ironflow's picture
Update the coverage line
58de237 verified
|
Raw History Blame Contribute Delete
15.3 kB
metadata
license: cc-by-4.0
pretty_name: Hyperliquid HIP-4 Outcome Markets
task_categories:
  - time-series-forecasting
tags:
  - hyperliquid
  - hip-4
  - prediction-markets
  - outcome-markets
  - crypto
  - defi
  - sports
  - tick-data
  - market-microstructure
  - finance
  - trading
  - time-series
size_categories:
  - 10M<n<100M
configs:
  - config_name: fills
    data_files:
      - split: train
        path: data/fills/*.parquet
    default: true
  - config_name: markets
    data_files:
      - split: train
        path: data/markets.parquet

Hyperliquid HIP-4 Outcome Markets

Every fill on Hyperliquid's HIP-4 outcome markets (its prediction markets) from 1 August 2026, one row per fill as Hyperliquid recorded it, with a new file every day, and a list of every contract with its event, outcome and settlement. Through 8 November 2026: 100 days, 15,171,948 fills, 41,741 contracts.

From 1 August to 5 November 2026: 13,893,235 fills on 39,260 contracts between 10,509 wallets, $160.9 million in trading volume and $486.9 million in notional over all fills. Volume grew from $12.0 million in August to $76.6 million in October; the busiest day is 1 November 2026, with 444,565 fills and $7.8 million.

Sports took 65% of the volume, $104.1 million over 1,004 events, from single games to season winners. The largest events are the 2026/2027 winners of the UEFA Champions League ($6.3 million), the NFL ($4.6 million) and the English Premier League ($4.4 million). Crypto and stock price contracts took 27% over 12,206 events, most of them BTC, HYPE and ETH price thresholds that settle within minutes to a day. Federal Reserve rate decisions (2.7%), Anthropic IPO market-cap contracts (2.4%) and the 2026 Ballon d'Or (0.9%) make up most of the rest.

The fills are the ones Ironflow records from its own Hyperliquid node, the same data its GET /v1/fills API serves. Names, events and Hyperliquid's own description of each outcome come from Ironflow's market registry, which reads them from Hyperliquid while a contract is listed. Hyperliquid stops listing a contract once it settles, so most of this metadata is no longer available from Hyperliquid.

What HIP-4 outcome markets are

HIP-4 is Hyperliquid's standard for outcome contracts. A contract pays 1 USDC if its outcome happens and 0 if it does not, so its price, between 0 and 1, reads as the market's probability. Every outcome has two sides, each a coin of its own (#N): Yes and No, Over and Under, or the two teams of a game. One of each side is always worth 1 USDC together, so the two prices add up to 1. A question with more answers, such as a football match (home win, draw, away win, other) or a season winner, bundles one two-sided outcome per answer.

Contracts are fully collateralized: a buyer pays the price up front, and that is the most the position can lose. There is no leverage and no margin, so there are no liquidations: none of the 21.3 million outcome fills Ironflow recorded from HIP-4's launch on 2 May to 5 November 2026 carries Hyperliquid's liquidation field, and none of the 10.2 million in the 30 days to 6 November.

Most outcomes are listed by deployers from Hyperliquid's templates: crypto price thresholds, sports games and seasons, IPOs, central bank decisions and awards. Hyperliquid's own recurring daily price contracts run alongside.

Get live data

Live Hyperliquid data is on Ironflow. The same fills, up to the latest block, are on Ironflow's API: GET /v1/fills for one contract, and GET /v1/markets?market_class=prediction for the contracts listed now. No key is needed for the last 24 hours; a free key gives 30 days.

Source and updates

  • Source: fills recorded by Ironflow's own Hyperliquid node. Block gaps, such as node outages, are refilled automatically from Hyperliquid's node archive on S3. Names and outcome descriptions from Ironflow's market registry.
  • Checks: each day's row count matches Ironflow's store, and in every hour the day's BTC and ETH trade counts are within 25 trades or 0.5% of Hyperliquid's own candles (a node gap would show there). For the 5 busiest outcomes Hyperliquid still lists each day, hourly trading volume matches Hyperliquid's own outcome candles: 7,492 outcome-hours on 63 days since 4 September 2026, all exactly equal (see Notes).
  • Updates: one fill file per UTC day, added daily for the day before; markets.parquet and the coverage line at the top are rebuilt with it.
  • Coverage: from 1 August 2026. Ironflow also holds HIP-4 fills from 2 May to 31 July 2026, but without the fill direction (dir) and the transaction hash, so settlements and conversions cannot be told apart from trades there; they are left out.

