Pass Window Arbitrage

Pass Window Arbitrage is the practice of matching inference billing models to the shape of an agency's workload rather than to its total token volume. Metered gateways charge per token, so cost scales linearly with usage; fixed-duration passes sell unlimited tokens inside a 1-hour to 3-day window, which means the effective rate collapses on bursty work and balloons on steady drip. The framework asks two questions before any client integration: how concentrated is the workload in time, and how tolerant is the deliverable to a hard cutoff. A retainer that runs nightly batch summarization across ten client accounts fits a pass model; a live client-facing assistant that must answer at 3am does not. endlesstokens illustrates the model with time-boxed passes on its own GPUs, while TokenDelivery.ai shows the metered alternative with deterministic outputs. Forrester's tokenomics warning about agentic workflows destroying margin is the reason this belongs in delivery planning, not procurement.