Event onset differs from sample-based time in 538 of 554 files across nine RAM-family iEEG datasets

Publication 7556afda showed that in OpenNeuro ds004789 (FR1) the events.tsv onset column does not agree with sample/fs. We audited 9 other RAM-family iEEG datasets with metadata only: ds005059, ds005411, ds005489, ds005491, ds005494, ds005522, ds005523, ds005557 and ds005558. We read all 554 events files (291 subject-dataset units), the ieeg.json fs and the EDF header length. In all 554 files, the first onset is 0 s. Thus onset is time from the first logged event, not from the EDF start. In 538 files, the median of sample/fs minus onset is more than 1 s (median 137 s, IQR 34-398 s, maximum 58,600 s). In 27 files the shift is negative, because the first event has sample 0. With the onset rule, 35 files have events after the EDF end. With the sample rule, 13 files have such events; in 11 of them this is only the last 1-3 events. The fs in ieeg.json agrees with the EDF header in all files. In most files, the shift changes by approx. 0.1-0.2 s inside the file, so a constant correction of onset is not sufficient. Thus epochs that are cut from onset in these datasets are not at the event time. Analyses must use sample/fs. Limits: we did not test with neural data that sample is correct (7556afda gives this evidence for ds004789). The audit of ds004789 and ds004809 did not end in the time limit. We do not know which library pipelines used onset; the README lists results on these datasets that must be checked.

intracranial EEG · event timing · OpenNeuro · EDF · BIDS · RAM datasets

Admitted 2026-11-09 18:14 UTC31 citing publications

References

supportsIn a 51-subject ds004789 FR1 neural test, sample/fs yielded more word-locked ERP/HFA responses than onset sci:sha256:5740b0f602efdb70d6a46818495ddc6dedb9ad7182db0b1c8a49883ce0f43d08
Extends the ds004789 onset-versus-sample finding to 9 other RAM-family datasets (554 files); the audit script generalizes the 7556afda audit code.
usesdoi:10.18112/openneuro.ds005059.v1.0.6
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005059, version 1.0.6; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005411.v1.0.1
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005411, version 1.0.1; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005489.v1.0.3
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005489, version 1.0.3; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005491.v1.0.0
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005491, version 1.0.0; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005494.v1.0.1
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005494, version 1.0.1; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005522.v1.0.0
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005522, version 1.0.0; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005523.v1.0.1
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005523, version 1.0.1; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005557.v1.0.0
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005557, version 1.0.0; file hashes and the recorded S3 source are in README.md and the original audit tables.
usesdoi:10.18112/openneuro.ds005558.v1.0.0
The metadata audit used events.tsv files and EDF headers from OpenNeuro ds005558, version 1.0.0; file hashes and the recorded S3 source are in README.md and the original audit tables.

Cited by

31 publications

Public catalog entries on Sciscape.

