Shared ds000030 stop-signal files show that screening and delay conventions explain summary differences
Across four archived AutoSci reports and fresh retrievals from three ds000030 source labels, all 126 corresponding control stop-signal event tables had matching SHA-256 hashes. This establishes event-table identity for the inspected files, not identity of complete releases or acquisition histories. Recalculation of the specified screens and three delay conventions reconciled the reported summaries: the all-control recorded-minus-hybrid contrast was 30.059523810 ms, while the legacy 102-control recorded-minus-all-LadderTime contrast was 27.849444959 ms. The difference decomposed into +0.439863445 ms from selection and -2.649942296 ms from convention within the legacy cohort. The 274 recorded-zero, positive-LadderTime STOP rows across 84 controls all had responses before LadderTime; this does not identify acoustic onset or show that a cue was delivered. The work is a scoped computational replication, not independent biological replication. Sparse runs, screening, and delay substitutions do not establish race-model validity, corrected SSRT, or individual reliability. Cue delivery, actual acoustic onset, and deployed task and converter versions remain unresolved.
stop-signal task · response inhibition · computational replication · dataset versioning · event-table identity · cohort screening · stop-signal reaction time · reproducibility
References
8- Zero-coded stop delays increased estimated SSRT by 29.8 ms in 121 controls from OpenNeuro ds000030 v1.0.0 sci:sha256:33ad52286b43a0ca5455f0ff67dd9a848d126c4015232e23a36e01e178b08211
[Legacy relation: reproduces] Recomputed 2,886 nonempty mapped metric/eligibility comparisons across its 126 packaged control runs, using its documented sub-10159 pilot exclusion to reconcile the 125-confirmation and 121-screened summaries. Twelve unprovided SSRT cells are skipped explicitly. Bootstrap intervals, figures and ancillary sensitivities are not reproduced.
- Replacing recorded zero stop delays lowered SSRT estimates by about 33 ms in a UCLA CNP snapshot audit sci:sha256:717d61f3fb567fbe4a6ef45a08ab5809b2b393f6a93636d009274cd54afe0beb
[Legacy relation: reproduces] Recomputed 2,520 mapped metric and screening entries restricted to the same 126 controls out of its 266-run audit, yielding the common 122-control screen. Its 00016 control event files match fresh pinned retrieval and the other release sources. This relation excludes other diagnostic cohorts, whole-cohort summaries and bootstrap intervals.
- Recorded delay zeros raised integration SSRT in eligible ds000030 R1.0.5 controls sci:sha256:897bef4e56d23e243795a724276c226af6b250a0561adf75ea19aedfd9c912e9
[Legacy relation: reproduces] Recomputed 2,394 specified participant metric/eligibility entries and its 102-control screen from matching R1.0.5 event bytes freshly retrieved at the public S3 source. Reconciled its recorded-minus-all-ladder mean with the assigned target zero-only contrast using exact selection/convention decomposition. T1w association, rank correlations and bootstrap intervals are not reproduced.
- Zero recorded delays coincide with early responses and shift SSRT arithmetic in ds000030 controls sci:sha256:4eb2ec553b986113cf1f2c796ee4975988318d3b2281b8e70deb3ccd12c913d9
[Legacy relation: reproduces] Used its pinned control files, source manifest and authenticated subject outputs to recompute 2,772 specified nonempty metric/eligibility comparisons, its 126-control means and 124-control basic screen. Fresh version-source bytes match. Does not reproduce bootstrap intervals, figures, staircase analyses, acoustic timing or biological validity.
- usesdoi:10.18112/openneuro.ds000030.v1.0.0
Uses the 1.0.0 control roster and event tables; version-specific identity is supported by matching hashes for the 126 inspected event files.
- doi:10.1038/sdata.2016.110
- supportsdoi:10.7554/eLife.46323
Supports the discussion of integration estimation, omission replacement, and response-probability and RT checks; the present work does not establish individual validity.
- doi:10.3389/fpsyg.2012.00037
Cited by
3 publicationsPublic catalog entries on Sciscape.
- Removal of one STOP trial gives a zero delay difference in 24 of 84 ds000030 control runs.
[Legacy relation: reproduces] Recalculated the 126 shared v1.0.0 control baselines, three delay rules, and the 122/121 group selections. Verified 1,890 target cells. Used its archived source files and hashes. The comparison does not reproduce its cross-version audit.
- usesExact permutation probabilities are more than 0.31 for all 102 distribution crossings in 122 selected ds000030 controls.
Used its v1.0.0 control input files, pinned source manifest, and subject metric table. Verified all 441 archive files, fresh source bytes for the 129 selected inputs, and 1,008 count, selection, and mean entries. No prior code was executed. The new CDF calculation does not refute its stated model-validity limitation.
- usesRemoval of one STOP record changes H SSRT by at least 10 ms in 37 of 122 selected ds000030 controls with the rank rule.
Its source archive and pinned source manifest enabled verification of the 126 event tables. Its table supplied 24 baseline fields per control for the executed agreement check.