Pre-Built Polymarket Datasets, Ready to Backtest
1.88B on-chain execution fills, 3.06M scored wallet entities, and complete resolution history in Apache Parquet. Instant signed Cloudflare R2 downloads for DuckDB, Polars, and ClickHouse.
The only mathematically reconciled database of Polymarket participants. Rebuilt directly from raw ERC-1155 on-chain logs to purge the $344M NegRisk conversion phantom profit bug and recover $1.84B in dropped relayer redemptions. 3.06M scored wallets evaluated on true settled PnL, Brier calibration, and Benjamini-Hochberg FDR q-values.
- • Complete lifetime PnL reconciled to exact Polygon contract balances
- • Purged of the $344M NegRisk convert gross-booking accounting bug
- • Recovered $1.84B in unindexed ERC-4337 and relayer payout redemptions
- • FDR q-values and ev_ci_low separating statistical edge from lucky streaks
- • Bot vs human execution latency signatures and Sybil cluster tags
Full-depth execution tape for BTC, ETH, SOL, XRP, BNB, and DOGE up/down contracts, macro indicators, and year-end prediction markets. Includes the unbackfillable Chainlink 30s/60s TWAP settlement tape recorded directly from Polymarket's RTDS socket.
- • Over 380M raw on-chain crypto & macro fills
- • Unbackfillable Chainlink 30s & 60s settlement TWAP ticks from RTDS socket
- • Lossless E18 fixed-point integer settlement values recorded continuously
- • Maker and taker wallet identities linked to transaction hashes
- • 5m and 15m cycle contracts joined to final settlement verdicts
- • Snappy block-compressed Apache Parquet format
Continuous 1-minute historical L2 orderbook depth snapshots across major crypto, political, and macro markets. Captures top-10 bid/ask price levels, resting token quantities, bid-ask spread, and mid-market prices. Unbackfillable CLOB websocket capture essential for slippage modeling and market making.
- • Continuous 1-minute L2 order book snapshots recorded via BookRecorder #819
- • Top-10 bid and ask price levels with resting token quantities
- • Precomputed bid-ask spread and volume-weighted mid-market prices
- • Unbackfillable CLOB websocket capture: historical depth not available from Polymarket API
- • Formatted in Snappy block-compressed Parquet partitioned by condition ID and date
Every trade, fill, and quote across the 2020 & 2024 Presidential Elections, Senate and House control, Supreme Court rulings, and macro policy contracts. Zero sampled rows: complete institutional execution record.
- • Over 420M on-chain election & policy fills
- • Trump vs Harris, 2020 Trump vs Biden, debates, and swing state markets
- • Maker and taker wallet identities with transaction hashes
- • Joined to market resolution timestamps & official outcomes
- • Clean Parquet partitions by year and election cycle
Point-by-point in-play execution dynamics for resolved sports and esports markets. Tennis (ATP, WTA, Slams), Soccer (Premier League, UCL, World Cup), NBA, NFL, CS2, Dota 2, and League of Legends.
- • Over 260M resolved sports and esports execution fills
- • Full in-play momentum, trade timing, and spread execution
- • Includes market condition IDs, team/player outcomes, and official payouts
- • Complete maker/taker attribution for flow and market maker analysis
- • Snappy-compressed Parquet, partitioned by sport and date
The complete on-chain entity graph mapping sybil rings, common funders, proxy clusters, and coordinated betting cartels. Traced across 5.4M typed relationship edges, CEX deposit trees (Binance, OKX, Kraken), and co-trade timing corroboration. Delivered to academic computational finance labs and market integrity teams.
- • 5.4M typed relationship edges with calibrated confidence weights
- • Common root funder attribution tracing USDC deposit origins
- • Cluster membership classifications: leader, satellite, and sybil roles
- • Co-trade timing matrix identifying synchronized multi-wallet execution
- • Gzip/Snappy Parquet archive with complete checksum manifest
The gold standard dataset for benchmarking LLM forecasting capability against real-money crowd intelligence. 2.13M resolved Polymarket events, continuous consensus probability curves up to resolution, question embeddings, domain tags, and pre-computed Brier calibration scores. Designed for frontier AI evaluation teams.
- • 2.13M resolved markets with ground truth outcomes and resolution timestamps
- • Microsecond consensus probability curves at t-30d, t-7d, t-24h, and t-1h before resolution
- • Pre-calculated Brier scores and logarithmic loss benchmarks per market
- • Categorized across macroeconomics, geopolitical elections, tech, and science
- • Structured JSON-L and Parquet partitions ready for OpenAI Evals and LM Evaluation Harness
Direct ClickHouse replicas, custom S3 feeds, or bulk Parquet delivery?
For prop trading desks, hedge funds, and institutional research labs: we provision dedicated Read-Only ClickHouse replicas, hourly S3 tick delivery, commercial redistribution rights, and custom multi-condition filters (e.g. Benjamini-Hochberg FDR q-value filters, Sybil funding cluster graphs, or raw TWAP execution streams).