Skip to content

Qualifying future integrations

Website discovery improvement (October 2026): search 14 ecosystem entries by research role and repository integration status, and create a locally generated provenance checklist. This improves findability without claiming new adapters or changing the published 1.2.0 API.

This roadmap is not a commitment to new dependencies or a claim of stable compatibility.

Priority Integration Qualification before adding a supported API
EyeLink SyeLink 2.0 SessionData handoff Synthetic genuine-dataclass preview implemented; requires cross-parser real-device ASC validation, display/segment/clock audits, and lossless event/mask provenance
EyeLink EDF vs ASC independent sources Compare EyeLinkIO binary EDF with vendor edf2asc and SyeLink results on consent-safe real recordings before a stable importer
1 pandas / Polars / PyArrow Example exists; qualify null/dtype parity, large-file behavior and version/platform matrix
2 pymovements sample and event handoff Pinned 0.28.0 synthetic-object preview implemented; still requires real-device fixtures, upstream version compatibility, trial/time mapping and detector provenance to qualify a stable API
3 Pandera optional schema checks Explicit nullable rules, strict failure contracts and no silent coercion or missing-row filtering
4 Eye-Tracking-BIDS reading Read-only JSON/TSV.GZ shape preview implemented; still requires full version-pinned BIDS conformance, sidecar inheritance, recorded-eye and event validations
5 PyNeon mobile recordings Native/cloud format fixtures, synchronization and coordinate provenance
6 eyekit/PupEyes Real reading/pupil fixtures, separate raw/derived modalities and clock contracts
7 scikit-fda/FDApy independent references Matched estimands, support, simulation-law and numerical tolerance audits

Optional Python plugin architecture

Later, versioned Python entry points could register independent data adapters. Entry points should declare the external package/version, source format, coordinate and time metadata, and explicit fail-closed rejection. They must not monkey-patch root APIs, silently install heavy dependencies, or imply valid confidence bands. No plugin registry or new package extra is introduced here.

A compatibility smoke test or an upstream release is not scientific inference qualification. The 1.2.0 API and open W3 scientific exclusions remain unchanged.

API stability · Research evidence · Tabular example