---
title: "The production judge's verdict on one showcase transcript (public, rate-limited)"
method: POST
path: "/api/demo/judge"
tags: ["Testing"]
---

# The production judge's verdict on one showcase transcript (public, rate-limited)

`POST /api/demo/judge`

## Request

- Base URL: `https://api.wordis-bond.com`
- URL: `https://api.wordis-bond.com/api/demo/judge`
- Auth: HTTP bearer

Returns the temperature-0 judge's scorecard for a golden-corpus transcript beside the committed calibration baseline for the serving tier. By default it serves the latest verdict from the daily calibration run (judged within the last 26 hours; `source: 'daily-canary'`, with `judgedAt`) and makes no judge call; with `?live=1` — or when no usable verdict from the last 26 hours exists (none was judged, it came from a different judge version, or the lookup failed) — the judge scores it now (`source: 'live'`). Run it live twice and watch the verdict repeat. Only allowlisted caseIds are accepted; visitor text never reaches the judge. Rate-limited per IP; live judge calls also share a global ceiling.

## Query parameters

- `live` '1'

## Request body

- object
  - `caseId` 'clear-pass-booking' | 'polite-but-wrong' | 'caller-frustrated-but-resolved', required

## Response `200`

Scorecard with judge provenance, plus baseline, tolerance, `source` ('daily-canary' | 'live') and `judgedAt` (when the judge produced this verdict).

## Other responses

- `400`: Unknown caseId or malformed body.
- `429`: Per-IP demo limit, or the global live-judging ceiling, reached.
- `503`: No LLM tier reachable — never a fabricated verdict.

---

[API](https://skmtc.dev/wordis-bond/apis/word-is-bond-voice-agent-testing-api.md) · [All operations](https://skmtc.dev/wordis-bond/apis/word-is-bond-voice-agent-testing-api/llms.txt) · [OpenAPI document](https://skmtc.dev/wordis-bond/apis/word-is-bond-voice-agent-testing-api/revisions/66c2d5480d84?raw)