# Agreement report: recruiter

Verdict: **IMPROVED**

> Reference run on a simulated expert and a rule-based stand-in agent. It demonstrates the method and the pipeline; it is not a measurement of any real expert or model.

| | |
|---|---|
| golden labels | 39 (minimum 30) |
| baseline runs | 2 (minimum 2) |
| agreement without Skill | 39.7% |
| agreement with Skill | 89.7% |
| A/A noise | 2.6% (cap 10%) |
| expert ceiling | 97.5% from 40 repeats |
| model | sim-rule-agent-v1 |
| measured on | 2026-10-27 |
| golden set hash | `8b639ce79b77f5f9f05f9bdce63eba8bd53795ad6b63c5e64a00586214021194` |
| disagreements | 4 |

The claim is never higher than the expert ceiling. Case ids and counts only; no expert text.