[
  {
    "query": "What is Python?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "What is the capital of France?",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "Who invented the telephone?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "What does HTTP stand for?",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "Translate 'hello' to Spanish",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "translation"
  },
  {
    "query": "What year was Python created?",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "Is Java object-oriented?",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "What is RAM?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual"
  },
  {
    "query": "Summarize this paragraph: The quick brown fox...",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "summarization"
  },
  {
    "query": "Explain what a variable is in programming",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "What is the difference between a list and a tuple in Python?",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "Write a simple hello world in JavaScript",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "What is recursion? Give a simple example",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "Translate this paragraph to French",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation"
  },
  {
    "query": "Summarize the key points of agile methodology",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "summarization"
  },
  {
    "query": "Explain binary search with a code example",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code"
  },
  {
    "query": "Write a Python function to check if a number is prime",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "code"
  },
  {
    "query": "Compare REST and GraphQL APIs",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "reasoning"
  },
  {
    "query": "Explain the CAP theorem in distributed systems",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "reasoning"
  },
  {
    "query": "Write SQL to find duplicate rows in a table",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code"
  },
  {
    "query": "Explain gradient descent step by step",
    "complexity": 0.5,
    "tier": "medium",
    "domain": "math"
  },
  {
    "query": "What is the time complexity of quicksort? Explain with an example",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "code"
  },
  {
    "query": "Design a rate limiter for an API",
    "complexity": 0.62,
    "tier": "hard",
    "domain": "code"
  },
  {
    "query": "Implement a LRU cache in Python",
    "complexity": 0.6,
    "tier": "hard",
    "domain": "code"
  },
  {
    "query": "Explain the Transformer architecture in detail",
    "complexity": 0.7,
    "tier": "hard",
    "domain": "science"
  },
  {
    "query": "Write a comprehensive tutorial on Docker and Kubernetes",
    "complexity": 0.68,
    "tier": "hard",
    "domain": "code"
  },
  {
    "query": "Analyze the pros and cons of microservices vs monoliths",
    "complexity": 0.65,
    "tier": "hard",
    "domain": "reasoning"
  },
  {
    "query": "Derive the backpropagation equations from first principles",
    "complexity": 0.8,
    "tier": "hard",
    "domain": "math"
  },
  {
    "query": "Design the Paxos consensus algorithm",
    "complexity": 0.92,
    "tier": "expert",
    "domain": "reasoning"
  },
  {
    "query": "Prove that P ≠ NP (or outline the key open problems)",
    "complexity": 0.98,
    "tier": "expert",
    "domain": "math"
  },
  {
    "query": "Design a distributed SQL database from scratch",
    "complexity": 0.95,
    "tier": "expert",
    "domain": "code"
  },
  {
    "query": "Implement a compiler for a simple language in Python",
    "complexity": 0.9,
    "tier": "expert",
    "domain": "code"
  },
  {
    "query": "Explain quantum entanglement and Bell's inequality with math",
    "complexity": 0.88,
    "tier": "expert",
    "domain": "science"
  },
  {
    "query": "Write a full-stack web app with React and FastAPI",
    "complexity": 0.85,
    "tier": "expert",
    "domain": "code"
  },
  {
    "query": "Analyze the ethical implications of AI in healthcare",
    "complexity": 0.72,
    "tier": "hard",
    "domain": "reasoning"
  },
  {
    "query": "Compare BERT and GPT architectures in depth",
    "complexity": 0.75,
    "tier": "hard",
    "domain": "science"
  },
  {
    "query": "What is async/await in Python?",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "Explain SOLID principles with examples",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "code"
  },
  {
    "query": "Write a regex to validate email addresses",
    "complexity": 0.3,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "Design a URL shortener system like bit.ly",
    "complexity": 0.7,
    "tier": "hard",
    "domain": "code"
  },
  {
    "query": "Implement a red-black tree in C++",
    "complexity": 0.8,
    "tier": "hard",
    "domain": "code"
  },
  {
    "query": "Explain Bayesian inference with an example",
    "complexity": 0.65,
    "tier": "hard",
    "domain": "math"
  },
  {
    "query": "Write a neural network from scratch in numpy",
    "complexity": 0.82,
    "tier": "expert",
    "domain": "code"
  },
  {
    "query": "Translate this legal document to Spanish",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "translation"
  },
  {
    "query": "Summarize this 50-page research paper",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "summarization"
  },
  {
    "query": "Debate the pros and cons of nuclear energy",
    "complexity": 0.6,
    "tier": "hard",
    "domain": "reasoning"
  },
  {
    "query": "Write a creative short story about time travel",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "creative"
  },
  {
    "query": "Explain what a closure is in JavaScript",
    "complexity": 0.3,
    "tier": "easy",
    "domain": "code"
  },
  {
    "query": "What is the difference between TCP and UDP?",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "factual"
  },
  {
    "query": "Prove the Pythagorean theorem",
    "complexity": 0.55,
    "tier": "medium",
    "domain": "math"
  },
  {
    "query": "What is Python?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "What is the capital of France?",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "factual",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Who invented the telephone?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual",
    "model": "llama3.2:3b"
  },
  {
    "query": "What does HTTP stand for?",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "factual",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'hello' to Spanish",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "What year was Python created?",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "factual",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Is Java object-oriented?",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "factual",
    "model": "llama3.1:8b"
  },
  {
    "query": "What is RAM?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "What is 2+2?",
    "complexity": 0.02,
    "tier": "trivial",
    "domain": "math",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Name one planet in our solar system",
    "complexity": 0.03,
    "tier": "trivial",
    "domain": "science",
    "model": "llama3.2:3b"
  },
  {
