Four objects
A data source is a connection — Redshift, Athena, S3, RDS, Salesforce, a file upload.
A dataset sits on top of a data source and defines the prepared table: selected fields, joins, filters, calculated fields, and whether the data is held in SPICE or queried directly.
An analysis is the authoring environment where you build visuals. It is editable and private to people you share it with.
A dashboard is a published, read-only snapshot of an analysis, shared with consumers. Editing the analysis does not change the dashboard until you publish again.
That last point trips people up constantly: changes are not live until republished.
Where logic belongs
Calculated fields can be defined at dataset level or analysis level.
Dataset-level fields are inherited by every analysis built on that dataset — define shared business logic there. Analysis-level fields exist only in that analysis.
The same reasoning as published data sources in Power BI or reusable sources in Looker Studio: define once, inherit everywhere.
Editions
Standard covers authoring and dashboards for individual users.
Enterprise adds row-level security, Active Directory / IAM Identity Center integration, VPC connectivity, encryption with customer-managed keys, hourly refresh, embedded analytics with anonymous users, and higher SPICE capacity.
Row-level security is Enterprise-only, which is usually the deciding factor. Pricing is per-user for authors and either per-user or per-session for readers — the session model is what makes broad embedding affordable.