Governing the Transition from Learning to Effect: A Target Architecture for Commit-Time Warrant and Governed Model Reliance

A governance determination can be warranted as learning material while the model produced from that material acquires a different role, purpose and institutional consequence. A valid beginning does not establish the authority of an altered destination. This research preprint specifies a target architecture for governing the transition from learning to effect. It distinguishes START, COMMIT, ACTIVATE and RELY as separate action claims, each bound to the actual model configuration, operative time, scope and authority. The central contribution is a learning-specific whole-path warrant architecture connecting governance-derived inputs to particular learning-produced effects. A governed warrant path (Π) uses typed handoffs to distinguish verified transformations from normatively justified transfers of entitlement. The architecture separates the standing of a current action (W_local) from the standing of the specific historical path invoked as its justification (W_path). Required-parent coverage, three-valued status assessment (PASS, OPEN, NO), an E1–E11 operational contract, conditional formal results and observational countermodels define the architecture's obligations and limits. The accompanying Supplement S1 documents PR-01, an executed synthetic omitted-condition conformance check. The experiment uses 22 synthetic training cases, two fitted decision-tree classifiers, 23 expected-status checks and five fixture mutations. The recorded results are 23/23 and 5/5 respectively. Supplement S1 is provided in two complementary formats: a readable PDF containing the original supporting files as embedded attachments, and a reproducibility ZIP containing executable scripts, synthetic fixtures, recorded outputs and integrity records. The experiment establishes bounded internal consistency under author-stipulated rules. It does not constitute independent adversarial validation, production assurance or proof of universal model safety. This work continues the research developed in Governing the Transition from Decision to Learning, extending the governance question from the authority to use a determination for learning to the authority governing what the resulting model may subsequently do. Research preprint — Version 1.0 — Not peer reviewed. Manuscript completed: 27 October 2026. Published: 28 October 2026.