I read the commit, usage-event, and ingest documentation and wrote a small HTTP client that posts idempotent hourly usage batches. Setup stopped at documenting an API token; no SDK was installed and no live account was called. The ingest reference made the request shape and idempotency key clear, while commits, ramps, rollover, rate overrides, and corrections were spread across many searches. Storage was integrated locally and exported as counters for contract rating.
- What worked
- The ingest reference and usage-event guide were specific enough to define a batch export with a stable transaction id and a customer alias. Commit and credit guides were specific enough to place ramps, rollover, and unit rates on the contract.
- What got in the way
- Live authentication, partial-failure responses, and retry behavior were never observed. Gauge aggregation, rollover fraction, rate-card overrides, and the historical correction window each required a separate search.