    "query": "Summarize this: 'The quick brown fox jumps over the lazy dog.'",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "summarization",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Explain what a variable is in programming",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "code",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "What is the difference between a list and a tuple in Python?",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a simple hello world in JavaScript",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "What is recursion? Give a simple example",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Translate 'Good morning, how are you?' to French",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Summarize the key points of agile methodology",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "summarization",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Solve for x: 2x + 5 = 15",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "math",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "What is photosynthesis? Explain briefly",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "science",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a haiku about spring",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "creative",
    "model": "llama3.1:70b"
  },
  {
    "query": "Why is the sky blue?",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "reasoning",
    "model": "mistral-large-latest"
  },
  {
    "query": "Explain binary search with a code example",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a Python function to check if a number is prime",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Compare REST and GraphQL APIs",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the CAP theorem in distributed systems",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write SQL to find duplicate rows in a table",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Explain gradient descent step by step",
    "complexity": 0.5,
    "tier": "medium",
    "domain": "math",
    "model": "deepseek-chat"
  },
  {
    "query": "What is the time complexity of quicksort? Explain with an example",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Translate this technical document abstract to German",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Summarize the plot of '1984' by George Orwell",
    "complexity": 0.37,
    "tier": "medium",
    "domain": "summarization",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a short story about a robot learning to dream",
    "complexity": 0.44,
    "tier": "medium",
    "domain": "creative",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the theory of relativity in simple terms",
    "complexity": 0.46,
    "tier": "medium",
    "domain": "science",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "What is the difference between supervised and unsupervised learning?",
    "complexity": 0.36,
    "tier": "medium",
    "domain": "factual",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Solve the quadratic equation: x^2 - 5x + 6 = 0",
    "complexity": 0.32,
    "tier": "medium",
    "domain": "math",
    "model": "llama3.1:70b"
  },
  {
    "query": "Design a rate limiter for an API",
    "complexity": 0.62,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Implement an LRU cache in Python",
    "complexity": 0.6,
    "tier": "hard",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the Transformer architecture in detail",
    "complexity": 0.7,
    "tier": "hard",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a comprehensive tutorial on Docker and Kubernetes",
    "complexity": 0.68,
    "tier": "hard",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Analyze the pros and cons of microservices vs monoliths",
    "complexity": 0.65,
    "tier": "hard",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "Derive the backpropagation equations from first principles",
    "complexity": 0.8,
    "tier": "expert",
    "domain": "math",
    "model": "deepseek-chat"
  },
  {
    "query": "Prove that the square root of 2 is irrational",
    "complexity": 0.58,
    "tier": "hard",
    "domain": "math",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate this legal contract summary to Japanese",
    "complexity": 0.55,
    "tier": "hard",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Summarize the latest advances in quantum computing (2023-2024)",
    "complexity": 0.66,
    "tier": "hard",
    "domain": "summarization",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a poem in the style of Edgar Allan Poe about AI",
    "complexity": 0.63,
    "tier": "hard",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the ethics of autonomous weapons systems",
    "complexity": 0.67,
    "tier": "hard",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "What is the role of the Golgi apparatus in cells?",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "science",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a script to scrape a website and extract all links",
    "complexity": 0.52,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the concept of idempotency in REST APIs",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Describe the water cycle for a 10-year-old",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "science",
    "model": "llama3.2:3b"
  },
  {
    "query": "What is a neural network? Give a simple analogy",
    "complexity": 0.24,
    "tier": "easy",
    "domain": "science",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Solve: If a train travels 60 mph for 2 hours, how far does it go?",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "math",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Name three data types in Python",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "code",
    "model": "llama3.2:3b"
  },
  {
    "query": "Translate 'Thank you very much' to Italian",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "What is the main ingredient in guacamole?",
    "complexity": 0.03,
    "tier": "trivial",
    "domain": "factual",
    "model": "mistral-small-latest"
  },
  {
    "query": "Write a for loop that prints numbers 1 to 10 in Python",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Explain what a deadlock is in concurrency",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Compare socialism and capitalism",
    "complexity": 0.47,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "What is the significance of the Higgs boson discovery?",
    "complexity": 0.56,
    "tier": "medium",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a SQL query to join three tables",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Create a simple JavaScript function that returns the current date",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "What is the Pythagorean theorem? Provide an example",
    "complexity": 0.23,
    "tier": "easy",
    "domain": "math",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Summarize the main ideas of Stoic philosophy",
    "complexity": 0.49,
    "tier": "medium",
    "domain": "summarization",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Write a recipe for chocolate chip cookies in poetic form",
    "complexity": 0.51,
    "tier": "medium",
    "domain": "creative",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Explain the difference between CPU and GPU",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "factual",
    "model": "llama3.1:8b"
  },
  {
    "query": "Solve: What is the derivative of x^3?",
    "complexity": 0.31,
    "tier": "medium",
    "domain": "math",
    "model": "deepseek-chat"
  },
  {
    "query": "Design a simple task scheduler in Python",
    "complexity": 0.59,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the concept of 'impostor syndrome' in the workplace",
    "complexity": 0.34,
    "tier": "easy",
    "domain": "reasoning",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "What are the three laws of robotics?",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "science",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Translate 'The food was delicious' to Mandarin Chinese",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a CSS snippet to center a div",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "Explain the Drake equation",
