Simulate payout of prediction-prophet-gpt4 if we were using Kelly

[gabrielfior]

• We have the positions the agent had when market was resolved (Dune and/or subgraph)
• We can call the get_kelly_bet and get a bet_amount we would use (let's make this variable, e.g. max_bet_amount = [1,2,5,10]
• We can then calculate the payout per position, average this out per month, etc and compare with our current payouts

[kongzii]

Don't we also need agent’s predicted probability to compute Kelly? If agent bought position when market was at 60%, it doesn't mean agents probability wasn't 51% or 99%.

You should be able to query it from Langfuse relatively easily.

[evangriffiths]

See #419 for simulation results. Looks like switching to kelly is a good idea!