Research · Methodology

Methods, written down.

Dataset construction, annotation, versioning, limitations, and reproducibility — documented before we ask anyone to trust the artifacts.

Selection criteria, sourcing, and construction steps for each public dataset.

Dual-track annotation: dense-tag and structured bilingual, with inter-annotator notes.

Semantic versions with published checksums; no silent edits after release.

Every release carries an explicit limitations section. Known gaps are stated, not discovered by users.

The full document lives on the lab site: /lab/methodology.