Files

Two tables:

Config Path One row per
fills (default) data/fills/YYYY-MM-DD.parquet fill, as Hyperliquid recorded it, in exchange order (time, block, position in the block)
markets data/markets.parquet contract (one side of one outcome) with at least one fill here

A day holds 14,000 to 460,000 fills, 0.5 to 17 MB. Each trade between two wallets appears as two fills, one per wallet.

fills

Column Type Meaning
timestamp timestamp (ms, UTC) fill time
block int Hyperliquid block
market_id int Ironflow market id, 200000 + N for coin #N
coin string Hyperliquid's name for the contract, e.g. #14840
market string Ironflow market name, e.g. ATL vs BKN-20261108-ATL; #N-OUTCOME when unnamed (see Notes)
event string the event or question, shared by all its contracts, e.g. ATL vs BKN-20261108; null when unnamed
outcome string what this contract pays on, e.g. ATL, Yes, Draw-Yes, Manchester City-No; null when unnamed
address string the wallet of this fill
side string buy or sell, for this wallet
dir string Hyperliquid's fill direction: Buy, Sell, Settlement, Split Outcome, Merge Outcome, Negate Outcome, Merge Question (see Notes)
price float contract price in USDC, 0 to 1
size float contracts; each pays 1 USDC if its outcome happens
notional_usd float price * size, in USDC
fee float fee paid, in fee_token
fee_token string USDC, or the contract's own coin as +N (some August 2026 fills)
crossed bool true when this fill took liquidity (the taker)
tid int Hyperliquid trade id
oid int Hyperliquid order id
hash string transaction hash

market is event and outcome joined with -.

markets

Column Type Meaning
market_id, coin, market, event, outcome as in fills
siblings list of string the other contracts of the same event; for an unnamed contract, the other side of its outcome
kind string Hyperliquid's template, e.g. binaryPrice, sportsContestWinner, sportsContestResult, sportsTournamentWinner; priceBinary and priceBucket are Hyperliquid's own daily contracts
venue string the HIP-4 deployer, e.g. out, fox, txyz, skew
outcome_id int Hyperliquid outcome id; its two sides are coins outcome_id * 10 and outcome_id * 10 + 1
side_index int 0 for the first side (Yes, Over, team A), 1 for the second
question_id int Hyperliquid question id, for an outcome that is one answer of a question
description, question_description string Hyperliquid's own description of the outcome and of its question, verbatim
expiry timestamp (UTC) the time a price or IPO contract is decided on
scheduled_start timestamp (UTC) the scheduled start of a sports event
resolution_deadline timestamp (UTC) the latest time it resolves, as Hyperliquid lists it
listed bool true when Hyperliquid still listed the contract at export time
settle_price float 1 if the contract paid, 0 if it did not, 0.5 for a few (see Notes); null if not settled here
settle_source string fills (its own settlement fills) or other side (1 minus the other side's price)
settled_at timestamp (UTC) time of the first settlement fill
first_fill, last_fill timestamp (UTC) first and last fill in this dataset
fills int fills in this dataset
notional_usd float sum of notional_usd over those fills
taker_notional_usd float notional of the taker Buy and Sell fills: the contract's trading volume
name_source string registry, fills (the name the fills were stored under; see Notes) or null when unnamed

Use

import pandas as pd

f = pd.read_parquet("hf://datasets/ironflow/hyperliquid-outcomes/data/fills/2026-11-05.parquet")
game = f[(f.event == "AC Milan vs Ferencvarosi TC-20261105") & f.dir.isin(["Buy", "Sell"])
         & f.outcome.str.endswith("-Yes")]
odds = game.set_index("timestamp").groupby("outcome").price.resample("1h").last().unstack(0).ffill()
print(odds.tail(4))  # implied probability of a Milan win, a draw and a Ferencvaros win, by hour
-- DuckDB: the 10 events with the most trading volume, and what paid
SELECT event, round(sum(taker_notional_usd) / 1e6, 2) AS usd_m,
       string_agg(outcome, ', ') FILTER (WHERE settle_price = 1) AS paid
FROM 'data/markets.parquet'
WHERE event IS NOT NULL
GROUP BY event
ORDER BY usd_m DESC
LIMIT 10;
-- DuckDB: are the prices calibrated? The last trade price an hour before
-- settlement, against how often the contract paid
WITH last AS (
  SELECT f.market_id, arg_max(f.price, f.timestamp) AS price, any_value(m.settle_price) AS paid
  FROM 'data/fills/*.parquet' f JOIN 'data/markets.parquet' m USING (market_id)
  WHERE f.dir IN ('Buy', 'Sell') AND m.settle_price IN (0, 1)
    AND f.timestamp < m.settled_at - INTERVAL 1 HOUR
  GROUP BY f.market_id)
SELECT round(price, 1) AS price, count(*) AS contracts, round(avg(paid), 3) AS paid_share
FROM last
GROUP BY 1
ORDER BY 1;

