id int64 1 12 | session_hash large_stringlengths 36 36 | tool_a_id large_stringclasses 2
values | tool_b_id large_stringclasses 2
values | chosen large_stringclasses 2
values | llm_id large_stringclasses 2
values | task large_stringclasses 2
values | goal large_stringclasses 2
values | timestamp large_stringdate 2026-04-10 10:37:42 2026-04-10 13:22:21 | competitor_type large_stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|
1 | 6f772c9f-9dae-4d49-8972-956d68d04907 | llamaindex_rag | langchain_rag | a | openrouter/google/gemma-3-4b-it:free | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T10:37:42.830615 | tool |
2 | f46fdeaa-7067-4272-a754-a4e0fca098be | llamaindex_rag | langchain_rag | a | openrouter/google/gemma-3-4b-it:free | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T10:44:35.275104 | tool |
3 | 1b3e6815-b716-4523-a5c6-780fb1039e1c | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T10:53:52.466172 | tool |
4 | 36a53241-a0bc-48df-a843-e0ae35f422be | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T10:55:38.798318 | tool |
5 | 64bda38c-6b73-4c39-b65f-29d347d4426c | llamaindex_rag | langchain_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T11:15:14.058075 | tool |
6 | e1e7bca9-fce9-40c2-887b-bc2797667f34 | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the key concepts of retrieval-augmented generation | A concise 3-sentence summary that covers indexing, retrieval, and generation steps | 2026-04-10T11:18:18.397315 | tool |
7 | 9a522f84-f787-4126-b4f1-a556d5c31316 | langchain_rag | llamaindex_rag | b | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T11:33:01.467988 | tool |
8 | 2401b163-06db-4d93-b7da-9825245e5e87 | llamaindex_rag | langchain_rag | b | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T11:40:14.880607 | tool |
9 | 9760920d-9c65-4ac0-8de6-7e28a0e1af8c | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T11:41:46.547588 | tool |
10 | 128d0078-7082-49f2-9744-2a9fca51306a | llamaindex_rag | langchain_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T11:42:58.461563 | tool |
11 | 8ba32eee-fa3c-42b8-b319-80f1348c3c7f | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T11:43:11.481040 | tool |
12 | ccdebe08-0714-4470-8416-c86d9ac69f92 | langchain_rag | llamaindex_rag | a | openrouter/mistralai/mistral-small-3.1-24b-instruct | Summarize the following document or content clearly and concisely, capturing the key points. | A concise, accurate summary that captures all key points | 2026-04-10T13:22:22.592288 | tool |
CompaRAG — Tool Votes Dataset
CompaRAG is a blind comparison platform for MCP (Model Context Protocol) tools, built by The Borges Graph as a spinoff of comparIA, a French government initiative (Ministère de la Culture / Beta.gouv.fr).
What is this dataset?
This dataset contains human preference votes collected on the CompaRAG platform. Users submit a task and a goal, two MCP tools respond anonymously, and the user votes for the best result — without knowing which tool produced which answer.
This is the MCP-tool equivalent of RLHF preference data: real tasks, real users, blind evaluation.
Dataset structure
| Column | Description |
|---|---|
id |
Unique vote ID |
session_hash |
Anonymous session identifier |
tool_a_id |
Name of tool A (revealed after vote) |
tool_b_id |
Name of tool B (revealed after vote) |
chosen |
User's choice: a, b, or tie |
llm_id |
LLM used as mediator (constant across both calls) |
task |
The task submitted by the user |
goal |
The success criterion defined by the user |
timestamp |
When the vote was recorded |
competitor_type |
Type of competition (tool) |
Methodology
- Blind evaluation: tool identities are hidden during voting, revealed only after
- Equifinality principle: same task, same goal, same LLM — only the tool varies
- Real tasks: user-submitted, not synthetic benchmarks
- Bradley-Terry scoring: votes feed into a statistically robust leaderboard
License
Licence Ouverte / Open Licence 2.0 (Etalab)
Citation
@dataset{comparag_tool_votes_2025,
author = {The Borges Graph},
title = {CompaRAG Tool Votes — Blind MCP Tool Preference Dataset},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/ArthurSrz/comparag-tool-votes}
}
- Downloads last month
- 29