Design token-based metering and billing for an LLM product, and explain why it must be decoupled from the call path.

The LLM billing round has three trapdoors: retries double-charging customers, metering latency leaking into inference, and month-end numbers that don't match the provider invoice. The design that survives all three is the answer.

Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.