Budget engine
The subsystem that allocates cognitive resources (compute, retrieval depth, reasoning complexity) to operations based on utility threshold gating. Controls fast-mode vs. slow-mode selection and enforces graceful degradation under resource exhaustion.
The budget engine is the operational discipline that prevents the cognitive loop from spending unlimited resources on every event regardless of its importance. It computes a utility score for each incoming operation - combining expected reward, urgency, goal relevance, and novelty - and gates the depth of processing against that score. Operations above the threshold get slow-mode processing: full agent dispatch, maximum retrieval depth, cross-encoder reranking, and world-model prediction with causal model consultation. Operations below the threshold get fast-mode processing: reduced agent set, shallower retrieval, no reranking. Under resource exhaustion - when concurrent event volume is high or infrastructure is degraded - the budget engine enforces graceful degradation by progressively restricting slow-mode eligibility rather than timing out or failing. This makes the system resilient to load spikes at the cost of temporary quality reduction, which is the correct tradeoff for an always-on cognitive substrate. Budget decisions are recorded in the session's telemetry and are visible in cognitive observability dashboards.