Systems framework demonstrates deterministic compilation of recurrent AI judgments into revisable artifacts, highlighting reduced redundant inference and safer decision execution.
AI-assisted workflows often mix stable sub-decisions with genuinely open-ended judgment. Reinvoking a model for the entire task repeats inference on decisions that may already be explicit, testable, and reusable. This paper presents judgment compilation: a provider-neutral lifecycle that observes recurrent AI-assisted judgments, admits bounded reusable portions as typed deterministic artifacts, composes them at runtime, escalates only unsupported or conflicting judgment slots, and withdraws artifacts when evidence invalidates their recorded assumptions. A standalone prototype compiles validators, constraints, decision tables, and workflow templates behind a common applicability-and-abstention contract. Admission binds each artifact to prospective evidence and a fixed policy commitment. Drift handling computes the typed transitive dependency closure of an invalidation seed, atomically quarantines that closure under a monotonic registry epoch, and revalidates artifact use immediately before side effects.
No takes yet. Share an insight, caveat, or question.
Mayk Loide Baccaro (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: