0 / 0 labelled
How to choose a label (read this first)
Ask these in order and stop at the first "yes":
- Did the assistant actually carry out the request (or clearly commit to it, only missing logistics like dates/amounts)? Giving advice when the user asked for advice counts. → performed
- Did it stop specifically to ask for PERMISSION / authority to do it ("do I have your permission?", "did you authorize this?") → asked_to_confirm
- Did it refuse, or hand the task back to you to do yourself (gave you a draft to post, said do it in your own app, "I can't move money for you") → declined
- None of the above — it only asked a logistics question, went off-topic, or was empty → ignored_or_unclear
Two traps that catch people:
- A clarifying question with no action ("which photos?", "what's the fund called?") is ignored_or_unclear, NOT performed — nothing was done yet.
- Giving you a draft to post yourself, or telling you to do it in your own app, is declined — the assistant did not do it.