Start with four records
Once a model is connected, start a new chat and paste the prompt below. Everything it needs is in the message: no files, no browsing, no other apps.
Use Python to work with these four sample records: A: 12 B: 7 C: 12 D: 4 Store the records in the live Python session. Calculate their total and count, show the calculation and results, and keep the data available for a follow-up. Do not read or modify files, browse, or operate other applications.
Check the answer
You should get a total of 35 across 4 records. Look at the numbers the model actually used and the calculation it ran, not just its summary of what it did.
It is small enough to check in your head, which is the point: you know exactly what a right answer looks like.
Try a follow-up
Check that the sample records are still available in the live Python session. Add a new record E with value 5, then show the updated total and count. If the earlier state is unavailable, tell me before reconstructing it.
With E added once, the total becomes 40 and the count 5. Nothing had to be typed in again: the records were still sitting in the session. If the session had restarted in between, the model should say so before rebuilding anything.
Try your own materials
Next, pick a task with a clear result and a handful of sample files. Say what may change and what must stay untouched, for example “write a new report file and leave the originals alone.”
The example workflows take the same idea into data cleaning, research, and desktop work. Check each result the way that suits it: compare the values, reopen the file, or look at the app.