# Survey a collection

> Let Clusters propose what a collection is about and the fields you can filter its books by.

A survey reads a sample of each document and proposes two things for the collection note: its `context` (what the collection is about) and, for books, each book's **fields**. For the nonfiction lens these include the author's role and whether they write from their own experience. Each proposal comes with a quote behind it.

Fields are what the search filters use.

## Steps

1. Run **Clusters: Survey a collection…** and pick the collection. Or select **Survey…** in the collection's dialog.
2. Read the dry run: the documents, the size of what would be sent, and the number of calls. No model has been called.
3. Set a spend limit if you want one, and run it.
4. Read what it would change in the collection note, shown as old → new.
5. Select **Write** to accept. Nothing is written until you do.

A collection of notes or Base rows gets `context` only, since its fields are its own properties.

If you close the dialog, the result stays reachable from **Clusters: Open jobs**, under **Review**.

## Related

- [Search a collection](search.md)
- [Undo a run](undo-a-run.md)