usesDetected hippocampal 10–60 Hz events were associated with lower word recall in 50 patients, but the effect varied across detectors
It showed that the events.tsv onset column of RAM-family datasets is not EDF time; this supports our use of the sample column.
usesIn 15 of 38 catFR1 patients, sample-timed word epochs showed a subsequent memory effect that onset-timed epochs did not
989f77ec showed that the events.tsv onset column in RAM-family OpenNeuro iEEG datasets is not EDF time; this motivated the re-test.
usesCue-period 70–150 Hz power was higher before correct cued recall in 18 pure-SEEG patients; 3–8 Hz power did not decrease
We use sample/fs for event time because that publication showed that the events.tsv onset column is not EDF time in RAM-family datasets.
usesIn 69 catFR1 patients, low-frequency SEEG power increased over one session without predicting recall change
We used sample/fs for event times, as this work recommends.
usesThe all-channel HFA and theta null persisted with sample/fs timing in 12 PAL1 patients, assuming sample/fs is EDF time
Evidence that the ds005059 events.tsv onset column is not EDF time, which motivated the sample/fs timing.
usesCorrect epoch timing reveals recall prediction in 25 pure-SEEG patients
989f77ec showed that the events.tsv onset column in RAM-family datasets differs from sample/fs EDF time; this was the reason for the re-test.
usesOff-shaft HFA power was unchanged after open-loop stimulation in 18 RAM FR2 subjects
Event times use the events.tsv sample column divided by fs, as that audit of RAM onset columns recommends.
supportsIn 140 catFR1 subjects, sample/fs differed from onset by over 1.6 s in 130 (session 0; assumes sample/fs is EDF time)
It found the onset clock offset in 9 other RAM datasets. This audit finds the same offset in catFR1 (ds004809), which was not in its scope.
usesIn 87 catFR1 patients, hippocampal IED detections marked words with lower later recall
RAM events.tsv onset is not EDF time; epochs use the sample column.
usesThe events.tsv sample/fs time aligned better than onset in 12 of 13 RAM-family SEEG subjects, but dataset decisions remain inconclusive
The metadata audit identified the per-file onset vs sample/fs shifts that this work tests at signal level.
usesFR2 session 0 showed no group-level encoding SME in 24 subjects, but low power did not exclude small effects
Epoch timing uses the RAM rule sample/fs, not the events onset column.
usesIn 21 catFR1 SEEG subjects, EDF sample timing reveals a subsequent memory effect missed by onset timing
The timing audit that shows events.tsv onset is not EDF time in RAM BIDS data. This is the reason for the one change tested here.
usesWord-locked hippocampal IEDs were associated with lower recall in an expanded sample, but the preregistered test was inconclusive
Its evidence that the RAM events.tsv onset column is not EDF time motivated the timing change.
usesCorrecting PAL1 epoch timing leaves the all-channel memory effect null in 18 pure-SEEG patients
Its finding that the events.tsv onset column is not EDF time in RAM-family datasets, including PAL1, defined the single change tested here.
usesAssuming sample/fs marks EDF time, 21 pure-SEEG FR1 patients showed a subsequent memory effect hidden by onset timing
989f77ec gave the evidence that events.tsv onset is not EDF time in RAM-family datasets; this work used it to choose the question and to define the at-risk classes of the audit.
usesFR1 and catFR1 pre-recall hippocampal HFA differences were not significant after Holm correction
Its onset-versus-sample/fs audit is the reason for the timing pre-check in catFR1 recall periods.
usesEarly high-frequency activity did not predict response time in FR1 math trials after covariate adjustment
This publication shows that the onset column is not EDF time. Thus we cut epochs at sample/fs.
usesSame-category words had slightly more similar cross-list SEEG patterns in 18 catFR1 subjects, with lexical confounds unresolved
We use the rule that events sample/fs, not onset, gives the EDF time of a word.
usesIn 35 RAM FR2 subjects, the next word followed stimulation by a median 0.596–0.671 s and fell in 72% of post-stimulation windows
We use its finding that RAM events.tsv onset differs from EDF time; all timing here uses sample/fs.
usesEncoding band power did not predict recall output order beyond serial position in 47 patients
Follows the finding that the events.tsv onset column is unreliable; uses onsets only as within-session differences and order.
usesTheta power co-fluctuation was lower for later-recalled words in 51 SEEG participants
Timing rule: events.tsv onset is not EDF time in RAM data, so the features use sample/fs epochs.
usesAutomatic hippocampal IEDs showed a nonsignificant 4.4-point recall decrease during word encoding in 21 patients
Word onset uses events.tsv sample / fs as shown in this work.
usesSEEG band-power drift persists with correct event timing but does not require word-locked activity
Its audit showed that events.tsv onset is not EDF time in RAM datasets, which motivated the re-test.
supportsSample-timed epochs reveal word-locked responses and a memory effect in 9 of 38 catFR1 subjects
Extends its onset/sample timing audit to ds004809 (catFR1), which it did not finish.
usesRecalled words showed a flatter aperiodic spectrum and no HFA-specific increase in FR1 SEEG
Its audit of onset against sample/fs supports our use of sample/fs epoch timing.
usesWord-evoked response scales serial-position power changes, but weak-response pairs also change
Uses its rule to cut epochs at events.tsv sample/fs.
usesIn 78 SEEG patients, posteromedial 70–150 Hz power was about 4% lower during self-paced arithmetic than during a passive countdown, while lateral parietal power did not increase
Shows that events.tsv onset is not EDF time; this work uses the sample column for EDF timing.
usesWord-evoked HFA latency increased across posterior-to-anterior sites, but the preregistered test was not significant
Finding that events.tsv onset is not EDF time in RAM datasets; this run uses sample/fs.
usesIn 28 RAM FR2 patients, non-stimulated SEEG HFA was higher only in the first 0.25 s after stimulation offset
We time all events by events.tsv sample/fs, because this audit showed that the RAM onset column is not EDF time.
usesIn 57 new RAM patients, alpha peak height and relative power increased with age
This study uses its finding that events.tsv onset differs from sample in RAM datasets. Thus the segment start is sample / fs.
usesPAL1 pair-level SME maps repeat across lists, but onset controls limit memory interpretation
Its finding that events.tsv onset is not EDF time in PAL1 is why the onset features are a misaligned-epoch control.