gramajo/nouns-proposals
Nouns DAO Proposals Dataset A dataset of all Nouns DAO governance proposals with pass/fail outcomes and LLM-assigned categories, designed for text classification and DAO governance research. Dataset Description This dataset contains all 983 proposals submitted to Nouns DAO from its inception (August 2021) through July 2026. Each proposal includes its title, full description, on-chain outcome (passed/failed), and a category label assigned by an LLM classifier. Key… See the full description on the dataset page: https://huggingface.co/datasets/gramajo/nouns-proposals.
Nouns DAO Proposals Dataset
A dataset of all Nouns DAO governance proposals with pass/fail outcomes and LLM-assigned categories, designed for text classification and DAO governance research.
Dataset Description
This dataset contains all 983 proposals submitted to Nouns DAO from its inception (August 2021) through July 2026. Each proposal includes its title, full description, on-chain outcome (passed/failed), and a category label assigned by an LLM classifier.
Key Statistics:
- Total proposals: 983 (982 decided, 1 active)
- Passed: 495 (50.4%)
- Failed: 487 (49.6%)
- Train split: 785 proposals (IDs 1–785)
- Test split: 197 proposals (IDs 786–982)
Data Source
All data was collected from the nouns.camp subgraph, which indexes on-chain Nouns DAO governance data. The subgraph provides structured access to proposal metadata, vote counts, and execution state.
Collection Methodology
- Scraping: All 983 proposals were queried from the nouns.camp GraphQL subgraph endpoint
- Status correction: The subgraph's
ACTIVEstatus was corrected for 344 proposals whose voting period had ended — these were reclassified asDEFEATEDbased on on-chain vote totals vs quorum requirements - Category labeling: Each proposal was classified into one of 5 categories using
gpt-4o-minivia OpenRouter API, with a structured prompt (see below)
Categories
Categories were assigned using gpt-4o-mini (OpenRouter) with temperature=0.0. The labeling prompt and model version are documented below for reproducibility.
Pass/Fail Definition
- Passed: On-chain status is
EXECUTEDorQUEUED - Failed: On-chain status is
DEFEATED,CANCELLED, orVETOED - Active: Voting period has not ended (1 proposal as of dataset creation)
Proposals that passed voting but were never executed (expired) are labeled as DEFEATED since they did not result in execution.
Train/Test Split
The split is chronological (not random):
- Train: Proposals 1–785 (August 2021 – early 2025)
- Test: Proposals 786–982 (2025 – July 2026)
This simulates the real-world use case of predicting future proposals from historical data and prevents temporal data leakage.
⚠️ Important: The test set has a significantly lower pass rate (28.4%) than the training set (55.9%), reflecting the DAO becoming more conservative over time. This is an intentional feature of the chronological split, not a bug.
Dataset Format
Two formats are provided:
JSONL (human-readable):
{"id": 1, "title": "Donate 5 ETH each to 6 Charities", "description": "...", "category": "Grants & Funding", "status": "EXECUTED", "passed": true}Parquet (HuggingFace datasets library):
from datasets import load_dataset
dataset = load_dataset("gramajo/nouns-proposals")Known Limitations
- Proposer addresses: The subgraph does not consistently return proposer addresses (null for most proposals). This field is excluded from the dataset.
- LLM category labels: Categories are machine-assigned, not human-verified. While ~95% accurate based on spot-checking, some edge cases may be misclassified.
- Temporal drift: The DAO's voting patterns, participant composition, and proposal norms have evolved significantly over 5 years. Models trained on this data may not generalize to future proposals.
- Quorum changes: Nouns DAO has changed its quorum requirements multiple times. Proposals are evaluated against the quorum in effect at the time of voting.
Labeling Prompt
The following system prompt was used with gpt-4o-mini (via OpenRouter) on 2026-07-09:
You are a classifier for Nouns DAO governance proposals. Assign each proposal to exactly ONE category.
Categories:
- "Art & Traits": Adding/modifying Nouns artwork, accessories, heads, bodies, traits, Noundry proposals, art contributions, fonts, CC0 assets, plush/merch design
- "Governance & Protocol": DAO-level decisions — parameter changes, DUNA/legal, delegate appointments, client ID approvals, treasury strategy, DAO structure, elections
- "Grants & Funding": Requests for ETH/USDC from treasury, donations, retroactive funding, project grants, school/community funding
- "Events & Marketing": Conferences, festivals, parties, IRL activations, esports sponsorships, swag production, content creation, social media, PR
- "Infrastructure & Dev": Technical tooling, smart contracts, websites, integrations, developer tools, protocol development
Key: Client ID approvals = GOVERNANCE. Esports = EVENTS. Plush DESIGN = Art, plush PRODUCTION = Events.
IMPORTANT: Use the EXACT proposal ID number from each proposal's "ID:" line.Intended Use
This dataset is intended for:
- Training text classification models to predict DAO proposal outcomes
- Research on decentralized governance patterns
- Educational purposes in NLP and Web3 governance
Not intended for: Financial advice, automated voting decisions, or production governance systems without human oversight.
Citation
@misc{nouns-proposals-2026,
author = {Juan Gramajo},
title = {Nouns DAO Proposals Dataset},
year = {2026},
publisher = {Hugging Face},
howpublished = {\\url{https://huggingface.co/datasets/gramajo/nouns-proposals}}
}License
This dataset is released under the MIT License. Nouns DAO proposal content is CC0 (public domain) per the Nouns DAO governance framework.
