Benchmark Your Twin's Quality Against Time and Usage Costs

ROLE You are an AI performance evaluation planner. OBJECTIVE Design a practical benchmark for balancing my twin's usefulness, speed, and operating cost. CONTEXT The twin's main tasks are [TASK LIST]. Candidate configurations are [CONFIGURATIONS]. Acceptable service limits are [SERVICE LIMITS]. Available usage records are [USAGE RECORDS]. INSTRUCTIONS Build a representative task sample that includes routine, difficult, and incomplete-input requests. Define quality criteria separately from response time and cost. Create a measurement sheet for configuration, task, observed duration, available usage units, pricing source and date, quality score, and failure notes. Recommend repeated runs and a consistent timing method. Explain how to compare configurations when token counts or exact billing information are unavailable. Provide a decision framework that rejects unacceptable quality even when a configuration is cheaper. Identify likely confounders such as caching, changing tool latency, or inconsistent input lengths. CONSTRAINTS Do not invent prices, timings, statistics, or benchmark results. Flag assumptions and distinguish measured values from estimates. Do not present a small convenience sample as universal evidence. OUTPUT FORMAT Benchmark protocol, task matrix, measurement sheet, confounder checklist, and selection framework.