Polyflux closed on 2 November 2026

Four years deep in Polymarket data.

We built and ran Polyflux — a real-time, wallet-level data layer for Polymarket — and traded the markets with it. The paid service has shut down. The experience, the research and the full on-chain history are still here.

Open the dataset Talk to us

What we did

Hands-on, end to end: from the node to the order book to the P&L.

Ran the infrastructure

Our own Polygon full node (Bor + Heimdall) and a mempool pipeline that streamed every Polymarket trade, deposit, withdrawal and oracle event to customers roughly 1.5 seconds before it confirmed on-chain.

Built the tools

A real-time WebSocket API and Python SDK, a Trades Inspector that reveals every wallet behind a market, wallet tracking with Telegram and Discord alerts, a disputed-markets monitor, and a wallet profiler.

Spotted insiders

Built tooling to flag wallets that look like they trade on information before it is public: timing against news and resolution, position size relative to history, wallet age and funding patterns. Wallet-level data makes this possible; aggregate feeds do not.

Modelled 5-minute crypto markets

Designed and ran a trading model for Polymarket's 5-minute crypto up/down markets: fast-resolving, high-frequency contracts where order flow, pricing lag and execution speed matter more than opinion.

Traded it ourselves

We used the same wallet-level order flow to trade Polymarket — profitably, across a range of markets. We know the microstructure, the resolution edge cases and the behaviour of the big wallets from the inside.

Published the research

Wallet-level studies, strategy write-ups and market analyses on the blog — wallet types, whale behaviour, copy-trading, forgotten wallets, panic-seller dynamics, UMA disputes. All still online.

Open dataset · CC BY 4.0

Every Polymarket trade, on-chain, since day one.

We published the complete on-chain history of Polymarket as a frozen archive on Hugging Face: every fill, split, merge and redemption from the first CLOB trade on 21 November 2022 to 26 October 2026, with no gaps. Decoded from our own Polygon node and cross-checked day by day against independent indexers.

4.9B

rows

1.32B

fills

$46.7B

traded volume

3.27M

wallets

10.1M

outcome tokens

4 yrs

Nov 2022 → Oct 2026

What's inside

trades — daily Parquet files, ~143 GB compressed. One row per wallet per on-chain event: wallet, asset, size, price, fee, counterparty, maker/taker role, tx hash, block.
assets — one row per outcome token with its market, outcome label and resolved payout.
markets, events, markets_full — market and event catalogues with questions, tags and full resolution rules.

Built for research: time-series, wallet behaviour, market microstructure, resolution analysis. Use it, cite it, build on it.

View on Hugging Face

quickstart.py

# pip install polars huggingface_hub
import polars as pl

# One day of fills — every row is one wallet's side of an on-chain event
df = pl.read_parquet(
    "hf://datasets/ieshuaganocry/polymarket-onchain-trades/data/year=2025/day_2025-11-04.parquet"
)
# Taker rows only, so volume isn't double-counted
takers = df.filter(pl.col("wallet_role") == "taker")
print(takers.select((pl.col("size") * pl.col("price")).sum()))

Happy to talk.

If you're building on Polymarket, researching prediction markets, or want a second pair of eyes from people who ran the data and traded on it, get in touch. No pitch, just a conversation.

hello@polyflux.io

Things we can help with

Wallet-level order flow and market microstructure
Insider and informed-trading detection
Trading models for 5-minute crypto up/down markets
Whale, smart-money and copy-trading analysis
Running Polygon nodes and real-time event pipelines
Resolution, disputes and UMA oracle behaviour
Working with the on-chain dataset