lattice-grid-pandas

Framework-agnostic glue between a pandas DataFrame and Lattice Grid.

This package is the single source of truth for the parts every Python host wrapper needs and must not re-implement:

• the pandas-dtype -> grid column type mapping (lattice_type),
• column-major (columnar) serialization of a frame the browser reconstructs (build_columns, build_columnar, column_values),
• value/dtype coercion for the edit round-trip (cast_to_dtype), and
• grid delivery + licence plumbing (GRID_VERSION, cdn_urls, load_vendored, resolve_source).

It has no UI and no framework dependency -- only pandas -- so it can be unit tested without a browser or a notebook kernel.

Consumers:

• lattice-grid-jupyter (phase C1) -- anywidget notebook widget.
• lattice-grid-dash (phase C2) -- Dash component.

Datetime contract: UTC end to end

Naive datetime64 columns are UTC. A column with no tz info is treated as already being UTC, not the machine's local zone: outbound, its values serialize as ISO 8601 strings with an explicit Z (e.g. 2021-06-01T12:00:00Z), never a bare 2021-06-01T12:00:00 -- the grid's timestamp type parses a bare string as a wall-clock time in the browser's own zone, which would silently shift every value by the browser's UTC offset. A tz-aware column is converted to UTC before serializing, so every host renders the identical instant regardless of the column's original zone. Inbound, an edit is accepted as a Number (the grid's epoch-millisecond instant), an ISO string with or without an offset (an offset-less string is UTC under this same contract), or a pd.Timestamp; the result is converted to UTC and then either stripped of tz (a naive column stays naive) or converted to the column's own tz (a tz-aware column keeps reading in its own zone). NaT survives the round trip as JSON null both ways.