Tenant-Aware Batch Processing for Document Summarization With Usage Accounting

The article argues that summarizing multiple documents should be treated as a small data pipeline rather than a simple loop over an API endpoint. It recommends tenant-aware batches with bounded inputs, idempotency keys, and usage recorded before results are shown, so each document maps to one accountable record. Usage should be a measured or deliberately estimated value tied to the amount of text processed rather than the document count. Manifests and usage events are kept longer than raw text, subject to privacy and contractual rules, and item-level failures should not halt the rest of a batch. Retries are permitted only when the previous attempt is known to have ended without a committed result or when an idempotency key protects the operation.