← DataDownloads
Prediction market data,
in bulk.
Five datasets covering live world state, 1.7M+ settled outcomes, monthly Brier calibration scorecards, the four-value SimpleFunctions Index, and the World Awareness benchmark — all CC-BY-4.0, mirrored on HuggingFace, refreshed on a known cadence.
from datasets import load_dataset
ds = load_dataset(
"SimpleFunctions/settled-markets"
)
# 1.7M outcomes · Kalshi + Polymarket
# Brier-eligible: 326K (t-24h price)
ds["train"].to_pandas().head()5
Datasets
1.7M+
Settled outcomes
180+
Days of history
CC-BY
License
Datasets
The five datasets.
World State Daily
Compressed snapshot of the prediction-market world model — top movers, divergences, consensus breaks, SimpleFunctions Index components — in roughly 800 tokens. The same payload that powers /api/agent/world. 180+ days of history.
- Format
- JSON
- Cadence
- Daily
- Rows
- 180+ days
- Size
- ~150 KB / day
API twin
/api/agent/worldLive world state
Best for
Agent context, daily archive, WAB-style eval traces
date · regime · sfIndex{disagree,geoRisk,breadth,activity} · movers[] · divergences[] · consensusBreaks[]huggingface.co/datasets/SimpleFunctions/world-state-dailySettled Markets
Monthly JSONL partitions of every prediction-market contract that resolved on Kalshi or Polymarket with meaningful volume (>$10K). The current and previous month refresh daily; older months are immutable.
- Format
- JSONL
- Cadence
- Daily
- Rows
- 1.7M+ outcomes (326K Brier-eligible)
- Size
- ~15-20 MB / active month
API twin
/calibrationLive calibration view
Best for
Brier scoring, event studies, forecasting research
venue · ticker · title · category · predicted_price · predicted_price_t24h · resolved_outcome · resolved_at · volumehuggingface.co/datasets/SimpleFunctions/settled-marketsCalibration Scorecards
Monthly Brier-score and log-loss calibration breakdowns for Kalshi + Polymarket, published with a 14-day delay after month end to capture late resolutions.
- Format
- JSON
- Cadence
- Monthly
- Rows
- Monthly scorecards
- Size
- ~2.5 KB / month
API twin
/api/calibrationLatest scorecard API
Best for
Model benchmarking, venue/category calibration checks
period · venue · category · brier · log_loss · hit_rate · price_buckethuggingface.co/datasets/SimpleFunctions/calibration-scorecardsSimpleFunctions Index History
Flat JSONL time series of the SimpleFunctions Index: disagreement, geo-risk, breadth, and activity. Computed every 15 minutes from the live market universe and exported daily.
- Format
- JSONL
- Cadence
- Daily
- Rows
- 15-min rows
- Size
- ~1.2 MB
API twin
/api/public/indexLive index API
Best for
Regime detection, macro overlay, feature engineering
timestamp · disagreement · geo_risk · breadth · activity · market_counthuggingface.co/datasets/SimpleFunctions/sf-index-historyWorld Awareness Bench
Monthly 100-question benchmark designed to measure how well an LLM reflects current-event probabilities. Each question pairs a held-out ground truth with a market-derived reference probability.
- Format
- JSON
- Cadence
- Monthly
- Rows
- 100 questions / month
- Size
- ~15-30 KB / version
API twin
/papers/world-modelWAB methodology
Best for
LLM eval, world-model probes, reasoning benchmark
question · ground_truth · market_probability · category · resolved_athuggingface.co/datasets/SimpleFunctions/world-awareness-benchQuick start
Three ways to load a dataset.
HuggingFace · datasets
Streaming and partition selection without an API key.
pip install datasets
python -c "from datasets import load_dataset; print(load_dataset('SimpleFunctions/settled-markets', data_files='2026-05.jsonl', split='train'))"Direct JSONL
Pull the monthly partition directly from HuggingFace.
curl -L https://huggingface.co/datasets/SimpleFunctions/settled-markets/resolve/main/2026-05.jsonl | head
Live API twin
Use the API when you want the current value, not a file snapshot.
curl https://simplefunctions.dev/api/agent/world curl https://simplefunctions.dev/api/public/index
Daily archive
Daily archive for research access.
Beyond the public HuggingFace datasets, a daily archive ships the internal compute tables — orderbook snapshots, indicator history, regime classifications, agent traces. Available via research collaboration.
Archive themes
- Markets & orderbook
- Indicators & regime
- Thesis & agent
- Macro & SimpleFunctions Index
FAQ
Frequently asked.
How often are the prediction-market datasets updated?
World State and SimpleFunctions Index export daily. Settled Markets uses monthly JSONL partitions, with the current and previous month refreshed daily. Calibration Scorecards publish monthly with a 14-day delay. World Awareness Bench publishes monthly on the first day of the month.
What license covers the SimpleFunctions datasets?
All public datasets ship under CC-BY-4.0 — free to use commercially or in research with attribution. See /data-license for the full text and citation guidance.
Which file formats are available?
JSONL for Settled Markets and SimpleFunctions Index History. JSON for World State Daily, Calibration Scorecards, and World Awareness Bench. All datasets are mirrored on HuggingFace for direct streaming via datasets, pandas, or polars.
How do I download a single day instead of the whole archive?
For dated JSON snapshots, fetch the file directly from HuggingFace, for example world-state-daily/resolve/main/2026-05-03.json. For live state, hit the matching public API endpoint: /api/agent/world for world state, /api/calibration for the latest scorecard, and /api/public/index for the live index.
Is the schema stable enough to build on?
Schemas are versioned. Additive changes (new optional fields) ship without notice. Breaking changes ship via a new dataset slug — the older slug stays readable for ≥6 months. Subscribe to /changelog for the announcement queue.
Adjacent surfaces
Historical data →
Per-market price history and Brier scorecards.
Real-time feeds →
World state, RSS, MCP — live, no signup.
Indicators →
Twelve quantitative metrics on every market.
Live API →
REST endpoints for the same content.
Calibration →
Brier scores rendered live across categories.
SimpleFunctions Index →
Daily disagreement / geo-risk / breadth time series.