wzsg/polymarket-orderfilled-v2

Polymarket CLOB V2 OrderFilled (Polygon)

中文说明 · CLOB V1 dataset

This dataset contains OrderFilled events emitted by Polymarket CLOB V2 Exchange contracts on Polygon mainnet. It covers both standard CTF markets and Neg Risk markets, follows the V2 event ABI, and is stored as ZSTD-compressed Parquet files partitioned by UTC month.

Explore V1 and V2 activity, trends, and summary metrics in the interactive live analytics dashboard.

The collection, processing, and dashboard source code is available in the open-source GitHub repository.

Updated daily

The incremental collector runs at 00:00 UTC and atomically publishes new Parquet shards together with the latest dataset state.

Item	Value
Last updated (UTC)	2026-08-04 00:10:40 UTC
Total rows	575,630,861
Standard CTF Exchange V2 rows	478,388,740
Neg Risk CTF Exchange V2 rows	97,242,121
BUY (side = 0) rows	494,473,779
SELL (side = 1) rows	81,157,082
UTC event time range	2026-04-28 11:04:48 UTC to 2026-08-03 23:47:33 UTC
Polygon event block range	86,127,122 to 91,397,218
Latest event month	2026-08
Latest safely scanned block	91,397,218
Safe block hash	0x4def3112eb625eb1596b9ea089107d8e8361ef5b43a841f67874c21d54875745
Next block to scan	91,397,219
Confirmation depth	512 blocks
Monthly partitions	5
Parquet files	123
Parquet data size	33,617,734,086 bytes (31.31 GiB)

Polymarket moved its production trading system from CLOB V1 to CLOB V2 on April 28, 2026. V2 introduced new Exchange contracts, pUSD collateral, a new order format, and a new on-chain fee model. See the official CLOB V2 migration guide for details.

Related dataset

• Polymarket CLOB V1 OrderFilled: V1 historical events from November 21, 2022 through April 28, 2026.
• This dataset: V2 events beginning on April 28, 2026.

The V1 and V2 OrderFilled ABIs differ. Do not concatenate identically named fields across versions without first normalizing asset and side semantics.

V2 contracts and event

Contract	Polygon address	Dataset contract value
CTF Exchange V2	0xE111180000d2663C0091e4f400237545B87B996B	CTF_EXCHANGE_V2
Neg Risk CTF Exchange V2	0xe2222d279d744050d28e00520010520000310F59	NEGRISK_CTF_EXCHANGE_V2

V2 OrderFilled event:

event OrderFilled(
    bytes32 indexed orderHash,
    address indexed maker,
    address indexed taker,
    Side side,
    uint256 tokenId,
    uint256 makerAmountFilled,
    uint256 takerAmountFilled,
    uint256 fee,
    bytes32 builder,
    bytes32 metadata
);

Side is ABI-encoded as uint8. Event topic0:

0xd543adfd945773f1a62f74f0ee55a5e3b9b1a28262980ba90b1a89f2ea84d8ee

The field layout is defined in the official Polymarket CTF Exchange V2 source.

Schema

Each Parquet file contains 16 physical columns:

Column	Parquet / DuckDB type	Description
timestamp	uint64 / UBIGINT	Unix timestamp in seconds
datetime	timestamp[us, UTC]	Event time in UTC
block_number	uint64 / UBIGINT	Polygon block number
transaction_hash	string	Transaction hash
log_index	uint32 / UINTEGER	Log index
contract	string	CTF_EXCHANGE_V2 or NEGRISK_CTF_EXCHANGE_V2
order_hash	string	V2 order hash
maker	string	Maker of the filled order
taker	string	Counterparty or Exchange execution address
side	uint8 / UTINYINT	Maker order side: 0 = BUY, 1 = SELL
token_id	string	Conditional Token ERC-1155 token ID
maker_amount_filled	string	Raw integer amount paid by the maker
taker_amount_filled	string	Raw integer amount received by the maker
fee	string	Raw V2 OrderFilled.fee value
builder	string	Canonical bytes32: 0x plus 64 lowercase hex digits
metadata	string	Canonical bytes32: 0x plus 64 lowercase hex digits

event_month is a Hive partition column derived from the directory name and is not stored inside each Parquet file:

event_month=2026-07/data_0.parquet

Side, asset, and price semantics

side is the side of the maker order:

• side = 0 (BUY): the maker pays pUSD collateral and receives the outcome token identified by token_id.
• side = 1 (SELL): the maker pays the outcome token identified by token_id and receives pUSD collateral.

Amounts and fees are stored as decimal strings to preserve full uint256 precision. pUSD and outcome token amounts normally use six raw decimal places.

