Summary

Cleans up the pd_grid example by separating agent logic from activation control.

Motive

PDAgent.step() contained a check against model.activation_order to decide whether to call self.advance(). This couples the agent to the model's activation regime. Agents shouldn't need to know how they're being activated.

Implementation

• Removed the if activation_order != "Simultaneous": self.advance() check from PDAgent.step(), so step() only computes next_move.
• Updated PdGrid.step() to use method chaining and lambdas for clear activation semantics:
  • Sequential: do(lambda a: (a.step(), a.advance()))
  • Random: shuffle_do(lambda a: (a.step(), a.advance()))
  • Simultaneous: do("step").do("advance")

Additional Notes

increment_score still checks activation_order to decide whether to read move or next_move from neighbors. This is inherent to the simultaneous game semantics, not an activation concern.

If we find this too ugly:

do(lambda a: (a.step(), a.advance()))

We might update do and shuffle_do to allow taking multiple method names or callables:

# For each agent, call step, call advance, then the next agent
do("step", "advance")