# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-140137
**Date:** Apr 02, 2026
**Category:** code
**Question ID:** CODE-020

---

## Question

Refactor this 'working but unmaintainable' code into clean, testable, well-structured code. Explain every design decision.

```python
def process(data, mode, flag1=False, flag2=True, output_type='json'):
    results = []
    for item in data:
        if mode == 'fast':
            if item.get('type') == 'A':
                val = item['value'] * 1.1 if flag1 else item['value']
                if val > 100:
                    if flag2:
                        results.append({'id': item['id'], 'val': val, 'status': 'high'})
                    else:
                        results.append({'id': item['id'], 'val': val * 0.9, 'status': 'adjusted'})
                else:
                    results.append({'id': item['id'], 'val': val, 'status': 'normal'})
            elif item.get('type') == 'B':
                val = item['value'] * 0.95
                results.append({'id': item['id'], 'val': val, 'status': 'discounted'})
            else:
                if flag1 and flag2:
                    results.append({'id': item['id'], 'val': 0, 'status': 'skip'})
        elif mode == 'careful':
            try:
                val = float(item.get('value', 0))
                if val < 0: raise ValueError('negative')
                results.append({'id': item['id'], 'val': val, 'status': 'validated'})
            except:
                results.append({'id': item['id'], 'val': 0, 'status': 'error'})
    if output_type == 'json':
        import json; return json.dumps(results)
    elif output_type == 'csv':
        return '\n'.join([f"{r['id']},{r['val']},{r['status']}" for r in results])
    return results
```

---

## Winner

**GPT-5.4** (openrouter)
- Winner Score: 9.18
- Matrix Average: 8.36
- Total Judgments: 69

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | MiniMax M2.5 | openrouter | 10.00 | 1 |
| 2 | GPT-5.4 | openrouter | 9.18 | 8 |
| 3 | Gemini 3 Flash Preview | Google | 8.96 | 7 |
| 4 | MiMo-V2-Flash | Xiaomi | 8.50 | 9 |
| 5 | DeepSeek V4 | openrouter | 8.46 | 9 |
| 6 | Claude Opus 4.6 | openrouter | 8.02 | 9 |
| 7 | Claude Sonnet 4.6 | openrouter | 7.99 | 8 |
| 8 | GPT-OSS-120B | OpenAI | 7.80 | 9 |
| 9 | Gemini 3.1 Pro | openrouter | 6.36 | 9 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | DeepSeek V4 | GPT-OSS-120B | MiMo-V2-Flash | GPT-5.4 | Claude Opus | Gemini 3.1 Pro | Claude Sonnet | Grok 4.20 | Gemini 3 | MiniMax M2.5 |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 | — | 9.8 | 9.8 | 9.8 | 9.8 | 8.6 | 10.0 | · | 9.8 | · |
| GPT-OSS-120B | 8.3 | — | 8.6 | 8.8 | 6.6 | 8.0 | 6.2 | · | · | · |
| MiMo-V2-Flash | 9.2 | 7.8 | — | 9.2 | 8.8 | 8.0 | 9.3 | · | 9.3 | 10.0 |
| GPT-5.4 | 6.3 | 3.8 | 6.2 | — | 4.3 | 2.0 | 4.8 | · | 8.0 | · |
| Claude Opus | 8.0 | 8.2 | 8.8 | 9.2 | — | 4.9 | 8.2 | · | 9.0 | · |
| Gemini 3.1 Pro | 8.2 | 7.5 | 8.0 | · | 7.2 | — | 7.5 | · | · | · |
| Claude Sonnet | 8.8 | 8.2 | 9.0 | 9.0 | 9.2 | 7.9 | — | · | 9.2 | · |
| Grok 4.20 | 8.8 | 8.7 | 8.8 | 9.0 | 8.8 | 7.9 | · | — | 8.8 | · |
| Gemini 3 | 9.8 | 9.6 | 9.6 | 9.6 | 10.0 | 6.7 | 9.8 | · | — | · |
| MiniMax M2.5 | 8.8 | 6.7 | 7.8 | 8.8 | 7.5 | 3.1 | 8.1 | · | 8.6 | — |

---

## 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: CODE-020. 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-20260402-140137/results
Full dataset: https://app.themultivac.com/dashboard/export