15btc_eth / README.md
kaboomfox's picture
Upload folder using huggingface_hub
a702ce9 verified
metadata
license: mit
task_categories:
  - time-series-forecasting
tags:
  - polymarket
  - prediction-markets
  - orderbook
  - btc
  - eth
  - trading
  - chainlink
  - binance
size_categories:
  - 1M<n<10M
pretty_name: Polymarket BTC/ETH 15-Minute Market Orderbook Data

Polymarket BTC/ETH 15-Minute Market Orderbook Data

Real-time orderbook snapshots from Polymarket 15-minute BTC and ETH prediction markets, collected via WebSocket with Chainlink oracle and Binance spot prices.

Dataset Description

Each row is a point-in-time snapshot of one side (YES or NO) of a 15-minute binary option market on Polymarket. Markets resolve based on whether BTC/ETH price goes up or down over 15 minutes, as determined by Chainlink oracle.

Collection method: WebSocket connection to Polymarket CLOB via shard manager, sampled at ~1 second intervals per market.

Data Format

Parquet shards (shard_XXXX.parquet) with the following schema:

Column Type Description
ts int64 Timestamp in milliseconds
progress float Market progress (0.0 = start, 1.0 = settlement)
outcome_up float 1.0 if this is the UP/YES token
outcome_down float 1.0 if this is the DOWN/NO token
best_bid float Best bid price (0.0-1.0)
best_ask float Best ask price (0.0-1.0)
best_bid_size float Liquidity at best bid (in dollars)
best_ask_size float Liquidity at best ask (in dollars)
oracle_price int64 Chainlink oracle price in cents
binance_price int64 Binance BTC/ETH price in cents
target_price int64 Market target/strike price in cents
imbalance float (bid_size - ask_size) / (bid_size + ask_size)

Settlement

Markets settle at the 15-minute mark based on Chainlink oracle:

  • YES wins if oracle_price > target_price (price went up)
  • NO wins if oracle_price <= target_price (price went down or flat)

The target_price is the oracle price at market open.

Usage

import pandas as pd

# Read a single shard
df = pd.read_parquet("shard_0001.parquet")

# Read all shards
import glob
dfs = [pd.read_parquet(f) for f in sorted(glob.glob("shard_*.parquet"))]
df = pd.concat(dfs, ignore_index=True)

# Filter BTC only (oracle_price > $50,000)
btc = df[df["oracle_price"] > 5_000_000]

# Get YES side only
yes_side = df[df["outcome_up"] == 1.0]

Use Cases

  • Backtesting prediction market trading strategies
  • Orderbook microstructure analysis
  • Price discovery dynamics in binary options
  • Oracle vs spot price divergence studies

Collection Details

  • Source: Polymarket CLOB (Central Limit Order Book)
  • Oracle: Chainlink price feeds on Arbitrum
  • Spot: Binance BTC/USDT and ETH/USDT
  • Shard rotation: Every 30 minutes
  • Update frequency: ~1 second per market side
  • Collection start: February 2026