{"version":"2.1.281","anchor":"side-queries-retry-after-budget-and-a-rebuilt-system-prompt","canonical_anchor":"side-queries-retry-after-budget-and-a-rebuilt-system-prompt","heading":"Side queries: Retry-After budget and a rebuilt system-prompt prefix","tier":"internal","area":"Models","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.281\/e\/side-queries-retry-after-budget-and-a-rebuilt-system-prompt","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.281","markdown":"### Side queries: Retry-After budget and a rebuilt system-prompt prefix\n\nSide queries accept a Retry-After time budget, and their attribution block is now built by a shared helper\n\n**Unclear.** The finding does not say which side queries set `retryAfterBudgetMs` or what `forceAttributionHeader` changes in the attribution block.\n\n**What**\n\nSide queries (`sideQuery`) are requests Claude Code sends to the model apart from the main exchange. They gain two changes:\n\n- `retryAfterBudgetMs`, a new option that wraps the network responses. Retry-After is a server's instruction to wait a given time before trying again, and this option sets a time budget in milliseconds for that waiting.\n\n- `onFetchAttempt` now actually wraps each network request, instead of having no effect on it.\n\nThe block at the start of the system prompt that carries attribution and fingerprint information is now built by a helper with a `forceAttributionHeader` option. The system prompt is the set of instructions sent to the model before the conversation.\n\n**Why**\n\nIn practice, side queries can be given a limit on how long they will wait when a server asks them to retry later.\n\n- Area: Models\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 2\/5"}