# Use-case presets "Realistic" means different things to different jobs, so name the job: ```python misata.generate_from_schema(schema, preset="demo") ``` ```bash misata generate --config misata.yaml --preset load ``` or in the schema itself: `preset: ml` in YAML, `"__preset__": "ml"` in a dict. | Preset | For | What it does | |---|---|---| | `demo` | sales demos, dashboards, screenshots | shifts declared date ranges so the latest ends today; coherence repairs on; no deliberate mess | | `test` | unit and integration tests | caps every table at 200 rows; seed 0 if none is set; no mess | | `load` | load and performance tests | multiplies row counts (x10; `apply_preset(..., scale=)`); turns expensive coherence passes off; keeps popularity skew | | `ml` | ML pipelines and feature code | about 3% nulls, 1% outliers, 0.5% typos, 0.2% duplicates, with keys protected; correlations inferred from column names | | `eval` | agent and text-to-SQL evaluation | `demo` dates plus 2% nulls, typos and duplicates | A preset only fills in what the schema did not say: a declared `noise` block, row count or realism setting is kept. `demo` and `eval` leave dates alone when curves or processes carry their own, because moving the columns without them would break the declaration. `misata.apply_preset(schema, name)` returns the adjusted copy if you want to inspect it before generating. See [A demo tenant that tells a story](../usecases/demo-tenant.md) for `preset: demo` in a complete recipe.