# Nimbus Studio reproduces reference BCI decoding pipelines: a validation study on public benchmarks across protocols and paradigms

Preprint · Sergey Musienko

- Archive id: arxiv:2610.13894
- Published: 2026-10-09
- arXiv: https://arxiv.org/abs/2610.13894
- Citations: 0
- Source URL: https://arxiv.org/abs/2610.13894
- PDF: https://arxiv.org/pdf/2610.13894
- Provenance: arxiv

## Abstract

Nimbus Studio is a visual pipeline platform for brain-computer interface (BCI) experiments: researchers assemble preprocessing, feature extraction, decoding, and evaluation over public benchmark datasets or their own recordings, with cross-validated evaluation protocols (including leave-one-subject-out), BIDS export, and bring-your-own Python model classes. Whether its numbers can be trusted is a fair question for any such platform. We answer it with a parity study: six standard BCI decoding pipelines (CSP+LDA, Riemannian MDM, tangent-space logistic regression through the platform's bring-your-own-code node, P300 xDAWN+LDA, and the deep ShallowConvNet and EEGNet) were evaluated on three public datasets (motor-imagery BNCI-2014-001/004; P300 BNCI-2014-009) under two protocols (per-subject session holdout and pooled leave-one-subject-out cross-validation via the platform's planned-CV driver), once through Nimbus Studio's execution engine and once through independent, hand-written reference implementations built directly on MNE-Python, scikit-learn, pyriemann, and braindecode. Every deterministic pipeline matched the reference exactly (0.00 points maximum difference across all 54 session-holdout subject-runs and all CSP/MDM grouped-CV folds), including pipelines through the bring-your-own-code node and the CV orchestration; the remaining grouped-CV cells agree to floating-point precision (worst case 0.06 points, about one trial per fold). The stochastic deep models agreed within seed-to-seed training variability, estimated with multi-seed reruns on each side. The platform's execution engine therefore computes the same science as the reference stack across protocols, paradigms, and extensibility surfaces. The platform is available at https://studio.nimbusbci.com/, and the complete harness and per-subject results accompany this paper for independent verification.

_Abstract reproduced under CC0-1.0 (arXiv descriptive metadata)._

Topics: [Brain-Computer Interfaces](https://space-kitty.com/topics/brain-computer-interfaces)

---
Space Kitty Archive · https://space-kitty.com · API: https://space-kitty.com/api/v1 · MCP: https://space-kitty.com/mcp · Premium: $0.20/request via x402 (Base Sepolia (testnet)). Metadata: arXiv (CC0), OpenAlex (CC0), Europe PMC (CC BY/CC0 works), ClinicalTrials.gov.