"""add_signal.py: your turn, add a feedback SIGNAL beyond the thumbs up/down.

Explicit thumbs are great but rare. Real systems also mine IMPLICIT signals from behaviour. Add one:

  1. edited_answer: if the user edited the agent's answer before sending it, the edit is a correction
     (treat like a thumbs-down + correction). Capture the edited text.
  2. resolved: did the user's session end successfully (no repeat question, no escalation)? A weak
     positive signal.
  3. regenerate: the user clicked "try again" -> a weak negative signal on the first answer.

Steps:
  1. Add your signal to a record (extend the shape in feedback_log.log_interaction).
  2. Teach curate.to_eval_cases / to_finetune_examples to use it (for example, an edited_answer
     becomes a correction).
  3. Add a check, then run it.

Worked curation (thumbs up/down + correction) is in ../solution/curate.py. Remember the M14 rule:
redact PII and respect consent before storing or training on anything.
"""

print(__doc__)