Scope the collection
Choose the tables and columns your Data Collection includes. Scope is what makes the collection grounded and answerable, and it is the input the catalog build reads.
Audience: builders
Open the Selection step
Setup → Selection is the middle step in the three-step workflow. The page header reads "Selection" (this is the user-facing name for the scope step). The step needs at least one attached source before it shows anything to pick from.
Pick what is in scope
Search across every attached source
The search input drives a smart-search across every table and column from every attached source. Try the phrases your team actually uses ("ap reconciliation", "vendor master", "GL close"). The search expands domain terms into the tokens that appear in your schemas, so a finance phrase fans out into invoice, vendor, payment, GL, payable.
The expansion strip surfaces what the search is actually looking for. If you want to search only the literal text you typed, click Use literal query instead.
Or browse by source
Clear the search to switch into the source-grouped browse tree. Each attached source collapses to its own group, so you can scan its tables without leaving the page. Search narrows. Browse stays the discovery path when you do not know what to ask for yet.
Add tables
Click a row to add a table to your selection. The relation moves to the right rail. Add as many as belong in the collection. Resist the temptation to add everything: a tighter scope produces a tighter catalog and faster, more accurate answers.
Trim columns
Open a selected table on the right rail to edit its column list. Toggle individual columns, or use All columns to keep everything and None to clear. The column editor opens as a drawer with type-ahead search.
Use Refine columns when the column list is long
Once you have selected some tables, the right rail offers Refine columns. It asks our AI to look at the columns under your current selection and propose which ones to keep and which to drop. You review the suggestions one at a time, accept the ones you want, and dismiss the rest. Manual edits and AI suggestions write through the same path, so you can mix them freely.
Capture Business context
Selection has an optional Business context disclosure below the picker. Use it for terms, edge cases, and rules our AI should remember on every question. Examples that work well:
- "Customer means a paying account, never a trial."
- "Renewals follow the contract anniversary, not the first invoice."
- "Treat tier = 'demo' as out of scope."
Business context is read on every question your collection answers and on every catalog build. Keep it tight. For one-off notes about the next build only, use Notes for this run on the Catalog step instead.
Scope tightens the catalog. Grounded answers depend on a clear selection. Twenty tables that matter beat two hundred tables that mostly do not, because the catalog build only reasons about what you selected.
Save and continue
The Selection step shows your unsaved changes inline. The continue button reads Save and continue when there are unsaved changes and Continue to catalog when everything is saved. The platform persists the selection before it moves you on to the catalog build.
Re-scope later
You can return to Selection at any time. Adding tables, dropping tables, or trimming columns marks the selection dirty. The next catalog build picks up the change automatically when you queue it from the Catalog step. Existing answers stay live against the current catalog until you publish a new build.
File uploads count too. Files you uploaded under Sources (CSV, Excel, Parquet, and the rest) appear in Selection alongside warehouse tables. Pick the columns you care about the same way.