<p>This article examines the legitimacy of AI-mediated governance through typed jurisdiction allocation, using the Meta-Layered Framework as a diagnostic vocabulary for heterogeneous disputes. It argues that legitimacy assessment should attach not primarily to model quality, individual determinations, or generic reviewability, but to the institutional settlement that fixes which reasons may count and which determinants may be revised. In AI-mediated regimes, binding constraint is often exercised through upstream determinants such as categories, thresholds, routing rules, and update practices. Review may therefore exist at the level of application while the determinants that organise treatment across cases remain insulated from revision. Existing literatures on review, contestability, redress, and standardisation identify related difficulties, but often leave under-specified the jurisdictional significance of routing. On the non-arbitrariness constraint adopted here, such arrangements can claim authorisation to bind only if relevant determinants are revisable under publicly stated standards and answerability extends to determinant-level remedy. Authorisation is defeated where answerability fails in practice or where relevant objections are excluded from admissibility with remedial force.</p>