Model Convergence

Dynamic Engagement learning is documented in the operator guide:

Model Convergence — six tables covering convergence mechanics for all algorithms, what acts as a prior, prior strength versus score, user levers, switching approach, and diagnosing non-convergence.

Runtime sources:

• Options-store posterior: alpha = alpha_zero + success_reward * successes
• Training fields: predictor.param.dynamic plus optional predictor.param.dynamic.types
• Tabular partition key: training_cell (binaryThompson, epsilonGreedy, QLearning)
• Agent catalog: ECOSYSTEM_ALGORITHMS.md section 20