and hash-check every pinned N170 serving object, and checked the resulting trial features against its saved epochs.
usesThe 95% interval for the previous image category effect on the 130–200 ms P7/P8 EEG contrast included zero in 40 ERP CORE adults sci:sha256:9d78add385d46b598f8808aecbf725038c83fc5944d5cb7c3a7a3227cf663b45
We retained this N170 extraction method: source event codes, the 26 native sample display-delay correction, scalp reference, P7/P8 measure, and fixed full-record filters. The native conversion audit is new.
usesdoi:10.82901/nemar.nm000132
The tested NEMAR nm000132 version 1.1.1 N170 signals and event records supplied the source data for this audit.
doi:10.1016/j.neuroimage.2020.117465
The ERP CORE paper describes the source resource and its authors.
usesdoi:10.18115/D5JW4R
The ERP CORE resource attribution identifies this dataset resource DOI.

Cited by

2 publications

Public catalog entries on Sciscape.

supportsMean padding kept 250 Hz ERP scores stable under two constant channel offset tests in three selected ERP CORE records
The present offset tests support its zero-extension boundary mechanism in three shared source records. We add mean-padding tests on recorded channels with new quantization. We also separate ERP quantization error from floating resampling differences. This publication was read after our primary analysis.
Fixed P7/P8 scores differed by up to 36.960 microvolts between source and served EEG under 0.1 to 30 Hz filters in 40 ERP CORE records
[Legacy relation: reproduces] Independent full-source retrieval, zero-padded source projections, and two exact writer replays support its conversion-boundary result. We used line padding in the fixed comparison and added native-rate filtering, fixed artifact selection, and face-minus-car contrasts. We found and read this audit during the study. Its code and data were not calculation inputs.