    "complexity": 0.44,
    "tier": "medium",
    "domain": "science",
    "model": "llama3.1:70b"
  },
  {
    "query": "What is the difference between error and exception in programming?",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Solve: 15% of 200",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "math",
    "model": "llama3.2:3b"
  },
  {
    "query": "Write a haiku about debugging",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Explain the concept of 'technical debt'",
    "complexity": 0.37,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "What are the benefits of using TypeScript over JavaScript?",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Summarize the plot of 'The Great Gatsby' in one paragraph",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "summarization",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate the following error message to Spanish: 'File not found'",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a Python decorator that measures execution time",
    "complexity": 0.55,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the concept of 'virtue ethics'",
    "complexity": 0.46,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "What is the function of the mitochondria?",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "science",
    "model": "llama3.1:8b"
  },
  {
    "query": "Solve for x: log(x) = 2",
    "complexity": 0.3,
    "tier": "medium",
    "domain": "math",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Design a simple REST API for a todo list",
    "complexity": 0.57,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the difference between let, const, and var in JavaScript",
    "complexity": 0.23,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a limerick about a programmer",
    "complexity": 0.32,
    "tier": "easy",
    "domain": "creative",
    "model": "llama3.1:70b"
  },
  {
    "query": "What is the greenhouse effect? Explain simply",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "science",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'I would like to book a flight' to French",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "What is the difference between a stack and a queue?",
    "complexity": 0.24,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Explain the concept of 'opportunity cost' in economics",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "reasoning",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a binary search algorithm in Python",
    "complexity": 0.43,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Summarize the key innovations of the Renaissance",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "summarization",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Solve: What is the area of a circle with radius 5?",
    "complexity": 0.1,
    "tier": "easy",
    "domain": "math",
    "model": "llama3.2:3b"
  },
  {
    "query": "Create a simple HTML page with a button",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Explain the concept of 'black swan events'",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "What is CRISPR used for?",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "science",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a regular expression to match an email address",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Translate 'Where is the nearest hospital?' to German",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Write a short story about a character who wakes up with amnesia",
    "complexity": 0.49,
    "tier": "medium",
    "domain": "creative",
    "model": "llama3.1:70b"
  },
  {
    "query": "Explain the difference between correlation and causation",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "reasoning",
    "model": "deepseek-chat"
  },
  {
    "query": "What is a SAT solver used for in computer science?",
    "complexity": 0.52,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "What is an API?",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "factual",
    "model": "llama3.2:3b"
  },
  {
    "query": "Who wrote 'Romeo and Juliet'?",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "factual",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "What does CSS stand for?",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "factual",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'good night' to German",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "What is the square root of 64?",
    "complexity": 0.03,
    "tier": "trivial",
    "domain": "math",
    "model": "mistral-small-latest"
  },
  {
    "query": "What color is the sky on a clear day?",
    "complexity": 0.02,
    "tier": "trivial",
    "domain": "science",
    "model": "llama3.1:8b"
  },
  {
    "query": "What does SQL stand for?",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "factual",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Name one mammal that can fly",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "science",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Translate 'I love you' to French",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "What is the largest ocean on Earth?",
    "complexity": 0.03,
    "tier": "trivial",
    "domain": "factual",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Explain what a function is in programming",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a simple for loop in C++ that prints 0 to 9",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "What is the difference between '==' and '===' in JavaScript?",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Translate 'The weather is nice today' to Italian",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Summarize the water cycle in two sentences",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "summarization",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Solve for y: 3y - 7 = 11",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "math",
    "model": "llama3.2:3b"
  },
  {
    "query": "Why do we have seasons?",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "science",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a two-line poem about the moon",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Explain why ice floats on water",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "reasoning",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "What is a stack overflow in programming?",
    "complexity": 0.23,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a Python function to find the maximum of three numbers",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "What is the difference between a class and an object?",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'Where is the bathroom?' to Spanish",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:70b"
  },
  {
    "query": "Summarize the main idea of the movie 'Inception'",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "summarization",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Solve: What is 25% of 80?",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "math",
    "model": "gpt-4o"
  },
  {
    "query": "Write a short story about a lost key",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "creative",
    "model": "gpt-4o"
  },
  {
    "query": "Explain how a memoization works with an example in JavaScript",
    "complexity": 0.44,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Compare TCP and UDP protocols in detail",
    "complexity": 0.43,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "What is the halting problem? Why is it important?",
    "complexity": 0.51,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "Derive the formula for the area of a circle",
    "complexity": 0.37,
    "tier": "medium",
