Reinforcement
The process that computes reward signals from operation outcomes and propagates reward-driven updates into retrieval priority, policy state, and identity state. Uses Hebbian compounding with a quality gate.
Reinforcement is the learning mechanism that connects outcomes to the memories and decisions that preceded them. After every cognitive loop iteration produces an observable outcome, the reinforcement engine computes a composite reward signal from seven normalized input channels: importance (18%), novelty (16%), prediction accuracy (14%), emotional weight (12%), goal relevance (14%), policy alignment (16%), and contradiction risk (10%, inverted). Each channel captures a different dimension of what makes an outcome valuable: importance reflects the stakes, novelty rewards learning something new, prediction accuracy rewards correct anticipation, goal relevance credits goal progress, policy alignment rewards staying on policy, and contradiction risk penalizes conflicting with established knowledge. The composite signal then propagates into four coupled outputs: retrieval priority update (making useful memories more accessible), decay rate adjustment (slowing the decay of memories that contributed to good outcomes), policy delta (adjusting the behavioral weights that led to this outcome), and identity impact (slowly shifting the identity state if the outcome reveals a persistent behavioral tendency). Hebbian compounding adds a logarithmic bonus for memories that accumulate positive reinforcement repeatedly.