NeuroDoc

NeuroDoc turns heterogeneous public EEG datasets into reusable benchmark units through rulebook-guided task documents and executable kernels.

Download standardized downstream datasets.

Using NeuroDoc, we created annotated, standardized downstream datasets ready for direct evaluation in EEG-Arena. ModelScope now hosts 17 datasets, 47 tasks, and 14 model-specific cache packages.

Coverage: EEG-Arena evaluates 23 datasets in total. Six remain benchmark-only because their licenses do not permit secondary redistribution; obtain those datasets from their original providers.

Download Data
14 model packages · 17 datasets · 47 tasks available on ModelScope.
ModelScope Source only Upload pending

Turn heterogeneous evidence into executable benchmark entries.

NeuroDoc builds context from dataset-local evidence and academic sources, proposes evidence-bounded tasks, and implements their requirements in a sandboxed coding loop. Each release pairs a human-readable task document with an executable, dataset-specific kernel.

Read the methodology
NeuroDoc workflow from dataset evidence and academic sources to a synchronized task document and dataset-specific task kernel.
Figure 2. NeuroDoc generation workflow. Dataset-local evidence and academic sources are synthesized into a synchronized task document and an executable, dataset-specific task kernel.

Review, correct, and continuously upgrade every entry.

NeuroAudit guides community review through a shared rulebook. Completed entries are archived, while unsupported claims, document conflicts, and kernel mismatches return to the NeuroDoc upgrade pipeline for evidence analysis and reconciliation.

Explore the review pipeline
NeuroAudit-assisted community review and NeuroDoc upgrade pipeline for reviewing documents, resolving conflicts, and updating task kernels.
Figure 3. NeuroAudit review and upgrade workflow. Community feedback is checked against the rulebook and routed back to NeuroDoc when document or kernel revisions are required.

Data Review

We are actively testing the data-review workflow and integrating it into the NeuroInfra library. We plan to officially release NeuroInfra in early October—stay tuned and try it when it launches.

Visit NeuroInfra