Random sample

A random sample step keeps a deterministic random subset of the input. Geometry and attributes pass through. The sample is stable across reads of the same recipe for the same source because it hashes each feature id with a persisted seed (not Postgres random()).

Inputs

• Source layer.
• Mode: - percentage — keep roughly value percent of rows (greater than 0, at most 100). Actual count may vary slightly because each row is accepted or rejected by a hash threshold. - count — keep exactly value rows (positive integer, soft max 100,000).
• Seed. Nonzero integer stored on the recipe. Changing the seed reshuffles which rows are kept; the same seed repeats the same sample.

Output

A subset of input rows, same schema. Order is not meaningful except that count mode orders by the hash before applying LIMIT.

How to use it on its own

1. New-item wizard → Derived layer.
2. Pick a Source.
3. Add a Random sample step.
4. Choose percentage or count, the value, and a seed.
5. Save.

From a data layer Analyze tab

On a data layer's Analyze tab, Random sample for this layer opens mode, value, and seed controls (with a client-side reshuffle) and creates a private derived layer with a single random-sample step (same PostGIS path as above). Expect a short server job, not an instant browser result. Changing the seed before create reshuffles which rows are kept; the saved seed keeps the sample stable across later reads.

Examples

• Ten percent QA sample. Mode: percentage. Value: 10. Seed: any stable integer.
• Exactly 500 inspection sites. Mode: count. Value: 500.

Notes

• Deterministic on purpose. Re-opening the derived layer does not roll a new sample unless you change the seed.
• Not stratified sampling. There is no group-by or stratum control on this step.
• Pair with Filter first when the sample should only come from a subset.