You’d like to see what Crunch can do before you connect your website or export anything from work. You don’t have to wait — try it on a toy dataset: a free, public dataset made for practice and learning, like store sales, bike rentals, flight delays or house prices. Thousands can be downloaded, and many come with a licence that lets anyone use them. Download one, upload it as a file, and ask it questions the way you would your own data.
The one thing to get right is the licence. A toy dataset being easy to download doesn’t mean you’re allowed to use it. This page shows where to look, which licences are fine, and which to pass on.
What you’ll need
- A BumbleB account. The Free plan is enough — sign up if you haven’t.
- A toy dataset you’re allowed to use. See Check the licence below before you download anything.
Where to find a toy dataset
- Kaggle Datasets — the largest collection, on every topic. You need a free Kaggle account to download. Each dataset’s page shows its licence in the License field.
- UCI Machine Learning Repository — older, well-documented datasets, often small and tidy. The licence is listed on each dataset’s page.
- Government open-data portals — such as data.gov (US), data.gov.uk and data.gov.in. Most datasets carry an open government licence, shown on the dataset’s page.
- Our World in Data — tidy country-by-year tables on health, energy, the economy and more.
- Google Dataset Search — searches all of the above and more at once. It points you to the source, so check the licence there.
Tip: pick a toy dataset that looks like the questions you’ll really ask. If you run a store, an orders or sales dataset tells you far more about Crunch than a table of flower measurements.
Check the licence
Read the licence on the dataset’s own page every time — two datasets on the same site can carry very different terms.
Usually fine to use:
- Public domain or CC0 — no conditions at all.
- CC BY (Creative Commons Attribution) — use it freely; credit the source if you share what you make from it.
- CC BY-SA, ODbL (Open Database License) and open government licences — free to use, with credit; if you publish a changed copy of the data, share it under the same terms.
Pass on these:
- No licence, “Unknown”, or “Other” with no terms you can read. No stated licence doesn’t mean free to use — it means you haven’t been given permission.
- Non-commercial (anything with NC) if you’re trying Crunch for your work. Evaluating a tool for a business can count as commercial use.
- Competition data on Kaggle. Data from a Kaggle competition is usually limited by the competition’s rules to use in that competition. Look under Datasets, not Competitions.
- Data about real, identifiable people — names, emails, phone numbers, addresses — even when the licence allows it. There’s no need for it to try Crunch, and it’s personal data you’d then be holding.
When you upload, you tick a box confirming you own the data or have the right to upload it. The licence is what gives you that right, and our Terms of Service ask that you only upload what you’re allowed to share. If you aren’t sure, choose another dataset — there are plenty.
Step by step
- Pick a toy dataset from one of the sources above, and check its licence.
- Download it. Kaggle and many portals hand you a
.zip— unzip it first, since Crunch reads the files inside, not the archive. - Choose one file in
.csv,.tsv,.xlsx,.xlsor.parquet, up to 100 MB and 300 columns. If the dataset comes as several files, start with the main one; you can add a second later. - Upload it. In a chat, click the datasets pill in the message box, choose Upload a file…, and follow the steps in Upload an Excel, CSV or Parquet file.
- Tell Crunch what it is. In your first message, say where the data comes from and what one row is — “this is a toy dataset of bike rentals in one city; each row is one day”. The dataset’s page usually explains its columns; pass on anything that isn’t obvious from the names.
- Ask a question in plain words, then the next small one.
How to tell it worked
When loading finishes, the dataset is selected for the conversation, and your first answer leads with the result and shows how it was counted. If the answer reads your question the way you meant it, you’re set — Read an answer explains each part.
Good to know
- Toy datasets can be messy. Missing values, odd codes and columns nobody explains are common. That’s a fair test: tell Crunch what you know, and notice how it handles what it can’t tell.
- Dates may load as text. If a date could be read day-first or month-first, Crunch keeps it as written. Ask about it anyway, or tell Crunch the format in your first message.
- Two files that share a column can be asked together — say, trips and stations sharing a station ID. See Ask one question across two datasets.
- Give credit when you share. If you post a chart or finding made from a CC BY or similar dataset, name the source, as its licence asks.
Related
- Upload an Excel, CSV or Parquet file — the upload steps, limits and versions in full.
- Ask about any dataset — what to tell Crunch up front, with a worked example.
- Solve a problem in a conversation — how to break a big question into small ones.