medium10 pts · first answer countsllm-observabilitymonitoring

What to log for each LLM call

For an LLM feature in production, what should each request trace capture?

A.Only the HTTP status codeB.The full raw prompt including user personal data, kept foreverC.Nothing, to save storageD.Prompt template version, model, token counts, latency, retrieved document IDs and tool calls, with personal data redacted

← Splitting requests into batches Fleet-wide p99 from many pods →