Cross-document answers with a locator on every claim, at a scale only a 1.05M window allows. Reasoning: high; reconciling many sources is deep work. API: reasoning.effort "high"; in ChatGPT, Work or Codex only, not Chat. Astra scored 100 percent on OpenAI's 8-needle retrieval test between 256K and 512K tokens and 96.3 percent from 512K to 1M, which is why the final re-check of key claims is there for the top of the window. Free template — copy and paste into ChatGPT, Claude, or Copilot.