[bot] Google GenAI batch generation (`:batchGenerateContent`) requests are not traced

[braintrust-bot]

<!-- provider-gap-audit: genai-batchgeneratecontent-not-traced -->

What instrumentation is missing

The Gemini API's Batch Mode endpoint (POST .../models/{model}:batchGenerateContent) submits an asynchronous batch generation job — a real model-execution surface (50% cheaper, higher-throughput content generation), not a CRUD/admin endpoint. This repo's GenAI tracing middleware does not recognize or trace this endpoint at all: requests to it pass through the wrapped http.Client completely untraced.

Where it breaks

trace/contrib/genai/tracegenai.go, genaiRouter (lines 143-157) dispatches based on containsGenerateContent/containsEmbedContent:

func containsGenerateContent(path string) bool {
	return strings.Contains(path, ":generateContent") ||
		strings.Contains(path, "/generateContent") ||
		strings.Contains(path, ":streamGenerateContent") ||
		strings.Contains(path, "/streamGenerateContent")
}

A batch request path is .../models/{model}:batchGenerateContent. The substring ":generateContent" requires a colon immediately before generateContent, but in :batchGenerateContent the colon precedes batch, not generateContent — so this check (and the /generateContent check) both return false. containsEmbedContent doesn't match it either. genaiRouter therefore returns nil and no span is created for batch generation requests.

Braintrust docs status: not_found

https://www.braintrust.dev/docs/integrations/ai-providers/gemini explicitly enumerates the traced Go methods as GenerateContent/GenerateContentStream plus embedding calls (EmbedContent/batchEmbedContents) — batch content generation is not listed as traced for any language on this page.

Upstream sources

• Gemini API Batch Mode / batchGenerateContent endpoint reference: https://ai.google.dev/api/batch-api
• Gemini API Batch Mode guide: https://ai.google.dev/gemini-api/docs/batch-mode
• Google Developers Blog announcement: https://developers.googleblog.com/scale-your-ai-workloads-batch-mode-gemini-api/

Local repo files inspected

• trace/contrib/genai/tracegenai.go — genaiRouter, containsGenerateContent, containsEmbedContent, isStreamingPath, extractModelFromPath (lines 142-212)
• trace/contrib/genai/generatecontent.go — generation tracer, only invoked for paths already matched by the router
• trace/contrib/genai/embedcontent.go — embedding tracer, confirms batchEmbedContents is handled while no equivalent batch-generation tracer exists