Skip to main content
PIO learns from the valued events your game already sends. There is no reward call: your project’s reward configuration defines which events matter, how their values combine, and over what window, and the platform derives the reward from that.

​
Choosing the signal

Optimise for the outcome you actually want, observed within a window you can wait for. Reward presets cover the common cases, such as first-time-user-experience completion; custom configurations combine your own valued events.

​
Pitfalls

  • Sparse signals: an event that fires for one player in a thousand gives the model little to learn from. Prefer outcomes most players reach.
  • Misaligned values: the model optimises exactly what the values say. If level_complete sends 1.0 for a win and 0.0 for a loss, wins are the objective, nothing subtler.
  • Dropped values: negative, NaN, and infinite event values never enter the queue, so a signal encoded as negative numbers silently vanishes.