Notes

  • One row per fill, not per trade. The two sides of an outcome share one order book, so a wallet buying Yes is often matched with a wallet buying No: each gets a buy fill on its own coin, with its own tid, at prices that add up to 1, and new contracts are created. Two sellers, one of Yes and one of No, likewise redeem contracts. In 51% of Buy and Sell fills the other party is on the other side's coin in this way, so fills are not paired into trades here (and ironflow/hyperliquid-trades leaves outcome markets out). On 20 October 2026, every one of the 72,801 taker fills without a partner in its own coin has a maker fill on the other side's coin with the same hash and side and a price that adds up to 1.

  • dir. Besides trades, Hyperliquid records conversions and settlements as fills, at fixed prices:

    dir fills what happened
    Buy, Sell 11,758,722 a trade, at prices from 0.00001 to 0.99999
    Settlement 724,858 the outcome resolved: each open position is closed at 1 or 0
    Merge Outcome 671,254 one Yes and one No turned back into 1 USDC (two sell fills at 0.5)
    Split Outcome 351,053 1 USDC turned into one Yes and one No (two buy fills at 0.5)
    Negate Outcome 235,789 the No of one answer exchanged for a Yes on each other answer of its question (a sell at 1 and buy fills at 1 / (answers - 1))
    Merge Question 151,559 a Yes on every answer of a question turned into 1 USDC (sell fills at 1 / answers)
  • Volume. To count each trade once, sum the taker fills of Buy and Sell (crossed and dir in Buy, Sell): this is taker_notional_usd in markets. Per outcome and hour, the size of those fills over both sides equals the volume of Hyperliquid's own outcome candles, which is the check above (Hyperliquid serves candles only for outcomes it still lists). notional_usd over all fills counts both wallets of a trade and includes conversions and settlements.

  • Settlement. When an outcome resolves, every open position is closed against a Hyperliquid system address (733 of them, 0x3200000000000000000000000000000000000001e3 and similar, in 362,429 fills): the holder's fill is a taker sell, the system address's a buy. 38,456 of the 39,260 contracts have a settlement price: 26 contracts (13 outcomes) settled at 0.5 on both sides, among them WAS vs NYG on 11 October 2026 and a few BTC price contracts, some settled days before their expiry; the data does not say why. 10 contracts nobody held at settlement take 1 minus the other side's price (settle_source = other side). 405 contracts Hyperliquid no longer lists have no settlement here: nobody held them when they settled, or they settled after 5 November 2026.

  • Names. 10,781 contracts (386,351 fills, 2.5% of the volume) are unnamed: market is #N-OUTCOME and event, outcome and the registry fields are null. They are short-lived, with a median of 14 minutes between first and last fill, almost all from October 2026: they opened and settled before Ironflow's registry stored their description, and Hyperliquid no longer lists them. Their siblings still give the other side. For 1,400 more contracts (4.1 million fills), the registry lost the name but the fills kept it: name_source is fills, event and outcome are split from the name by the same rules, and kind, venue, description and the times are null.

  • Registry fields. expiry is set for 24,226 contracts, scheduled_start for 2,703 and resolution_deadline for 2,853, where Hyperliquid's description has them. venue is known for 14,064 contracts, those Hyperliquid still listed after it started publishing deployers in October 2026; it is null for Hyperliquid's own daily contracts and older ones.

  • event and outcome are the market name split at the side label, which Ironflow's names always put last; for an answer of a question the answer stays with the side (Draw-Yes). The fill files keep the name a contract had on the day it was exported; markets.parquet is rebuilt daily with the latest one.

  • Fees. fee is in fee_token: USDC, except 461,833 fills in August 2026 that paid the fee in the contract they received (+N, the contract's coin #N).

  • No duplicate (market_id, tid, address, side, dir) rows; every fill has a block and a hash.

  • Wallet addresses are public on-chain data. To object to, or ask us to remove, a wallet address, write to hello@ironflow.sh. Files already published cannot be recalled; later files leave the address out.

Changes

  • v1, 6 November 2026. First release: 1 August to 5 November 2026.

License and citation

CC BY 4.0. Credit "Ironflow (ironflow.sh)" when you use or share the data.