Tutorial¶
This tutorial works through the three operations that define lairs, in order, on a single running example: reading a published corpus from a Personal Data Server (PDS), materializing its records into columnar views, and authoring new records and computing a publish plan. Each chapter builds on the last. Follow them in sequence. Every step is meant to run as written.
The running example is a small English corpus: one expression record holding the sentence The cat sat on the mat., with a part-of-speech annotation layer over it. The same expression and layer reappear in every chapter, so the reading chapter shows you the records that the materializing chapter turns into Arrow tables, and the authoring chapter rebuilds equivalents from scratch.
Prerequisites¶
Install lairs:
The reading chapter loads records over ATProto. Public reads need no
authentication: a PDS serves the com.atproto.repo.listRecords and
com.atproto.repo.getRecord methods to anyone. To follow the reading chapter
against the live network you need an AT-URI whose authority publishes
pub.layers.* records. The chapter also shows how to drive the loader from an
injected client so the example runs without any network at all.
The materializing chapter writes Parquet files and reads them back with pyarrow, which lairs already depends on. The authoring chapter writes to a local store on disk and computes a publish plan offline. It sends nothing to any PDS.
What you will not find here¶
The tutorial is a single guided path, not a catalog. It does not enumerate the load sources, the dataset transforms, or the exporter back ends. For those, read the task-oriented Guides. For the design behind the generated models, the anchor system, and the store, read the Concepts. For per-symbol signatures, read the API reference.