Datasets
Owned datasets, immutable definitions, evaluated versions, and capture runs.
A dataset is a user-owned tabular collection captured from a connection (or entered directly). Charts bind to datasets — live or as pinned snapshots.
Definitions and versions
A dataset has an immutable definition (its schema: ordered, typed fields) and a sequence of evaluated versions. Each capture run evaluates the current definition against the source and records a version — so a chart can always reference the exact version it was built against.
Capture runs
A capture run reads from the source and stores the result as a new version. Runs can be triggered manually or by an authorized schedule. Source reads are deduplicated (identical concurrent requests share one read) and bounded per connection.
Field types
Fields are strongly typed — dimensions and measures alike. The type system is what lets chart validation reject a bad mapping before render time: a category channel will not silently accept a date, and a measure will not accept text.
Live data behavior
Charts bound live to a dataset refresh in two ways:
- Manual refresh — explicit, always available.
- Schedules — authorized refresh intervals on the binding.
If a source read fails during a refresh, the prior capture is retained and the failure is surfaced; a chart never silently drops to empty because its source hiccupped. See chart bindings.