inference

type Complete func(context.Context, Request) (string, error)

type OutputThroughput struct {
	OutputTokens     uint64 `json:"output_tokens"`
	DurationNanos    uint64 `json:"duration_ns"`
	MeasuredRequests uint64 `json:"measured_requests"`
}

type Request struct {
	Model, Effort, Instructions, Input string
}

type Usage struct {
	// TotalsKnown distinguishes reported zero totals from absent totals.
	TotalsKnown         bool
	InputTokens         uint64
	UncachedInputTokens uint64
	CacheWriteTokens    uint64
	OutputTokens        uint64
	ReasoningTokens     uint64

	// Preserve inconsistent raw categories before normalizing input for existing consumers.
	Inconsistent bool
	// Missing categories remain zero for existing normalized capture consumers,
	// but cannot establish complete thread totals or a reference cost.
	Incomplete bool
	// The terminal provider tier overrides the requested tier for pricing only.
	ServiceTier string
}