    "domain": "math",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Explain the second law of thermodynamics",
    "complexity": 0.53,
    "tier": "medium",
    "domain": "science",
    "model": "llama3.1:70b"
  },
  {
    "query": "Write a SQL query to get the top 5 highest paid employees",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Translate this business email to Japanese: 'Dear Sir, we appreciate your prompt response.'",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Summarize the plot of 'The Odyssey' in 100 words",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "summarization",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Write a sonnet about artificial intelligence",
    "complexity": 0.54,
    "tier": "medium",
    "domain": "creative",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the concept of race conditions in multithreading",
    "complexity": 0.46,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "What is the difference between deep learning and traditional machine learning?",
    "complexity": 0.47,
    "tier": "medium",
    "domain": "science",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Solve the integral of x^2 dx",
    "complexity": 0.36,
    "tier": "medium",
    "domain": "math",
    "model": "llama3.1:8b"
  },
  {
    "query": "Implement a binary tree traversal in Python (in-order)",
    "complexity": 0.56,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Design a simple load balancer algorithm",
    "complexity": 0.58,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Write a regular expression to validate a US phone number",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "Explain the prisoner's dilemma and its implications",
    "complexity": 0.49,
    "tier": "medium",
    "domain": "reasoning",
    "model": "deepseek-chat"
  },
  {
    "query": "What is the Golden Ratio? Provide examples",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "math",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate a medical prescription summary to German",
    "complexity": 0.52,
    "tier": "medium",
    "domain": "translation",
    "model": "llama3.1:70b"
  },
  {
    "query": "Summarize the key arguments in Plato's 'Republic'",
    "complexity": 0.57,
    "tier": "hard",
    "domain": "summarization",
    "model": "gpt-4o"
  },
  {
    "query": "Write a dark fantasy short story about a cursed mirror",
    "complexity": 0.64,
    "tier": "hard",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Implement a concurrent web scraper in Python with asyncio",
    "complexity": 0.69,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the proof of Fermat's Last Theorem at a high level",
    "complexity": 0.78,
    "tier": "hard",
    "domain": "math",
    "model": "gpt-4o"
  },
  {
    "query": "Analyze the performance implications of columnar vs row-based storage",
    "complexity": 0.66,
    "tier": "hard",
    "domain": "reasoning",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Describe the architecture of a distributed key-value store like DynamoDB",
    "complexity": 0.72,
    "tier": "hard",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "What is the Bellman equation in reinforcement learning?",
    "complexity": 0.63,
    "tier": "hard",
    "domain": "science",
    "model": "mistral-large-latest"
  },
  {
    "query": "Translate a complex legal disclaimer to French",
    "complexity": 0.61,
    "tier": "hard",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Design a real-time chat system supporting 1 million concurrent users",
    "complexity": 0.75,
    "tier": "hard",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the concept of zero-knowledge proofs",
    "complexity": 0.7,
    "tier": "hard",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a detailed guide on optimizing Python code with C extensions",
    "complexity": 0.68,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Prove that there are infinitely many prime numbers",
    "complexity": 0.59,
    "tier": "hard",
    "domain": "math",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "What is the 'butterfly effect' in chaos theory?",
    "complexity": 0.55,
    "tier": "medium",
    "domain": "science",
    "model": "llama3.1:8b"
  },
  {
    "query": "Write a memoization decorator in Python",
    "complexity": 0.53,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Explain the difference between a process and a thread",
    "complexity": 0.31,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'Congratulations on your new job' to Portuguese",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "What is the role of the nucleus in a cell?",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "science",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Write a Java program to reverse a string",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "code",
    "model": "llama3.2:3b"
  },
  {
    "query": "Solve for x: 4x^2 - 16 = 0",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "math",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a haiku about winter",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Explain why the sky is red at sunset",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "reasoning",
    "model": "gpt-4o-mini"
  },
  {
    "query": "What is an abstract class in Java?",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a simple HTML form with two input fields",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate 'I need a doctor' to Korean",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Summarize the plot of 'Hamlet' in three sentences",
    "complexity": 0.32,
    "tier": "easy",
    "domain": "summarization",
    "model": "gpt-4o"
  },
  {
    "query": "What is the difference between GET and POST in HTTP?",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a C program to calculate factorial using recursion",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the concept of 'sharding' in databases",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "What is the difference between L1 and L2 regularization?",
    "complexity": 0.54,
    "tier": "medium",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Solve the system of equations: 2x + y = 10, x - y = 2",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "math",
    "model": "llama3.1:70b"
  },
  {
    "query": "Write a Python script to download an image from a URL",
    "complexity": 0.36,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Explain the Turing test and its criticisms",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "reasoning",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Translate 'The system is currently offline' to Russian",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "translation",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Summarize the main findings of the Human Genome Project",
    "complexity": 0.5,
    "tier": "medium",
    "domain": "science",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a villanelle about lost time",
    "complexity": 0.62,
    "tier": "hard",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Implement a connection pool in Go",
    "complexity": 0.71,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the concept of 'eventual consistency' in distributed systems",
    "complexity": 0.65,
    "tier": "hard",
    "domain": "reasoning",
    "model": "gpt-4o"
  },
  {
    "query": "Derive the Black-Scholes equation for option pricing",
    "complexity": 0.85,
    "tier": "expert",
    "domain": "math",
    "model": "gpt-4o"
  },
  {
    "query": "Design a distributed consensus protocol like Raft",
    "complexity": 0.88,
    "tier": "expert",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Explain the holographic principle in theoretical physics",
    "complexity": 0.92,
