[aaronkanzer]

In a Neuroscope XML, acquisitionSystem/samplingRate is the rate of the wideband .dat file. The downsampled .lfp and .eeg files are written at fieldPotentials/lfpSamplingRate. NeuroScopeRawIO always uses samplingRate, so LFP files get the wideband rate. Their times and durations come out too short by the ratio of the two rates, 16x for the common 20000 / 1250 Hz setup.

Repro with neuroscope/dataset_1 from ephy_testing_data. The XML has samplingRate 20000 and lfpSamplingRate 1250:

from neo.rawio import NeuroScopeRawIO
r = NeuroScopeRawIO(filename="YutaMouse42-151117.eeg"); r.parse_header()
print(r.header["signal_channels"]["sampling_rate"][0], r.get_signal_size(0, 0, 0), r.segment_t_stop(0, 0))

Observed on 0.14.6 and master: 20000.0 10 0.0005

Expected: 1250.0 10 0.008

Fix: when the data file ends in .lfp or .eeg and the XML has fieldPotentials/lfpSamplingRate, use that rate. .dat files are unchanged. A test covers both suffixes.