The explorer's boundary-layer sidecar: the registry joined with the release's spatial table

metadata/spatial_layers.csv is the registry of the boundary layers (Erin's sheet: one row per drawable layer with its PMTiles group, default symbology, filter expression and provenance), and the archives at {pmtiles_base}{dataset_group}.pmtiles carry the features. The explorer must not hard-code that list nor fetch the CSV from GitHub at runtime (plan 2026-08-31 D23), so each release ships spatial_layers.json: the registry verbatim plus what only the release knows — each layer's feature count, bbox, its sorted distinct names when there are at most names_max (the by-name palette, D24; NULL above that, and the app falls back to an id-hash palette), and n_memberships (distinct root samples in sample_spatial, so the Regions lens can list exactly the layers that can summarize something).

Usage

build_spatial_layers(
  con,
  registry_csv,
  version,
  pmtiles_base,
  built = NULL,
  names_max = 200
)

Arguments

con

DuckDB connection holding spatial (and, if built, sample_spatial).

registry_csv

Path to metadata/spatial_layers.csv.

version

Release version string, stamped into the sidecar.

pmtiles_base

URL prefix of the PMTiles archives (source-layer = dataset_group).

built

When the archives were last built (the ingest_spatial manifest's mtime) — version skew between releases and archives is accepted but must be visible.

names_max

Above this many distinct names a layer's names is NULL.

Value

A list ready for jsonlite::write_json(auto_unbox = TRUE): version, pmtiles_base, built, and layers[] with id (the registry dataset_id), group, name (the human layer name), source, geom, filter (the registry expression verbatim, as parsed JSON), the symbology defaults, name_field, description, attribution, n_features, bbox, names, n_memberships.