    "tier": "expert",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a full compiler frontend for a small language in Rust",
    "complexity": 0.95,
    "tier": "expert",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Prove the Riemann Hypothesis (outline the main approach)",
    "complexity": 0.99,
    "tier": "expert",
    "domain": "math",
    "model": "gpt-4o"
  },
  {
    "query": "Analyze the security of the TLS 1.3 handshake",
    "complexity": 0.82,
    "tier": "expert",
    "domain": "reasoning",
    "model": "llama3.1:70b"
  },
  {
    "query": "Implement a B+ tree index from scratch",
    "complexity": 0.87,
    "tier": "expert",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate a complex patent document to Chinese",
    "complexity": 0.76,
    "tier": "hard",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "What is the difference between Bayesian and frequentist statistics?",
    "complexity": 0.67,
    "tier": "hard",
    "domain": "math",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a high-performance WebSocket server in C++",
    "complexity": 0.79,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the many-worlds interpretation of quantum mechanics",
    "complexity": 0.73,
    "tier": "hard",
    "domain": "science",
    "model": "llama3.1:70b"
  },
  {
    "query": "Design a URL shortening service like TinyURL (full design)",
    "complexity": 0.69,
    "tier": "hard",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Prove the central limit theorem",
    "complexity": 0.84,
    "tier": "expert",
    "domain": "math",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Write a distributed task queue using Redis",
    "complexity": 0.74,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the concept of 'transfer learning' in neural networks",
    "complexity": 0.56,
    "tier": "medium",
    "domain": "science",
    "model": "gpt-4o"
  },
  {
    "query": "Solve the traveling salesman problem using dynamic programming",
    "complexity": 0.77,
    "tier": "hard",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate a software license agreement to Spanish",
    "complexity": 0.58,
    "tier": "hard",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a detailed critique of the OpenAI GPT-4 architecture",
    "complexity": 0.7,
    "tier": "hard",
    "domain": "reasoning",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "What is the curse of dimensionality in machine learning?",
    "complexity": 0.51,
    "tier": "medium",
    "domain": "science",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Fix this Python code: `def add(a,b): return a-b` – it should add, not subtract.",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a Python one-liner to reverse a string.",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "code",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Why does this JavaScript code print 'undefined'? `var x; console.log(x);`",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "code",
    "model": "llama3.2:3b"
  },
  {
    "query": "Implement a function `is_palindrome(s)` in Python that ignores spaces and case.",
    "complexity": 0.32,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Debug this SQL: `SELECT * FORM users WHERE name = 'John';`",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "Write a recursive function to compute the nth Fibonacci number in Java.",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "What is the output of `console.log(1 + '2' + 3)` in JavaScript? Explain.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Write a C function to swap two integers using pointers.",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "code",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Fix the memory leak in this C++ snippet: `int* p = new int; p = new int; delete p;`",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Implement a queue using two stacks in Python.",
    "complexity": 0.52,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Write a regex to extract all email addresses from a text.",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Why does this infinite loop happen? `for i in range(10): i -= 1`",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a Python decorator that caches return values of a function.",
    "complexity": 0.58,
    "tier": "hard",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Convert this list comprehension to a for loop: `[x**2 for x in range(10) if x%2==0]`",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a SQL query to find employees who earn more than their managers.",
    "complexity": 0.44,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Implement a binary search in a sorted array (any language).",
    "complexity": 0.36,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Debug this Python: `print('Hello' + 123)` – what error and how to fix?",
    "complexity": 0.1,
    "tier": "trivial",
    "domain": "code",
    "model": "llama3.2:3b"
  },
  {
    "query": "Write a function that merges two sorted lists into one sorted list (O(n)).",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Explain why `[1,2,3].map(parseInt)` returns `[1, NaN, NaN]` in JavaScript.",
    "complexity": 0.47,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Write a simple HTML page that fetches data from a REST API and displays it.",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Implement a singleton pattern in Python.",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Fix the race condition in this multi‑threaded Python code (pseudo).",
    "complexity": 0.63,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a recursive descent parser for simple arithmetic expressions (+, -, *, /).",
    "complexity": 0.71,
    "tier": "hard",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Why does this C code crash? `int arr[5]; arr[10] = 42;`",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "code",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a Python generator that yields the Fibonacci sequence infinitely.",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Implement an LRU cache using `OrderedDict` in Python.",
    "complexity": 0.59,
    "tier": "hard",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "What does `*args` and `**kwargs` do in Python? Give an example.",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "Write a JavaScript function that throttles another function (limit calls per second).",
    "complexity": 0.55,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Debug this SQL injection vulnerability: `\"SELECT * FROM users WHERE id = \" + user_id`",
    "complexity": 0.3,
    "tier": "easy",
    "domain": "code",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Implement a simple HTTP server in Python using `sockets`.",
    "complexity": 0.62,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain the output: `console.log([] + []); console.log([] + {}); console.log({} + []);`",
    "complexity": 0.49,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Write a C++ program that reverses a linked list.",
    "complexity": 0.46,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Convert this `try/except` to using `contextlib.suppress` in Python.",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "code",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Implement the Sieve of Eratosthenes in Java.",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Why does `0.1 + 0.2 !== 0.3` in JavaScript? Explain floating point.",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "code",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a Python script to find duplicate files in a directory (by hash).",
    "complexity": 0.51,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Fix the deadlock in this pseudocode: two threads lock A then B, and B then A.",
