Cut Noise in Gemini 1M-Token Prompts With Scope Limits

Google Cloud documentation notes that while flagship Gemini models support a 1-million-token context window, long context sequences can introduce noise, latency, and context degradation. Explicitly scope which sections of the context the model should attend to at the bottom of the prompt.

<context>{{MONOREPO_SOURCE_DUMP}}</context> <task>Based on the information above, trace the payment retry state machine. Focus strictly on files under src/billing/ and ignore deprecated v1 modules to avoid context noise.</task>