Polymarket Fills Substreams

Source: streamingfast/substreams-chain-modules · prediction-markets/polymarket-fills-substreams

Unified Polymarket order-fill history on Polygon (chain ID 137): CLOB v1 (CTF Exchange, Neg Risk CTF Exchange) merged with both CLOB v2 exchanges (CTF Exchange V2 via colindickson/polymarket-exchange, and Neg Risk CTF Exchange V2, decoded natively here) into one ordinal-sorted OrderFilled/OrdersMatched stream with an exchange_version column, so a single query spans every era.

Why this exists

Two third-party Polymarket packages were pulled from substreams.dev because they didn't work:

• polymarket-orderbook-substreams had no block-level index filter on any fill-extraction module, so it scanned and billed every block in range regardless of whether it contained a matching event.
• polymarket-pnl's map_order_fills mixed checksummed contract addresses into an eth_common:index_events filter that is lowercase-keyed, so the filter never matched anything — verified against 2,000 real fills over blocks 93,000,000–93,002,000 that map_order_fills returned zero rows for.

This package fixes both: every fill-extraction module has a lowercase-address eth_common:index_events block filter (verified with substreams info), and output was cross-checked against colindickson/polymarket-exchange's map_all_events over an identical block range (138,447 OrderFilled + 50,418 OrdersMatched, exact match).

Contracts

Contract	Address	Deploy block	Verified
CTF Exchange (v1)	0x4bfb41d5b3570defd03c39a9a4d8de6bd8b8982e	33,605,403	eth_getCode binary search against https://137.rpc.thirdweb.com, 2026-09
Neg Risk CTF Exchange (v1)	0xc5d563a36ae78145c45a50134d48a1215220f80a	50,505,492	same method
CTF Exchange (v2)	0xe111180000d2663c0091e4f400237545b87b996b	84,902,353	from colindickson/polymarket-exchange
Neg Risk CTF Exchange (v2)	0xe2222d279d744050d28e00520010520000310f59	85,058,176	eth_getCode binary search + confirmed live via raw eth_getLogs (matching OrderFilled/OrdersMatched topic0 and data width)

Neither v1 deploy block was taken from prior documentation — both were re-derived from Polygon RPC because an earlier, unsourced figure (57,000,000) had propagated through both packages this replaces. Neg Risk CTF Exchange V2 was named in the substreams-dev landing-page FAQ draft alongside the v2 cutover but wasn't read by any published package — colindickson/polymarket-exchange only covers CTF Exchange V2 above; this package covers all four.

Modules

Module	Kind	Output
map_v1_events	map	polymarket.fills.v1.V1Events — native decode of the two v1 contracts
map_v2b_fills	map	polymarket.fills.v1.UnifiedFills — native decode of Neg Risk CTF Exchange V2 (OrderFilled/OrdersMatched only)
map_fills	map	polymarket.fills.v1.UnifiedFills — OrderFilled/OrdersMatched/OrderCancelled, all four contracts
map_v2b_fee_events	map	polymarket.fills.v1.UnifiedFeeEvents — native decode of Neg Risk CTF Exchange V2 (FeeCharged only)
map_fee_events	map	polymarket.fills.v1.UnifiedFeeEvents
map_admin_events	map	polymarket.fills.v1.UnifiedAdminEvents
map_pause_events	map	polymarket.fills.v1.UnifiedPauseEvents

Neg Risk CTF Exchange V2's admin/pause events are not decoded: its OrderFilled/OrdersMatched/FeeCharged shapes were each empirically confirmed (topic0, indexed-param count, non-indexed data width) against real eth_getLogs output from that address, matching CTF Exchange V2 exactly. Its admin/pause events never appeared in any sampled block range, so there was nothing to confirm indexed-ness against — shipping them anyway would mean guessing at an ABI, which is exactly the failure mode this package exists to avoid.

V2's side + token_id encoding is normalized into maker_asset_id / taker_asset_id (side BUY (0): maker's asset is "0", i.e. collateral; side SELL (1): reversed — confirmed against Polymarket/ctf-exchange-v2 Structs.sol/CalculatorHelper.sol), so v1 and v2 rows union without a schema mismatch.

v1 has no equivalent for v2's FeeReceiverUpdated, MaxFeeRateUpdated, OrderPreapproved/OrderPreapprovalInvalidated, or per-user UserPauseBlockIntervalUpdated — those fields are only ever populated from the v2 leg. v1's global TradingPaused/TradingUnpaused are normalized into UserPaused/UserUnpaused with an empty user (v1 has no per-user pause targeting).

Quickstart

substreams run polymarket-fills-substreams@v0.1.0 map_fills -e polygon --start-block -1

Validation

# 1. Every fill-extraction module is index-filtered on lowercase addresses.
substreams info ./polymarket-fills-substreams-v0.1.0.spkg | grep -A2 "Name: map_"

# 2. Non-empty v2-era output, cross-checked against colindickson/polymarket-exchange.
substreams run ./polymarket-fills-substreams-v0.1.0.spkg map_fills -e polygon --start-block 93000000 --stop-block +2000 -o jsonl

# 3. Non-empty v1-era output.
substreams run ./polymarket-fills-substreams-v0.1.0.spkg map_fills -e polygon --start-block 40000000 --stop-block +2000 -o jsonl

# 4. Processed-block density is well below 100% during a quiet historical range.
substreams estimate ./polymarket-fills-substreams-v0.1.0.spkg map_fills -e polygon --start-block 33605403 --stop-block 45000000 --yes

Neg Risk CTF Exchange V2's OrderFilled/OrdersMatched/FeeCharged decode was validated the same way as everything else in this package: raw eth_getLogs against that address over blocks 93,000,000–93,003,000 was bucketed by (topic0, indexed-param count, non-indexed byte width), then map_fills/map_fee_events over the identical range were checked for an exact count match — 58,490 OrderFilled, 24,471 OrdersMatched, 25,906 FeeCharged, all exact.

Built with Substreams Skills (substreams-dev, substreams-ethereum).