switchboard-100.pages.dev

get_provenance

Explain where the numbers on this dashboard come from and how far they can be trusted. Call this before making a recommendation so you can state the limits of your evidence. Returns the measurement date, the run and call counts, and per-model sample sizes. Read samplingCaveat: the measurements are spread across a whole catalogue on a single structured-output task, so per-model sample counts are small — treat medianLatencyMs as indicative and p95LatencyMs as coarse, and say so when a recommendation turns on a latency margin of a second or less. Every field here is prose and counts for you to paraphrase, not a machine-readable prohibition list.

Input schema

No input properties (empty object).

{
  "type": "object",
  "properties": {}
}