Freeze a closed comparison period

Begin with a completed period whose records have largely arrived, and align application and account timezones. Specify whether the period is based on request start, completion, or posting time. List long-running tasks that cross the boundary separately. Save export dates and filters so a changing live view is not repeatedly compared with a fixed snapshot.

Separate task records, API calls, and billing entries. They need not correspond one to one: a task may create several requests, and posting can be delayed. Join them using permitted identifiers without copying authentication headers, secrets, or full user inputs into a reconciliation sheet.

Explain quantities before money

Group usage by public model identifier and actual billing category with matching units. Do not combine input, output, and other categories into one total multiplied by a single rate. Record when the pricing rules were checked, and handle historical rate changes according to applicable rules rather than repricing all history with today’s figure.

Classify differences as missing application records, duplicate imports, in-flight calls, unclear categories, or period mismatches. Keep absent usage unknown. Inspect execution when requests were retried or cancelled; a frontend failure message alone does not prove that no call took place or no usage was recorded.

Repair collection and retain an explanation

Assign an owner, evidence location, and finding to each discrepancy. Share only redacted times, public model identifiers, necessary request identifiers, and a concise difference summary with support. Any accounting adjustment should follow established authorization rather than being bundled into a diagnostic production write.

After fixing collection or import behavior, rerun the same period and check for duplicate entries. Distinguish reconciled items, explained but unposted items, and unresolved questions in the report. The goal is a reproducible explanation, not a spreadsheet forced to match. Preserve uncertainty instead of manufacturing a precise result without evidence.

Sources & further reading

An owned publication maintained by NexoToken. Product references reflect the publisher’s commercial interest and are not independent certification.