Module telemetry_source 

Wire-format identifiers for ProviderTelemetry::source. Keep these in sync with the matching strings in docs/src/observability/* and the The host eval aggregator.

Constants

ANTHROPIC_USAGE

Anthropic Messages API — usage counts only; no timings.

GEMINI_USAGE

Google Gemini usageMetadata block from generateContent.

LLAMACPP_TIMINGS

llama.cpp usage.timings extension (prompt_ms, predicted_ms, predicted_n, prompt_n, …). Preserved verbatim from the non-OpenAI subset.

OLLAMA_CHAT

Ollama /api/chat NDJSON stream — full timing breakdown.

OLLAMA_GENERATE

Ollama /api/generate (raw) — full timing breakdown.

OPENAI_USAGE

OpenAI-style usage block (prompt/completion tokens, optional cache details). No server-side timings unless the runtime extends the schema.

UNKNOWN

Provider responded but we did not capture anything beyond what already lives on LlmResult (e.g. mock / fake providers, or a stream that finished without a usage frame).