chatbot1-11  Exercise 2: Your Own Golden Conversation
=======================================================

ex2.py (save in the project root):

  import asyncio, json, pathlib, tempfile
  from bot.golden import check

  CASE = {
      "name": "my first golden conversation",
      "script": ["We open at nine.", {"error": "ServerError"}, "Back again."],
      "options": {"safety": {"disclosure": None}},
      "turns": [
          {"say": "When do you open?"},
          {"say": "And on Sundays?"},
          {"say": "Are you there?"},
          {"say": "/help"},
      ],
  }

  async def main():
      with tempfile.TemporaryDirectory() as d:
          path = pathlib.Path(d) / "mine.json"
          path.write_text(json.dumps(CASE, indent=2), encoding="utf8")
          print("1. First run, no expectations written yet (every turn 'differs'), with update=True to record them:")
          await check(path, update=True)
          for turn in json.loads(path.read_text(encoding="utf8"))["turns"]:
              print(f"      {turn['say']!r:<26} -> {turn['reply'][:60]!r}")
          print("2. Check again with no changes:", await check(path) or "passes")
          print("3. Someone changes what the bot says for /help (simulated by editing the expectation):")
          data = json.loads(path.read_text(encoding="utf8"))
          data["turns"][3]["reply"] = "Commands: just ask me anything!"
          path.write_text(json.dumps(data), encoding="utf8")
          for p in await check(path):
              print("     ", p.replace("\n", "\n     "))

  asyncio.run(main())

Run with:  py ex2.py

Output:

  1. First run, no expectations written yet (every turn 'differs'), with update=True to record them:
        'When do you open?'        -> 'We open at nine.'
        'And on Sundays?'          -> '<raised ServerError>'
        'Are you there?'           -> 'Back again.'
        '/help'                    -> 'Commands:\n/remember <text>   ask me to remember something ab'
  2. Check again with no changes: passes
  3. Someone changes what the bot says for /help (simulated by editing the expectation):
        mine.json turn 4 ('/help'):
           expected: 'Commands: just ask me anything!'
           got:      'Commands:\n/remember <text>   ask me to remember something about you\n/facts             list what I remember\n/forget <number>   forget one thing (or /forget all)\n/reset             clear this conversation\n/export            show everything I have stored for you as JSON\n/privacy           what I store and for how long\n/deleteme          erase your chats and notes'

What you must do before committing a file produced by --update: READ every changed expectation, and decide for each one whether the new text is what you want the bot to say. --update records what the code does now, so it will record a bug as readily as a deliberate change. If a diff shows a reply you did not mean to change, fix the code, not the golden file.
Note the second turn: the scripted ServerError reaches the caller (the bot does not catch provider errors; the channel adapters do), and the golden player records it as "<raised ServerError>" so that the transcript still has a line for that turn.

WHY THIS WORKS AS AN ANSWER
---------------------------
The case lists the model's scripted replies (a text, an error, a text), the settings, and the turns. check() plays the turns through a fresh Bot with a fake clock and compares each reply with the expected one. With update=True it writes the actual replies into the file. A later change to any reply (here a hand edit to the /help text) produces a message naming the turn and showing the expected and actual text,
which is exactly what a reviewer needs to decide whether the change was intended.