Curiosity engine
The subsystem that proposes bounded experiments to reduce uncertainty in high-value areas. Curiosity score is a weighted combination of information gain (40%), novelty (25%), uncertainty (25%), and visit count (10%). All experiments pass constitutional and budget checks before execution.
The curiosity engine prevents the cognitive system from becoming purely reactive - executing requests and refining existing knowledge without ever probing what it does not know. It maintains a map of uncertainty across domains and scores potential experiments by how much information they would yield relative to their cost. Information gain is weighted most heavily (40%) because the goal is reduction of meaningful uncertainty, not mere novelty-seeking. Novelty and uncertainty together account for another 50%, ensuring that experiments target genuinely unknown territory rather than re-exploring familiar ground. The visit count term (10%) acts as a recency penalty: directions that have been explored recently score lower even if their novelty score has not yet updated, preventing the engine from repeatedly proposing the same experiment. All proposed experiments pass through the constitutional engine (to confirm they don't violate invariants) and the budget engine (to confirm the cost is justified by the expected information gain) before execution. Experiments that are accepted but produce surprising results - outcomes that contradict the world model's predictions - are high-signal events that trigger accelerated world-model updates.