Expose a dataset to dashboards
Developer preview
Publish a table schema once, then bind its fields in the no-code plot editor.
Export a validated table
Use huggingbody.dataset_contracts.write_dataset(output_dir, filename, definition, rows, producer=...) in a file module and return the resulting descriptor on a declared output port.
Declare a namespaced semantic contract, title, explicit time base and 1–24 columns with at least one numeric column. Numeric columns require units; use 1 for dimensionless values. Declare nullability, and use None for missing values rather than silently coercing them.
Publish, bind and import
- In Workspace → Dashboards → Dataset library, select the output JSON or standalone schema.
- Publish the dataset schema. Only the definition is registered; patient rows are not published in the library.
- Choose the published dataset in a custom plot and bind its numeric and categorical fields.
- In the assigned patient dashboard, import an output against the exact pinned schema and patient session, then explicitly load it.
Dataset interchange limits
Producer metadata is an author declaration. Structural validation does not prove execution, verify scientific accuracy or attest to an algorithm. Full-size originals remain separate SDK artifacts.
| Boundary | Limit or rule |
|---|---|
| Envelope | huggingbody/tabular-dataset@1.0.0 |
| Columns | 1–24, including at least one numeric column |
| Rows | 1–5,000 exported rows |
| Normalized JSON | 400,000 bytes |
| Cells | Exactly the declared keys and types; finite numbers and declared nulls |
| Sampling | Explicit description when the original row count is larger; no automatic resampling |