The original title is "How to Test Structured Outputs Across Multiple AI Models" which is 10 words. Let me check the rules:

Short summary

A practical guide to testing structured JSON outputs across multiple LLMs in production, separating concerns into syntax validity, schema validity, and semantic validity. It covers building a workflow test set with edge cases like multilingual inputs and prompt injections, and tracking failure-mode metrics (parse rate, schema validation rate, semantic accuracy, p95 latency, cost per task) per model. The core message: a valid JSON parse does not mean a correct business outcome — test the full production path, not just playground responses.

• •Structured output testing requires three layers: syntax, schema, and semantic validity
• •Build a workflow test set with edge cases including multilingual inputs and prompt injections
• •Track per-model failure metrics: parse rate, schema rate, semantic accuracy, latency, cost

Generated with AI, which can make mistakes.