Markdown source

Agent Evaluations Credibility

Use Case

Rank people or sources for eval design, observability, review loops, and production agent quality.

View Signal Role

Low. Eval credibility comes from repeatable artifacts, production failure coverage, and trace evidence.

Scoring Weights

| Signal | Weight |

| --- | ---: |

| `topic_fit` | 15 |

| `domain_practice` | 25 |

| `evaluation_artifacts` | 25 |

| `production_scale` | 15 |

| `source_depth` | 15 |

| `public_attention` | 5 |

Evaluation Fixtures

| Name | Score | Expected Min | Result |

| --- | ---: | ---: | --- |

| Aparna Dhinakaran | 91.75 | 84 | pass |

| Laurie Voss | 87.25 | 82 | pass |

Policy Boundary

This policy scores credibility for one topic and use case. Do not reuse it for unrelated topics where public fame, domain credentials, implementation depth, or primary-source evidence should be weighted differently.

Change Rule

Change one policy file at a time, rerun python3 scripts/generate_synthesis_layers.py, and inspect credibility policy evals before applying new scores.