    "complexity": 0.66,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Implement a simple event emitter in JavaScript (Node.js style).",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "What is tail recursion? Convert this factorial to tail‑recursive: `def fact(n): return 1 if n==0 else n*fact(n-1)`",
    "complexity": 0.37,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Write a Go routine that computes the sum of squares concurrently.",
    "complexity": 0.54,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'The server is down' to Arabic",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Summarize the concept of 'reference counting' in memory management.",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "summarization",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Solve `∫ x e^x dx`",
    "complexity": 0.43,
    "tier": "medium",
    "domain": "math",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain why black holes evaporate (Hawking radiation).",
    "complexity": 0.61,
    "tier": "hard",
    "domain": "science",
    "model": "gpt-4o"
  },
  {
    "query": "Write a haiku about a segfault",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "creative",
    "model": "llama3.2:3b"
  },
  {
    "query": "What is the difference between `malloc` and `calloc` in C?",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "factual",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a Python script to fetch JSON from an API and pretty‑print it.",
    "complexity": 0.24,
    "tier": "easy",
    "domain": "code",
    "model": "mistral-small-latest"
  },
  {
    "query": "Debug this React hook: `useEffect(() => { setCount(count+1) }, [])` – why infinite loop?",
    "complexity": 0.4,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Implement a deep copy function for nested dictionaries in Python.",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Explain the `volatile` keyword in Java.",
    "complexity": 0.44,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a SQL query to delete duplicate rows keeping one copy.",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Why does this Python code raise `UnboundLocalError`? `x = 10; def foo(): print(x); x=5`",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Implement a simple key‑value store with TTL (time‑to‑live) in Python.",
    "complexity": 0.57,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'The warranty is void if seal is broken' to Mandarin.",
    "complexity": 0.36,
    "tier": "medium",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Summarize the paper 'Attention is All You Need' in 5 bullet points.",
    "complexity": 0.49,
    "tier": "medium",
    "domain": "summarization",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Solve `det([[1,2],[3,4]])` (determinant).",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "math",
    "model": "llama3.1:8b"
  },
  {
    "query": "Explain the chemical process of rusting.",
    "complexity": 0.23,
    "tier": "easy",
    "domain": "science",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a limerick about a null pointer",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "What is a JIT compiler? Give an example runtime.",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "factual",
    "model": "mistral-large-latest"
  },
  {
    "query": "Write a C# method that reads a CSV file and returns a list of objects.",
    "complexity": 0.42,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Fix the JSON syntax error: `{name: 'John', age: 30}`",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "code",
    "model": "llama3.2:3b"
  },
  {
    "query": "Implement a rate limiter using the token bucket algorithm in Python.",
    "complexity": 0.65,
    "tier": "hard",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the output `(0 == '0')` vs `(0 === '0')` in JavaScript.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a Bash script to find all `.log` files older than 7 days and compress them.",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Debug this Python multiprocessing code that hangs: `p = Pool(); p.map(f, range(10)); p.close()` (missing join).",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Write a regular expression to match a valid IPv4 address.",
    "complexity": 0.48,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate 'Please sign here' to Russian.",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Summarize the concept of 'copy‑on‑write' in operating systems.",
    "complexity": 0.37,
    "tier": "medium",
    "domain": "summarization",
    "model": "deepseek-chat"
  },
  {
    "query": "Solve `dy/dx = y` with initial condition y(0)=1.",
    "complexity": 0.3,
    "tier": "medium",
    "domain": "math",
    "model": "gpt-4o"
  },
  {
    "query": "Explain the Doppler effect with an example.",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "science",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a tanka (5 lines) about a broken build pipeline.",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "What is the difference between `INNER JOIN` and `LEFT JOIN`?",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "factual",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Write a Python script that watches a directory for new files and processes them.",
    "complexity": 0.56,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Fix the off‑by‑one error: `for i in range(1, len(arr)): if arr[i] > arr[i-1]:` – correct.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Implement a Bloom filter in Python (simple version).",
    "complexity": 0.6,
    "tier": "hard",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Explain why `'5' - 3` works but `'5' + 3` gives different results in JavaScript.",
    "complexity": 0.31,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Write a Golang function that reads a file line by line.",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "code",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'This is a confidential document' to German.",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Summarize the key differences between TCP and UDP.",
    "complexity": 0.24,
    "tier": "easy",
    "domain": "summarization",
    "model": "gpt-4o"
  },
  {
    "query": "Solve the eigenvalue problem for matrix `[[2, 1], [1, 2]]`.",
    "complexity": 0.53,
    "tier": "medium",
    "domain": "math",
    "model": "llama3.1:70b"
  },
  {
    "query": "Explain the double‑slit experiment in quantum mechanics.",
    "complexity": 0.58,
    "tier": "hard",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a sonnet about a race condition (14 lines).",
    "complexity": 0.63,
    "tier": "hard",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "What is the purpose of `__slots__` in Python?",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "factual",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a Rust function that takes a string and returns the first word.",
    "complexity": 0.32,
    "tier": "medium",
    "domain": "code",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Fix the SQL injection in this PHP code: `$query = \"SELECT * FROM users WHERE id = $_GET[id]\";`",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "code",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Implement a priority queue using a binary heap in Java.",
    "complexity": 0.55,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Why does `a = []; a.append(a)` cause infinite recursion when printed? Explain.",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "code",
    "model": "llama3.1:8b"
  },
  {
    "query": "Write a JavaScript function that deep freezes an object.",
    "complexity": 0.46,
    "tier": "medium",
