Abstract
This document discloses, as enabling public prior art, a method and system for enforcing a periodic utilization cap over heterogeneous large-language-model (LLM) usage without maintaining a separate rate-limit counter store. The consumption of models with different price/performance profiles is normalized onto a single synthetic currency — base-equivalent tokens — by assigning each model a drain weight pinned to the ratio of that model's provider output price to a chosen base model's output price. The cap for a given account is enforced before each call by computing the account's current-window utilization on demand, as a query-time SQL CASE aggregation over the very same append-only, idempotent meteredusage table that (a) bills the customer and (b) drives the customer-facing usage meter. On breach, the caller receives an HTTP 429 response carrying used, cap, and resetat. Because the aggregation reads the same fact rows that the exactly-once metered debit writes — keyed on a caller-supplied idempotency identifier — a retried turn can neither be double-billed nor double-drain the cap. Limit, invoice, and meter therefore share one tamper-consistent source of truth, and the cap is historically re-derivable when provider prices change, without migrating any stored counter. The mechanism is disclosed as a dependent claim under an exactly-once metering-debit protocol.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "Reference-Model-Equivalent Weekly Utilization Cap Computed Directly from the Billing Ledger", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10967