Execution price can be calculated as:

CASE
    WHEN side = 0 THEN
        TRY_CAST(maker_amount_filled AS DECIMAL(38, 0)) /
        NULLIF(TRY_CAST(taker_amount_filled AS DECIMAL(38, 0)), 0)
    WHEN side = 1 THEN
        TRY_CAST(taker_amount_filled AS DECIMAL(38, 0)) /
        NULLIF(TRY_CAST(maker_amount_filled AS DECIMAL(38, 0)), 0)
END AS price

Do not assume that taker is always an end user. One matching transaction can contain multiple maker fills, and mint, merge, and direct transfer paths may expose different counterparties.

Partition layout

.
├── event_month=2026-04/
│   ├── data_0.parquet
│   └── ...
├── event_month=2026-05/
├── event_month=2026-06/
├── event_month=2026-07/
├── manifest.json
├── README.md
└── README.en.md

Storage settings:

• Parquet
• ZSTD compression
• UTC monthly partitions
• 1,048,576-row row groups
• Historical export shards target about 1 GiB; block-bounded incremental shards can be smaller

DuckDB example

import duckdb

dataset = "event_month=*/*.parquet"

con = duckdb.connect()
con.execute("SET TimeZone='UTC'")

result = con.sql(f"""
    SELECT
        event_month,
        contract,
        side,
        count(*) AS fills,
        min(datetime) AS first_fill,
        max(datetime) AS last_fill
    FROM read_parquet(
        '{dataset}',
        hive_partitioning=true,
        union_by_name=false
    )
    WHERE event_month = '2026-07'
    GROUP BY event_month, contract, side
    ORDER BY contract, side
""")

print(result)

The dataset is large and continues to grow. See the live status section above for its current size. Filter by event_month, datetime, block_number, or contract as early as possible.

V2 filtering and field processing

The dataset was selected from a combined V1/V2 archive using:

WHERE contract IN ('CTF_EXCHANGE_V2', 'NEGRISK_CTF_EXCHANGE_V2')

The combined source archive retained V1-style physical column names. They were losslessly mapped to the V2 ABI during export:

Combined source column	V2 dataset column
maker_asset_id	side
taker_asset_id	token_id
maker_amount_filled	maker_amount_filled
taker_amount_filled	taker_amount_filled
maker_fee	fee
taker_fee	builder
protocol_fee	metadata

In the source archive, builder and metadata were stored as unsigned decimal representations of their corresponding 32-byte values. They were losslessly restored to canonical 66-character bytes32 hex strings during export. No other normalization was applied to hashes, address casing, amounts, or timestamps.

Validation

The initial historical export and every incremental release validate that:

• V2 ABI-mapped fields contain no NULL values
• builder and metadata match 0x[0-9a-f]{64}
• every new Parquet shard uses the same 16-column schema and ZSTD compression
• contract, side, row, block, and timestamp totals agree with cumulative shard statistics
• the stored safe block hash can be verified again through Polygon RPC

The initial export also compared BUY, SELL, non-zero builder, and non-zero metadata samples against raw Polygon RPC logs. Indexed fields and all seven non-indexed data words matched the dataset mapping.

Current cumulative statistics are in the live status section above. See manifest.json for per-month row counts, time ranges, block ranges, file counts, and byte sizes.

Incremental updates

The incremental collector scans confirmed Polygon blocks every day and adds schema-compatible immutable Parquet shards under the corresponding event_month=YYYY-MM directory. Existing shards are never overwritten. Events are deduplicated with a key containing transaction_hash, log_index, and contract.

Parquet shards, manifest.json, collector_state.json, and the live status blocks in both Data Cards are published in the same atomic Hub commit. The current progress and latest event boundary are recorded in the live status section and those two JSON files.

Known limitations

1. The current UTC month partition is still growing and does not represent a complete calendar month.
2. timestamp and datetime are preserved from the upstream archive. This dataset does not re-fetch every block timestamp.
3. The dataset contains on-chain fills only. It does not include market questions, outcome labels, categories, or Gamma API metadata. Join token_id to Polymarket market/token metadata when those fields are needed.
4. A single on-chain matching transaction can emit multiple OrderFilled events. A row should not automatically be interpreted as an independent user-to-user trade.
5. Polygon can experience short chain reorganizations. Incremental collectors should overlap the confirmed boundary by a small number of blocks and deduplicate by event key.
6. The data is derived from public blockchain events. This dataset description does not grant additional rights to underlying contract source code, platform trademarks, or third-party material. Users are responsible for determining applicable terms.

Sources

• Polymarket contract documentation
• Polymarket CLOB V2 migration guide
• Polymarket CTF Exchange V2 source
• Polygon Mainnet, chain ID 137