chatbot1-8  Exercise 2: What Buffering Costs
==============================================

ex2.py (save in the project root):

  import asyncio, time
  from bot.core import Bot
  from bot.persona import Persona
  from bot.providers import FakeProvider

  PERSONA = Persona("Rowan", "a library bot.", "brief", "library questions", "I can only help with library questions.")
  REPLY = " ".join(f"word{i}" for i in range(1, 21))              # 20 words, 0.05 s between pieces


  async def run(live):
      bot = Bot(FakeProvider([REPLY], chunk_delay=0.05), persona=PERSONA, stream_live=live)
      start = time.perf_counter()
      first = None
      count = 0
      async for event in bot.stream_message("s", "hello"):
          count += 1
          if first is None:
              first = time.perf_counter() - start
      total = time.perf_counter() - start
      return first, total, count


  async def main():
      print(f"{'mode':<9} {'first text on screen':>21} {'reply complete':>15} {'events':>7}")
      for live in (True, False):
          first, total, count = await run(live)
          print(f"{'live' if live else 'buffered':<9} {first:>20.2f}s {total:>14.2f}s {count:>7}")

  asyncio.run(main())

Run with:  py ex2.py

Output (timings vary by a few hundredths of a second between runs):

  mode       first text on screen  reply complete  events
  live                      0.06s           1.24s      21
  buffered                  1.25s           1.25s       1

Which mode for a children's library bot (judgement): buffered. If a model says something unsuitable, the guards and moderation can only act once they have the whole reply; in live mode the child's screen would already have shown it, even if it is then replaced. Waiting a second or two for a reply is a small price for never showing unchecked text to that audience.
For a low-stakes bot where speed matters more, live is the better experience.

WHY THIS WORKS AS AN ANSWER
---------------------------
Live mode sends one event per word (20 deltas plus the final: 21 events), so the first word is visible after one pause of 0.05 s and the reply completes after 20 pauses, about 1.2 s. Buffered mode must wait for the complete reply before it can run the checks, so it sends a single event at about the same total time but nothing is visible before then. The checks, the stored text and the final
text are identical in both modes; only what the user can see while waiting differs.