edfcore: Random-Access EDF/BDF Infrastructure for Large Biosignal Models

edfcore is a dependency-free TypeScript reader for EDF, EDF+, BDF, and BDF+ built around random access, explicit file semantics, and verifiable decoding.

The live demo will use several real and adversarial physiological recordings to exercise the parts of EDF ingestion that become consequential at scale. From a multi-hour polysomnography recording, attendees will request a short window deep inside the study and observe the byte range actually read, without loading the full recording. A discontinuous EDF+D recording will demonstrate gap-aware temporal access, where intervals spanning recording breaks remain structurally discontinuous. BDF examples will exercise signed 24-bit decoding, and EDF+ recordings will expose annotations on their exact event timeline.

The second part of the demo focuses on failure behavior. Malformed or contradictory recordings will be passed through edfcore's validation layer, which reports structured diagnostics tied to the relevant EDF field and, where applicable, its byte offset. Decoded physical samples can also be compared against independently generated pyEDFlib reference values used by the project's numerical validation suite.

For large-scale neural modeling, the file reader is part of the data pipeline. It determines how timestamps, physical scaling, discontinuities, per-channel sampling rates, and malformed metadata are represented. edfcore makes that behavior explicit and consistent across browser, Node, local-file, and other range-backed sources.

Its role is to provide a trustworthy and scalable boundary between an EDF/BDF recording and the model or analysis system that consumes it.

edf2csv: Faithful Conversion of Heterogeneous Biosignal Recordings

edf2csv is a local command-line converter for EDF, EDF+, BDF, and BDF+ recordings.

The live demo will focus on cases where straightforward conversion is not actually straightforward. Attendees will work with recordings containing channels sampled at different rates, EDF+D discontinuities, annotations, and 24-bit BDF signals. Recording gaps remain visible in the exported time axis, physical units are retained, and annotations and channel metadata are exported separately from the signal values.

A mixed-rate recording will be used to show the difference between preserving the recorded samples and implicitly creating new ones during conversion. The same recording will then be exported in wide and long layouts, with attendees able to select specific channels or time windows and inspect the corresponding output directly in CSV form.

The demo will also show batch conversion across multiple recordings, validation behavior, compressed output, input checksums, and machine-readable inspection. These features make the conversion step usable in larger data pipelines while keeping the relationship between the original recording and the exported values auditable.