Punting seems to be a problem that can be solved through machine learning.

https://en.wikipedia.org/wiki/Reinforcement_learning

1. There are rewards - a good punt (positive) a bad punt (negative) and certain services provided can be quantified. Every agent has a slightly different value function
2. There are uncertainties around a girl's mood and availability, and which shop
3. You can choose to share your information. You can give false information. You can choose to update your priors with information on this forum - you need to decide if other agents are giving false information.
4. Your strategy impacts other agent's strategies and vice versa
5. There are Constraints to work with: money, location, time of the day etc.

Any machine learning experts on the forum to give this problem a go?