    "domain": "code",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'Temperature exceeds safe limit' to Japanese.",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Summarize the plot of 'The Metamorphosis' by Kafka.",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "summarization",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Solve `lim x→0 (sin x)/x`.",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "math",
    "model": "mistral-small-latest"
  },
  {
    "query": "Explain why the sky appears blue (Rayleigh scattering).",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "science",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Write a seven‑line poem about a memory leak.",
    "complexity": 0.38,
    "tier": "medium",
    "domain": "creative",
    "model": "llama3.2:3b"
  },
  {
    "query": "What is the difference between a thread and a coroutine?",
    "complexity": 0.43,
    "tier": "medium",
    "domain": "factual",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Write a Python script that converts a CSV to JSON.",
    "complexity": 0.31,
    "tier": "medium",
    "domain": "code",
    "model": "deepseek-chat"
  },
  {
    "query": "Fix the race condition in this Python async code: `async def f(): global x; x+=1` (use lock).",
    "complexity": 0.51,
    "tier": "medium",
    "domain": "code",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'The quick brown fox jumps over the lazy dog' to Spanish.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Paraphrase this sentence: 'Despite the rain, the event was a huge success.'",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'Where is the nearest metro station?' to French.",
    "complexity": 0.1,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "Correct the grammar: 'He don't know nothing about that.'",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'I would like to order a vegetarian pizza' to Italian.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Rewrite this in formal English: 'Hey, can you send me the doc ASAP?'",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'The system is under maintenance' to German.",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:8b"
  },
  {
    "query": "Convert this passive voice to active: 'The report was written by John.'",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "translation",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Translate 'What is your name?' to Mandarin Chinese (pinyin).",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Paraphrase the idiom 'It's raining cats and dogs' into plain English.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'Please call back later' to Japanese.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Change this sentence to past tense: 'I go to the gym every day.'",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Translate 'The price does not include tax' to Portuguese.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Summarize this paragraph in 10 words: 'Machine learning is a subset of artificial intelligence that enables systems to learn from data.'",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "summarization",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate 'I have a meeting at 3 PM' to Russian.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Explain the difference between 'affect' and 'effect' with examples.",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "factual",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'Your session has expired' to Arabic.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Rewrite this sentence more concisely: 'Due to the fact that it was raining, we canceled the picnic.'",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'Can you help me with this problem?' to Hindi (Romanized).",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Change this to a question: 'She knows the answer.'",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "Translate 'The file is corrupted' to Korean.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Paraphrase this technical sentence: 'Authentication via OAuth 2.0 provides delegated access.'",
    "complexity": 0.34,
    "tier": "medium",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Translate a short poem from English to French (preserve rhyme).",
    "complexity": 0.45,
    "tier": "medium",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Detect the language of this text: 'Bonjour, comment ça va?' and translate to English.",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Write a grammatically correct sentence using 'their', 'there', and 'they're'.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "creative",
    "model": "llama3.1:8b"
  },
  {
    "query": "Translate 'The application has encountered an unexpected error' to German.",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Rewrite this in the imperative mood: 'You should read the instructions carefully.'",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "translation",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'I apologize for the inconvenience' to Spanish.",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Convert this direct speech to indirect: He said he would come tomorrow.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate 'Please proceed to gate B12' to Japanese.",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Simplify this legalese: 'The party of the first part shall indemnify the party of the second part.'",
    "complexity": 0.41,
    "tier": "medium",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'The deadline is Friday at 5 PM' to Italian.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Explain what a 'malapropism' is and give an example.",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "factual",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate a business email subject: 'Q3 Financial Results Attached' to French.",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Change this to future perfect tense: 'I finish the project.'",
    "complexity": 0.1,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "Translate 'The server will reboot in 5 minutes' to Portuguese.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Paraphrase this scientific abstract (2 sentences) for a general audience.",
    "complexity": 0.39,
    "tier": "medium",
    "domain": "summarization",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate 'Your payment was successful' to Russian.",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Rewrite this sentence in the negative form: 'Everyone attended the meeting.'",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "translation",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Translate 'Please do not touch the glass' to Mandarin.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Correct the spelling: 'recieve', 'seperate', 'definately'",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.1:8b"
  },
  {
    "query": "Translate 'The operation was completed successfully' to Arabic.",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Convert this bullet list into a coherent paragraph.",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "summarization",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'What time does the store close?' to German.",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Explain the difference between 'who' and 'whom' with examples.",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "factual",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate 'This feature is not yet implemented' to Spanish.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Rewrite this sentence using a simile: 'Her voice was loud.'",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "creative",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'Please verify your email address' to Japanese.",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "Change this from first person to third person: 'I think this solution is optimal.'",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Translate a chat message: 'BRB, gonna grab coffee' to formal English.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Paraphrase 'The new update includes several security patches' without changing meaning.",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Translate 'The connection has timed out' to Italian.",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Write a haiku about the translator's dilemma.",
