Domain-Agnostic Dataset Workflow¶

English · Русский

This compatibility page retains existing section links. Current instructions now have a single owner for each task, rather than a second copy of commands.

Mental Model¶

Profile safe metadata, review a DatasetSpec, generate new values and validate the result. See Profiles And Specs.

Folder Layout¶

Use one CSV per table. File stems identify entities. See Related Tables.

Commands¶

Follow the reviewed profile/spec/generate/validate flow. Commands live in one canonical reference. See CLI Workflows.

Large Inputs And Profile Cache¶

Profiling and inference are resource-bounded. Cache only safe metadata; use --no-cache when a fresh profile is needed. See Configuration.

Large Trino Tables¶

Keep profiling aggregate-only, allowlisted and bounded. Never download source rows for generation. See Profile Through Trino.

What Gets Written¶

Keep the specification, manifest and validation reports with the generated data. See Review The Output.

How Relationships Are Preserved¶

Generated child references use generated parent keys. Review inferred links and ambiguous identifiers. See Relational Synthesis Contract.

How Constraints Are Preserved¶

Typed, executable rules enforce formulas, temporal order and cross-table conditions. See Business Rules.

Safety Notes¶

Generation uses the specification and seed, never a pool of source rows. Privacy review remains necessary. See Safety Model.