transcript_policy - Judge API

Overview

Checks transcripts for required talking points and prohibited language. Requires at least one configuration parameter - Provide talking_points or language_to_avoid.

Configuration

config.talking_points

array

Required talking points to include.

config.language_to_avoid

array

Phrases that must not appear.

config.min_talking_points

integer

Minimum number of talking points required.

config.max_language_to_avoid

integer

Maximum number of forbidden phrases allowed.

Example Configuration

{
  "detector_name": "transcript_policy",
  "config": {
    "talking_points": ["battery life", "fast charging"],
    "language_to_avoid": ["free giveaway"],
    "min_talking_points": 1,
    "max_language_to_avoid": 0
  }
}

Result Schema

{
  "detector_name": "transcript_policy",
  "pass_check": true,
  "score": 0.75,
  "rationale": "2/2 talking points covered. No forbidden language detected (1 checked)",
  "metrics": {
    "talking_points_hit": ["battery life", "fast charging"],
    "language_to_avoid_hit": [],
    "talking_points_count": 2,
    "language_to_avoid_count": 0,
    "talking_points_total": 2,
    "language_to_avoid_total": 1,
    "matches": {
      "talking_points": [{"item": "battery life", "texts": ["battery life"]}],
      "language_to_avoid": []
    },
  }
}

Interpreting Results

talking_points_hit/language_to_avoid_hit: Items found (or avoided) in the transcript.
talking_points_count/language_to_avoid_count: Totals used for pass/fail.
matches: Evidence snippets supporting each detected item.

​Overview

​Configuration

​Example Configuration

​Result Schema

​Interpreting Results

Overview

Configuration

Example Configuration

Result Schema

Interpreting Results