Skip to content
Lattice Lattice

Datasets and charts

Analytical .dataset packages store facts as Parquet and query them locally with DuckDB. Results cross the native boundary as bounded Arrow IPC rather than large JSON object arrays.

Open a dataset

Section titled “Open a dataset”

Select a .dataset package and choose a panel:

  • Preview opens the bounded result in Perspective.
  • Chart builds a Vega-Lite view over the same result.
  • Profile runs DuckDB SUMMARIZE for column statistics.
  • Plan shows DuckDB EXPLAIN output.
  • Map plots rows when longitude and latitude columns are detected.

Changing panels cancels work that no longer matters. A failed specialized viewer falls back to an inspectable Arrow summary instead of hiding the data.

Create and import from the CLI

Section titled “Create and import from the CLI”
Terminal window
lattice dataset create Data/Events.dataset --title "Events"
lattice dataset import-csv Data/Events.dataset --csv Data/events.csv \
  --partition year=2026 --partition month=07
lattice dataset show Data/Events.dataset

Run workspace-scoped SQL with DuckDB:

Terminal window
lattice query --engine duckdb \
  --sql "select region, sum(revenue) from read_parquet('Data/Orders.dataset/facts/**/*.parquet') group by region"

Dataset annotations live in a separate SQLite overlay so human notes or review state do not rewrite source Parquet facts.

Open a chart resource

Section titled “Open a chart resource”

Vega-Lite .vl.json resources may bind to a dataset query. Open the chart from the tree, a page embed, an interface, or a canvas node. Keep the query bounded and place reusable transformation logic in the dataset or task layer.