Zero-Shot First: Prompting GPT-5.2 Thinking and o3

OpenAI's reasoning best practices and 2026 prompt guides state that reasoning models work best with brief, clear instructions and a defined end goal: try zero-shot first, and if a zero-shot prompt fails, add structure before adding few-shot examples.

Goal: Design an idempotent webhook ingestion architecture capable of absorbing 50,000 events/second with zero duplicate downstream side effects. Output Format: Architecture specification table followed by failure-recovery sequence.