Data downloads

Polymarket calibration benchmark, 2026 Q4

Market prices against outcomes for every Polymarket market that resolved between 2025-10-01 and 2026-09-30, excluding voided and disputed markets, scored at four fixed distances before resolution. Method notes and the full discussion are on the benchmark page.

• Horizon decay, constant universe (CSV, 4 rows): The 30,973 markets that had a traded price at all four horizons, scored at each. The only table in the set that reads as a decay curve, because the market population is held fixed across rows.
• Horizon table, full per-horizon universes (CSV, 4 rows): Each horizon over every market that had a price at that distance. Comparable to how public accuracy dashboards count. Rows are different market sets, so they do not form a decay curve.
• Calibration curves, 24 hours and 30 days (CSV, 20 rows): Stated price against resolved frequency by price decile, both horizons in one file with a horizon column.
• Accuracy by category, 24 hours (CSV, 12 rows): Twelve categories, every one with at least 2,000 markets at 24 hours, with market counts, stated price, resolved frequency, Brier score, and accuracy.

Polymarket calibration benchmark, 2026 Q3

Market prices against outcomes for every Polymarket market that resolved between 2025-07-01 and 2026-06-30, excluding voided and disputed markets, scored at four fixed distances before resolution. Method notes and the full discussion are on the benchmark page.

• Horizon decay, constant universe (CSV, 4 rows): The 21,992 markets that had a traded price at all four horizons, scored at each. The only table in the set that reads as a decay curve, because the market population is held fixed across rows.
• Horizon table, full per-horizon universes (CSV, 4 rows): Each horizon over every market that had a price at that distance. Comparable to how public accuracy dashboards count. Rows are different market sets, so they do not form a decay curve.
• Calibration curves, 24 hours and 30 days (CSV, 20 rows): Stated price against resolved frequency by price decile, both horizons in one file with a horizon column.
• Accuracy by category, 24 hours (CSV, 8 rows): Eight categories with market counts, stated price, resolved frequency, Brier score, and accuracy.

Graded alert ledger

Every position OVERROUND has alerted, with its grade after resolution, losses included. The human-readable ledger is on the receipts page; the complete ledger, every graded row in one response, is available as JSON at api.overround.pro/receipts/export. Unresolved alerts are sealed and publish in full at resolution, and voided markets are excluded from the record.

How to cite

• One line, for a chart caption or footnote: Data: OVERROUND, Polymarket Calibration Benchmark 2026 Q4 (markets resolved October 2025 to September 2026), overround.pro/benchmarks/polymarket-calibration-2026-q4 (CC BY 4.0). For the 2026 Q3 edition (markets resolved July 2025 to June 2026), name it and its page the same way.
• License: CC BY 4.0. Republish, chart, and build on any of it, including commercially, with attribution to OVERROUND.
• File URLs carry their edition and stay put, so a link in a published article keeps resolving after the next edition ships.
• For academic citation, each benchmark edition has its own DOI, which resolves to a Zenodo archive of that edition’s CSVs and charts. 2026 Q4: 10.5281/zenodo.23263367. 2026 Q3: 10.5281/zenodo.21790936. Cite the DOI of the edition you used.
• Method behind every number: the methodology page. Want a cut that is not here, by category, horizon, or period? Email hello@overround.pro and we will run it.