ExperimentArmResult
One arm's results (1.55.0): Calls and Turns, ended reasons, the transfer and Tool error rates with intervals, Policy firings per 100 Turns (all and compliance-tagged), duration and voice-to-voice latency.
An object with 14 fields.
Fields
| Field | Type | Required | Description |
|---|---|---|---|
agent_version | integer or null | yes | |
arm | ExperimentArmKey | yes | |
calls | integer | yes | The finalized Calls the arm received. |
compliance_firings_per_100_turns | number or null | yes | |
duration_ms | ExperimentDistribution | yes | |
ended_reasons | array of PromptEvalEndedReasonCount | yes | |
goal | PromptEvalRate or null | no | The arm's finalized Calls that met the experiment's goal (1.63.0), with its interval and sample size; null when the experiment has no goal. Absent from an api before 1.63.0. |
judged | ExperimentJudged or null | yes | Null until live Calls can be judged under live_call_judging; from 1.63.0 the judgments of the arm's sampled Calls when the experiment names a suite and the profile has it on, null otherwise. |
policy_firings_per_100_turns | number or null | yes | |
share_percent | integer | yes | |
tool_error_rate | PromptEvalRate | yes | |
transfer_rate | PromptEvalRate | yes | |
turns | integer | yes | |
voice_to_voice_ms | ExperimentDistribution | yes |