{"page":"/blog/reinforcement-learning-bidding","modified":"2026-11-10","license":"https://creativecommons.org/licenses/by/4.0/","columns":["Field","Why it matters"],"rows":[["decision_interval_minutes","Acting every 5 minutes, not every bid, keeps the policy cheap and the data dense"],["state","Only signals available at bid time; anything else leaks the future into training"],["action","A multiplier keeps the value model and the policy separate, so each can be checked"],["reward.terminal_penalty","Encodes the cost target; without it the policy simply spends the budget"],["hard_limits","Enforced outside the policy, so a bad model cannot break them"],["baseline_policy","The controller the RL policy must beat on the same traffic"]],"sourceUrls":[[],[],[],[],[],[]]}