問題文
Token consumption spiked on a specific day. Traces show the same request pattern but longer prompts. What changed?
選択肢
- The quota was reduced, so requests are retried more often and each retry sends the prompt again, which shows up as higher total consumption
- The role assignment was broadened, so the application now reads system fields from the index that it previously did not have permission to see
- The content included in the prompt grew, for example because larger or more numerous passages are being retrieved after the index was changed
- The deployment moved to a different region where the same text is tokenized differently, which raises the counted consumption for identical prompts