barissozudogru/belnap-debate-corpus

Belnap Real-Debate Corpus (110 Propositions)

What this is

110 contested propositions across 31 domains, used as a real-debate evaluation corpus for paraconsistent (Belnap-Dunn) debate aggregation. Each row is a single proposition sourced from established controversy databases.

Distribution

Source	Count
Kialo	52
ProCon.org	40
Classical philosophy	18

Controversy level	Count
high	79
medium	25
low	6

31 domains including: economics, bioethics, ethics (9 each); labor, education, technology_policy, animal_ethics, environment, political_philosophy, free_speech (5 each).

Schema

Column	Type	Description
id	string	Corpus row identifier (rc_001 ... rc_110)
proposition	string	Single-sentence claim under debate
domain	string	Topic domain (one of 31 categories)
controversy	string	high / medium / low — degree to which informed debaters disagree
source	string	ProCon.org / Kialo / philosophy

Source attribution

Propositions are short factual statements about real debate topics, curated from publicly accessible controversy databases:

• ProCon.org (now part of Encyclopedia Britannica): topic statements about U.S. political and ethical debates
• Kialo: public debate-map root claims on contemporary issues
• Classical philosophy: long-standing propositions from the philosophical canon (e.g., free will, the problem of evil)

Only the proposition statements (not full pro/con argument trees) are included. The curation, schema, and domain/controversy labeling is original work licensed CC-BY-4.0.

Use cases

• Evaluating paraconsistent (Belnap-Dunn) debate aggregators against classical majority-vote baselines
• Benchmarking LLM-as-judge systems on genuinely contested claims rather than toy propositions
• Measuring agreement rates between LLM debaters as a function of controversy level
• Training/evaluating disagreement-detection systems

Load it

from datasets import load_dataset
ds = load_dataset("barissozudogru/belnap-debate-corpus")
print(ds["train"][0])
# {'id': 'rc_001', 'proposition': 'The death penalty should be abolished in all circumstances.', 'domain': 'criminal_justice', 'controversy': 'high', 'source': 'ProCon.org'}

Or directly with pandas:

import pandas as pd
df = pd.read_csv("hf://datasets/barissozudogru/belnap-debate-corpus/propositions.csv")
print(df[df["controversy"] == "high"].head())

Caveats

• Single-sentence propositions. Full pro/con argument trees from the source sites are NOT included; this corpus is only the root claims.
• "Controversy" labels are author judgments, not measured agreement rates. Treat as ordinal categories, not calibrated scores.
• ProCon.org content underwent an editorial transition during Encyclopedia Britannica acquisition (2022). Propositions reflect the topics covered, not Britannica's current editorial framing.

Related

• ADK Agent Playground — the Belnap debate adjudication system this corpus evaluates
• Belnap Paraconsistent Logic Visualizer — interactive four-valued logic demo
• Belnap, N. D. (1977). A Useful Four-Valued Logic. In Modern Uses of Multiple-Valued Logic.
• SEP: Paraconsistent Logic — Priest, Tanaka & Weber (2025)