Abstract
This publication discloses, in enabling detail, a License-Anchored LLM Capability Governor: a system that governs which large-language-model any principal (a human user or an autonomous persona/agent) may cause to run, by resolving exactly one entitlement lever — a license tier — into a concrete model, and then enforcing that resolution wherever a model can be selected. Resolution walks a fixed precedence (per-principal override → role-exception regex → license-to-tier map → legacy role map → global default) and passes through an ordinal capability ladder so that models become comparable, not merely listed. The resolved model is then clamped to the physical seat's capability ceiling, with a distinctive property: an override id whose capability is unknown (absent from the ladder) is treated as unprovable and fails closed to the seat ceiling rather than being trusted. Enforcement is asymmetric ordinal substitution applied in three planes — an inference route guard that rewrites an over-capable per-call request down to the ceiling (never up), a workstation injector that pushes the resolved model plus a per-tier operational envelope into agent command-line runtimes, and a worker sync that writes the resolved model to background cognitive workers — with a drift auditor reconciling stored state against policy. This document, its diagrams, and its clean-room reference implementation are published to establish prior art and bar patenting of the mechanism.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "License-Anchored LLM Capability Governor with Ordinal Down-Only Route Clamp and Tri-Plane Enforcement", Technical Disclosure Commons, (July 13, 2026)
https://www.tdcommons.org/dpubs_series/10871