# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260315-055905
**Date:** Mar 15, 2026
**Category:** reasoning
**Question ID:** EVAL-20260315-055905

---

## Question

A disease affects 1 in 10,000 people. A test is 99% sensitive (true positive rate) and 99.5% specific (true negative rate). A patient tests positive. (1) What is the probability they have the disease? (2) If they test positive twice with independent tests, what is the probability? (3) A doctor says 'You tested positive, so you almost certainly have it.' Critique this reasoning. (4) Design a testing protocol that achieves >95% positive predictive value.

---

## Winner

**Gemma 3 27B** (openrouter)
- Winner Score: 9.59
- Matrix Average: 8.79
- Total Judgments: 65

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Gemma 3 27B | openrouter | 9.59 | 8 |
| 2 | Qwen 3 8B | openrouter | 9.53 | 8 |
| 3 | Phi-4 14B | openrouter | 9.05 | 8 |
| 4 | Devstral Small | openrouter | 9.02 | 7 |
| 5 | Granite 4.0 Micro | openrouter | 8.94 | 8 |
| 6 | Llama 4 Scout | openrouter | 8.72 | 8 |
| 7 | Mistral Nemo 12B | openrouter | 8.61 | 8 |
| 8 | Qwen 3 32B | openrouter | 8.25 | 1 |
| 9 | Kimi K2.5 | openrouter | 8.25 | 1 |
| 10 | Llama 3.1 8B | openrouter | 7.91 | 8 |

---

## 10×10 Judgment Matrix

Rows = Judge, Columns = Respondent. Self-judgments excluded (—).

| Judge ↓ / Resp → | Qwen 3 32B | Devstral Small | Gemma 3 27B | Llama 4 Scout | Phi-4 14B | Granite 4.0 | Qwen 3 8B | Mistral Nemo | Llama 3.1 8B | Kimi K2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen 3 32B | — | · | 10.0 | 8.4 | 9.0 | 7.8 | 10.0 | 7.5 | 7.6 | · |
| Devstral Small | · | — | 9.8 | 8.1 | 9.8 | 9.4 | 9.8 | 8.6 | 6.7 | · |
| Gemma 3 27B | · | 8.3 | — | 9.3 | 8.3 | 9.3 | 9.8 | 8.8 | 6.8 | · |
| Llama 4 Scout | · | 9.8 | 9.8 | — | 9.8 | 9.4 | 9.8 | 9.4 | 8.3 | · |
| Phi-4 14B | · | 10.0 | 10.0 | 8.9 | — | 10.0 | 10.0 | 9.0 | 9.6 | · |
| Granite 4.0 | 8.3 | 8.8 | 8.8 | 8.8 | 8.8 | — | 8.8 | 8.8 | 8.2 | 8.3 |
| Qwen 3 8B | · | 7.8 | 9.4 | 9.3 | 8.8 | 7.4 | — | 7.4 | 7.5 | · |
| Mistral Nemo | · | 9.7 | 9.1 | 8.3 | 8.8 | 8.7 | 9.2 | — | 8.7 | · |
| Llama 3.1 8B | · | 8.8 | 9.8 | 8.8 | 9.1 | 9.6 | 8.8 | 9.3 | — | · |
| Kimi K2.5 | · | · | · | · | · | · | · | · | · | — |

---

## Methodology

- **10×10 Blind Peer Matrix:** All models answer the same question, then all models judge all responses.
- **5 Criteria:** Correctness, completeness, clarity, depth, usefulness (each scored 1–10).
- **Self-judgments excluded:** Models do not judge their own responses.
- **Weighted Score:** Composite of all 5 criteria.

---

## Citation

The Multivac (2026). Blind Peer Evaluation: EVAL-20260315-055905. app.themultivac.com

## License

Open data. Free to use, share, and build upon. Please cite The Multivac when using this data.

Download raw JSON: https://app.themultivac.com/api/evaluations/EVAL-20260315-055905/results
Full dataset: https://app.themultivac.com/dashboard/export