# The Multivac — Evaluation Report

**Evaluation ID:** EVAL-20260402-132144
**Date:** Apr 02, 2026
**Category:** code
**Question ID:** CODE-014

---

## Question

Implement a Last-Writer-Wins Element Set (LWW-Element-Set) CRDT in Python. It should support add, remove, lookup, and merge operations. Include a proof that merge is commutative, associative, and idempotent. Write tests demonstrating conflict resolution between two divergent replicas.

---

## Winner

No winner determined.

---

## Rankings

| Rank | Model | Provider | Avg Score | Judgments |
|------|-------|----------|-----------|----------|
| 1 | Gemini 3 Flash Preview | Google | 8.97 | 6 |
| 2 | GPT-5.4 | openrouter | 8.75 | 5 |
| 3 | DeepSeek V4 | openrouter | 8.49 | 4 |
| 4 | GPT-OSS-120B | OpenAI | 8.17 | 5 |
| 5 | MiMo-V2-Flash | Xiaomi | 8.00 | 5 |
| 6 | Claude Sonnet 4.6 | openrouter | 7.48 | 5 |
| 7 | Claude Opus 4.6 | openrouter | 7.39 | 4 |
| 8 | Gemini 3.1 Pro | openrouter | 4.19 | 4 |

---

## 10×10 Judgment Matrix

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

| Judge ↓ / Resp → | MiMo-V2-Flash | Gemini 3.1 Pro | Claude Sonnet | Grok 4.20 | DeepSeek V4 | GPT-OSS-120B | MiniMax M2.5 | Gemini 3 | GPT-5.4 | Claude Opus |
|---|---|---|---|---|---|---|---|---|---|---|
| MiMo-V2-Flash | — | 1.6 | 7.6 | · | 8.8 | 8.6 | · | 8.6 | 8.8 | · |
| Gemini 3.1 Pro | 7.9 | — | 6.0 | · | 8.9 | 6.8 | · | 9.8 | · | · |
| Claude Sonnet | 8.0 | 5.1 | — | · | · | 8.8 | · | 8.8 | 8.7 | 7.7 |
| Grok 4.20 | 7.7 | · | 8.6 | — | 7.4 | · | · | 8.6 | 8.6 | 8.4 |
| DeepSeek V4 | 8.8 | 7.6 | 8.6 | · | — | 9.6 | · | 9.6 | 9.4 | 8.8 |
| GPT-OSS-120B | · | 2.4 | 6.6 | · | 8.8 | — | · | · | 8.3 | 4.7 |
| MiniMax M2.5 | 7.7 | · | · | · | · | 7.1 | — | 8.4 | · | · |
| Gemini 3 | · | · | · | · | · | · | · | — | · | · |
| GPT-5.4 | · | · | · | · | · | · | · | · | — | · |
| Claude Opus | · | · | · | · | · | · | · | · | · | — |

---

## 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-014. 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-20260402-132144/results
Full dataset: https://app.themultivac.com/dashboard/export