A dataset is a table Crunch can read — a file you uploaded, your clickstream (the stream of events your site sends), or a table you linked. Before you ask, you tell the conversation which datasets to read, and which version of each. Once it has read them, they stay fixed, so a figure you got this morning doesn’t quietly change because new data arrived at lunch.
What you’ll need
- At least one dataset, or a file to upload — see Upload files.
Pick datasets
- Click the datasets pill (the database icon) in the message box.
- Search datasets and tick the ones you want — or choose Upload a file… to add a new one.
- For a selected dataset, you can choose:
- Use as clickstream — read it as events, which is what funnels, retention and paths need.
- Snapshots — which version to read (see below).
- Clear selection — start the pick again.
Choose a version
- A table you replace with each upload. Crunch keeps each upload as a version. You can read As first read (the default), Newest, or a specific one.
- A table that grows over time. You can read All snapshots, or a chosen set of them.
How to tell it worked
Once the conversation has read its data, the pill shows In play (fixed for this conversation). From then on, its datasets and versions are fixed, and answers don’t shift under you.
Good to know
- Heads up: Because the pick is fixed once read, ask about new data in a new conversation.
- Every upload, and every dataset a chat makes, lives in your own workspace.
Related
- Upload files — add a dataset from a file.
- Your own data lake — read tables where they already live.
- Data home and lineage — browse every dataset and where it came from.