Dream engine
The subsystem that synthesizes hypothetical scenarios from memories, goals, contradictions, and curiosity targets to test world-model predictions offline. Synthetic events are tagged as synthetic to prevent confabulation - imagined scenarios becoming indistinguishable from facts.
The dream engine gives the system a way to learn from experience it has not had. By synthesizing hypothetical scenarios - drawn from combinations of existing memories, current goals, unresolved contradictions, and curiosity engine targets - it generates predictions that the world model can be tested against without incurring the cost or risk of real-environment experimentation. These synthetic scenarios are structurally identical to real experience events, which is why the confabulation prevention is critical: every synthetic event carries a permanent synthetic tag that prevents it from being written to the episodic truth layer, prevents it from seeding the reinforcement pipeline as if it were a real outcome, and prevents it from influencing trust score updates. The dream engine's output feeds directly into the open-ended evolution pipeline: before a capability proposal reaches constitutional review, the dream engine simulates what would happen if that capability were active, allowing the world model to pre-screen proposals at low cost. It also generates training scenarios for the curiosity engine's high-uncertainty domains, helping the world model build predictions for situations it has not yet encountered in real operation.