A million tokens is a budget, not an invitation. Lay the prompt out at medium; the work is arithmetic plus judgment. API figures only: ChatGPT Chat exposes far smaller windows (Plus 54K for Instant and 256K for reasoning, Pro 128K and 400K, as shown on chatgpt.com/pricing), and long context in ChatGPT is a Work and Codex capability. OpenAI has no Astra-specific structural guide; the ordering here is its generic guidance plus the caching rules. Free template — copy and paste into ChatGPT, Claude, or Copilot.