    "complexity": 0.31,
    "tier": "medium",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'Access denied' to French.",
    "complexity": 0.06,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Explain the concept of 'cognates' in linguistics with examples.",
    "complexity": 0.35,
    "tier": "medium",
    "domain": "science",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate a short warning: 'High voltage, risk of electric shock' to Spanish.",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:70b"
  },
  {
    "query": "Rewrite this in colloquial English: 'I am extremely fatigued.'",
    "complexity": 0.1,
    "tier": "trivial",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'The package will arrive within 2 business days' to German.",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Change this sentence to exclamatory: 'It is a beautiful day.'",
    "complexity": 0.05,
    "tier": "trivial",
    "domain": "translation",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Translate 'Please enter your password' to Korean.",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Summarize the difference between British and American English spelling (5 examples).",
    "complexity": 0.27,
    "tier": "easy",
    "domain": "summarization",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'I don't understand this instruction' to Portuguese.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:8b"
  },
  {
    "query": "Paraphrase this corporate jargon: 'We'll circle back on that action item.'",
    "complexity": 0.22,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'Your account has been locked' to Arabic.",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Write a sentence using the word 'ubiquitous' correctly.",
    "complexity": 0.12,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'The meeting was rescheduled to Monday' to Italian.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Change this from active to passive: 'The chef cooked a delicious meal.'",
    "complexity": 0.08,
    "tier": "trivial",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'Please wait while we process your request' to Japanese.",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Explain the term 'portmanteau' and give three examples.",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "factual",
    "model": "llama3.2:3b"
  },
  {
    "query": "Translate a legal disclaimer: 'Not responsible for lost or stolen items' to French.",
    "complexity": 0.33,
    "tier": "medium",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Rewrite this sentence without using the word 'very': 'She was very tired.'",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "translation",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'Your session will expire in 10 minutes' to Russian.",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "translation",
    "model": "deepseek-chat"
  },
  {
    "query": "Convert this to a rhetorical question: 'You should know better.'",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'The file size exceeds the limit' to Spanish.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Summarize the plot of 'The Tower of Babel' story in two sentences.",
    "complexity": 0.25,
    "tier": "easy",
    "domain": "summarization",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Translate 'Please accept our sincere apologies' to German.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:70b"
  },
  {
    "query": "Paraphrase this proverb: 'A bird in the hand is worth two in the bush.'",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Translate 'The system will restart automatically' to Mandarin.",
    "complexity": 0.18,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Fix the subject‑verb agreement: 'The list of items are on the table.'",
    "complexity": 0.07,
    "tier": "trivial",
    "domain": "translation",
    "model": "llama3.2:3b"
  },
  {
    "query": "Translate 'I've attached the document for your review' to Italian.",
    "complexity": 0.17,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-3.5-turbo"
  },
  {
    "query": "Explain the difference between denotation and connotation with examples.",
    "complexity": 0.29,
    "tier": "easy",
    "domain": "factual",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Translate 'Your request has been received' to Korean.",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  },
  {
    "query": "Rewrite this sentence using alliteration: 'The dog ran fast.'",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "creative",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate 'The store is closed on Sundays' to Portuguese.",
    "complexity": 0.13,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-5-haiku-20241022"
  },
  {
    "query": "Change this from singular to plural: 'The child is playing.'",
    "complexity": 0.04,
    "tier": "trivial",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Translate 'Please note that prices are subject to change' to French.",
    "complexity": 0.24,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:8b"
  },
  {
    "query": "Paraphrase this headline: 'Tech Giant Announces Record Profits Amid Layoffs'",
    "complexity": 0.26,
    "tier": "easy",
    "domain": "summarization",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'The shipment is delayed due to weather' to German.",
    "complexity": 0.19,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-small-latest"
  },
  {
    "query": "Write a sentence that is grammatically correct but semantically nonsensical.",
    "complexity": 0.23,
    "tier": "easy",
    "domain": "creative",
    "model": "claude-3-haiku-20240307"
  },
  {
    "query": "Translate 'Your feedback is valuable to us' to Japanese.",
    "complexity": 0.16,
    "tier": "easy",
    "domain": "translation",
    "model": "gemini-1.5-flash"
  },
  {
    "query": "Explain what a 'double negative' is and why it's often avoided in standard English.",
    "complexity": 0.2,
    "tier": "easy",
    "domain": "factual",
    "model": "deepseek-chat"
  },
  {
    "query": "Translate 'The server is experiencing high load' to Arabic.",
    "complexity": 0.21,
    "tier": "easy",
    "domain": "translation",
    "model": "gpt-4o"
  },
  {
    "query": "Rewrite this sentence as a conditional: 'You didn't water the plant, so it died.'",
    "complexity": 0.14,
    "tier": "easy",
    "domain": "translation",
    "model": "llama3.1:70b"
  },
  {
    "query": "Translate 'Congratulations on your promotion' to Spanish.",
    "complexity": 0.11,
    "tier": "easy",
    "domain": "translation",
    "model": "claude-3-5-sonnet-20241022"
  },
  {
    "query": "Summarize the concept of 'code‑switching' in linguistics in one sentence.",
    "complexity": 0.28,
    "tier": "easy",
    "domain": "science",
    "model": "gpt-4o-mini"
  },
  {
    "query": "Translate 'Please do not disturb' to Italian.",
    "complexity": 0.09,
    "tier": "trivial",
    "domain": "translation",
    "model": "gemini-1.5-pro"
  },
  {
    "query": "Convert this sentence to use the subjunctive mood: 'I wish I was there.' (correction)",
    "complexity": 0.15,
    "tier": "easy",
    "domain": "translation",
    "model": "mistral-large-latest"
  }
]