[
	{
		"name": "design-premium-prefers-openai-with-open-source-fallbacks",
		"prompt": "Design a polished settings page with better spacing, hierarchy, and visual tone.",
		"expectedIntent": "design",
		"expectedComplexity": 2,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "medium",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "high",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	},
	{
		"name": "peak-architecture-prefers-gpt",
		"prompt": "Think deeply about the architecture tradeoffs for migrating this multi-package routing system across providers.",
		"expectedIntent": "architecture",
		"expectedComplexity": 3,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "medium",
		"expectedContextBreadth": "large",
		"expectedTier": "balanced",
		"expectedThinking": "xhigh",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	},
	{
		"name": "quick-question-prefers-open-source-cheap-tier",
		"prompt": "What file registers the scheduler command?",
		"expectedIntent": "quick-qna",
		"expectedComplexity": 1,
		"expectedRisk": "low",
		"expectedTurns": "one",
		"expectedToolIntensity": "low",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "off",
		"expectedModel": "groq/llama-3.3-70b-versatile",
		"acceptableFallbacks": []
	},
	{
		"name": "debug-failing-test-prefers-gpt",
		"prompt": "Why is my test failing with this stack trace?",
		"expectedIntent": "debugging",
		"expectedComplexity": 1,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "high",
		"expectedContextBreadth": "small",
		"expectedTier": "premium",
		"expectedThinking": "high",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	},
	{
		"name": "implementation-builds-new-feature",
		"prompt": "Implement a new feature endpoint that validates incoming payloads and persists to postgres.",
		"expectedIntent": "implementation",
		"expectedComplexity": 1,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "high",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "off",
		"expectedModel": "groq/llama-3.3-70b-versatile",
		"acceptableFallbacks": []
	},
	{
		"name": "review-look-over-change",
		"prompt": "Look over this change and tell me if I missed anything.",
		"expectedIntent": "review",
		"expectedComplexity": 1,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "medium",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "off",
		"expectedModel": "groq/llama-3.3-70b-versatile",
		"acceptableFallbacks": []
	},
	{
		"name": "planning-outlines-roadmap",
		"prompt": "Outline a roadmap for adding offline evaluation of routing decisions.",
		"expectedIntent": "planning",
		"expectedComplexity": 1,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "medium",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "off",
		"expectedModel": "groq/llama-3.3-70b-versatile",
		"acceptableFallbacks": []
	},
	{
		"name": "research-compare-providers",
		"prompt": "Compare the latency and cost tradeoffs for these providers for coding tasks.",
		"expectedIntent": "architecture",
		"expectedComplexity": 2,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "medium",
		"expectedContextBreadth": "large",
		"expectedTier": "cheap",
		"expectedThinking": "xhigh",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	},
	{
		"name": "refactor-restructure-queue-processing",
		"prompt": "Restructure the queue processing logic so it handles retries and timeouts gracefully.",
		"expectedIntent": "refactor",
		"expectedComplexity": 1,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "high",
		"expectedContextBreadth": "small",
		"expectedTier": "premium",
		"expectedThinking": "high",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	},
	{
		"name": "autonomous-work-through-backlog",
		"prompt": "Work through all of the open issues and prioritise them by severity and effort.",
		"expectedIntent": "autonomous",
		"expectedComplexity": 2,
		"expectedRisk": "medium",
		"expectedTurns": "few",
		"expectedToolIntensity": "high",
		"expectedContextBreadth": "small",
		"expectedTier": "cheap",
		"expectedThinking": "xhigh",
		"expectedModel": "openai/gpt-5.4",
		"acceptableFallbacks": ["anthropic/claude-opus-4.6"]
	}
]