chatbot1-7  Exercise 1: Cost by Hand and by Report
====================================================

The prices are EXAMPLES (per million tokens), not any vendor's. Copy real ones from your provider's pricing page and note the date.

ex1.py (save in the project root):

  import asyncio
  from bot.core import Bot
  from bot.costs import Price, cost_of
  from bot.limits import Limits
  from bot.providers import FakeProvider, Reply, Usage

  # EXAMPLE prices (per million tokens), not any vendor's
  PRICES = {"example-small": Price(1.0, 5.0), "example-large": Price(4.0, 20.0)}
  CALLS = [("example-small", 1200, 300), ("example-large", 3000, 800), ("example-small", 800, 150)]

  by_hand = 0.0
  for model, tin, tout in CALLS:
      price = PRICES[model]
      cost = tin / 1e6 * price.input_per_mtok + tout / 1e6 * price.output_per_mtok
      by_hand += cost
      print(f"{model:<14} {tin:>5} in {tout:>4} out  = {cost:.6f}")
  print(f"by hand total: {by_hand:.6f}")

  async def main():
      fake = FakeProvider([Reply("ok", Usage(i, o), model=m) for m, i, o in CALLS])
      bot = Bot(fake, limits=Limits(burst=100, prices=PRICES))
      for n in range(3):
          await bot.handle("s", f"question {n}")
      print()
      print(bot.report())
      print(f"report total : {sum(r[5] for r in bot.store.usage_report(0)):.6f}")
  asyncio.run(main())

Run with:  py ex1.py

Output:

  example-small   1200 in  300 out  = 0.002700
  example-large   3000 in  800 out  = 0.028000
  example-small    800 in  150 out  = 0.001550
  by hand total: 0.032250

  served by     model          calls   in tok  out tok      cost
  fake          example-large      1     3000      800  0.028000
  fake          example-small      2     2000      450  0.004250
  report total : 0.032250

WHY THIS WORKS AS AN ANSWER
---------------------------
The cost of a call is input tokens x input price + output tokens x output price, each per million. By hand the three calls cost 0.002700, 0.028000 and 0.001550, total 0.032250. The Bot recorded each call as it was answered, keyed by the model name the reply reported,
and the report grouped them: the two small-model calls add up to 0.004250 and the large-model call is 0.028000, and the total is again 0.032250. Both routes agree because they use the same arithmetic on the same recorded tokens. If a call's model had not been in the price table, the report would have shown "unpriced"
for that row instead of quietly leaving it out.