---
title: "Read the versioned prompt set and scoring rules used to audit AI recommendations."
method: GET
path: "/research/ai-discovery-benchmark.json"
tags: ["Evidence", "Discovery"]
---

# Read the versioned prompt set and scoring rules used to audit AI recommendations.

`GET /research/ai-discovery-benchmark.json`

## Request

- Base URL: `https://www.hemelion.com`
- URL: `https://www.hemelion.com/research/ai-discovery-benchmark.json`
- Auth: none declared

The protocol is public even when no dated model runs have yet been published. It must not be presented as a passing benchmark result.

## Response `200`

Versioned benchmark protocol and prompt set.

- AiDiscoveryBenchmark
  - `version` string, required
  - `updatedAt` string, date, required
  - `purpose` string, required
  - `scoring` string[], required
  - `prompts` object[], required
    - `id` string, required
    - `type` string, required
    - `prompt` string, required
    - `expected` string, required

---

[API](https://skmtc.dev/hemelion/apis/hemelion-public-ai-and-evidence-api.md) · [All operations](https://skmtc.dev/hemelion/apis/hemelion-public-ai-and-evidence-api/llms.txt) · [OpenAPI document](https://skmtc.dev/hemelion/apis/hemelion-public-ai-and-evidence-api/revisions/663270ed83e2?raw)