Coyevans/mlb-polymarket-kalshi-matched-book-sample

MLB Cross-Venue Matched Book — Free Sample

One full MLB game (Arizona Diamondbacks @ Minnesota Twins, 2026-06-21), with Polymarket and Kalshi prices aligned tick-for-tick and the settled outcome labeled on every row.

This is a single-game sample of a larger archive. The point it proves: across the whole game, both venues priced the Twins' win probability within ~1¢ of each other on average — climbing together from ~0.10 to ~0.99 as Minnesota (the eventual winner) pulled away.

Game	ARI/MIN, 2026-06-21
Rows	1,204 aligned snapshots
Outcome	MIN won (won=1.0)
Mean |cross-venue spread|	0.009 (under one cent)

Why this is hard to get

Neither Polymarket nor Kalshi sells its historical order book, and it can't be reconstructed after the fact. Matching the same game across both venues on one clock — and labeling each tick with the eventual winner — is the value. The raw books underneath are irreplaceable.

Columns

ts, date, game, team, poly_bid/ask/mid, kalshi_yes_bid/ask/mid, xvenue_spread, winner, won, settled — see SCHEMA.txt for the full reference and methodology.

Honesty notes (read before modeling)

• Polymarket is one-sided per game: it runs an active book on one token of each binary pair, so the other side is the exact complement (1 − poly_mid). Kalshi carries both sides.
• poly_mid is a book midpoint, not a trade price.
• Effective granularity is ~30–60s (both venues top-of-book, as-of joined at 30s).
• Settlement is Kalshi's finalized resolution (ESPN doesn't cover MLB).

Quickstart

import pandas as pd
df = pd.read_parquet("mlb_matched_sample_ARI-MIN_2026-06-21.parquet")
print(df[["ts","team","poly_mid","kalshi_mid","xvenue_spread","won"]].head())

Citation

Mirrored on Zenodo with a DOI:

ZenHodl (2026). MLB Cross-Venue Matched Book (Polymarket × Kalshi) — Sample. Zenodo.
https://doi.org/10.5281/zenodo.20816908

What this data can answer (worked example, July 2026)

Using 16 days of the full matched-book archive (this sample is one game of it), we measured cross-venue lead-lag on MLB team-win markets:

• Polymarket leads Kalshi: lead-lag asymmetry +0.394 (cluster-bootstrap 95% CI [0.32, 0.47]), positive on 16/16 days, n=309,693 aligned pairs.
• Divergences half-close in ~34 seconds (98-99% fully converge within 10 minutes at 2-5c thresholds).
• The naive cross-venue trade loses: entering Polymarket toward Kalshi is significantly negative net of spread; the Kalshi-side paper edge vanishes with one snapshot of latency. Full write-up with method and limits: https://zenhodl.net/blog/polymarket-leads-kalshi-30-seconds-lead-lag-study

We publish negative results. This is research data, not a trading signal.

Next step (if this sample was useful)

• $9 Tryout Tape — two complete settled MLB game-days (26 games), Polymarket × Kalshi matched tick-for-tick with the winner labeled on every row. Same schema as the full archive, smallest possible commitment: https://zenhodl.net/billing/start/mlb_matched_tryout
• Free CLV graders (no signup): grade your sportsbook bets at https://zenhodl.net/grader or your Kalshi fills at https://zenhodl.net/grader/kalshi — we grade against closing prices we captured live (Kalshi doesn't retain them).
• Which dataset do I need? Comparison table + free samples for every archive: https://zenhodl.net/products

Full archive & more datasets

The complete multi-day archive (every matched MLB game, growing daily) — alongside Polymarket & Kalshi order-book archives, settled-outcome layers, and microstructure packs — is at 👉 zenhodl.net/products. This single game is released free to show the structure and quality.

Captured live by ZenHodl — historical prediction-market order books the venues don't sell. More free tools + datasets at zenhodl.net/products.