Interactive laboratory · observe real results
“Can you automate our expense report?”
A manager wants a department summary from a CSV file. There are eight records, including a repeated ID, a missing amount and a refund. Work through the decisions before running the implementation.
Inspect the eight input records
| ID | Department | Amount (AUD) |
|---|---|---|
| TX001 | Sales | 120.00 |
| TX002 | Sales | 80.00 |
| TX003 | Operations | 45.00 |
| TX003 | Operations | 45.00 |
| TX004 | Operations | — |
| TX005 | Research | 35.00 |
| TX006 | Sales | -20.00 |
| TX007 | Operations | 90.00 |
Ask the questions code cannot decide.
These are role-play decisions, not a live chat. There is no single universal refund or duplicate rule; choose a policy and make it explicit.
Make the agreement testable.
A specification guides the code. Tests check it.
The example implementation is prepared in advance and instantiated from your decisions. This page makes no live AI calls. The prompt shows how to brief your own AI tool; actual Python execution happens in your browser.
Inspect the implementation prompt
Edit the implementation, then run it against the fixture expectations. Changing stakeholder decisions clears confirmation, regenerates the example and invalidates the previous result.
Python output & traceback
First execution loads about 14 MB of Python. Runs stop after 10 seconds. No files are uploaded, source data stays unchanged, and edits are not saved after refreshing.