How are you deduping partial fills against order_id when CLOB reuses IDs across batches?

[manja316]

Looking for input from anyone who's written their own Polymarket fill reconciliation. This is the corner where I burned the most hours building pnl-truthteller and I want to know if I solved it the same way everyone else does or if I'm still missing something.

The problem

For every market order I post, the CLOB API returns a single order_id even when the order fills across N maker orders. When I later pull /data/trades and /data/transactions to reconcile, I get:

• Multiple trade rows that share the same parent order_id
• Multiple transaction rows that point to the same underlying maker fill in some edge cases
• Occasionally a single fill split into two rows because of how OrderFilledQuantity events get emitted when a sweep retry lands on the same maker

If you sum naively, you double-count shares on the trades-side and triple-count fees on the transactions-side. Net effect: reported slippage looks better than reality, which is the opposite of the bug I was trying to surface.

What I'm doing in pnl-truthteller right now

Dedup key = (order_id, maker_order_id, transaction_hash, log_index) collapsed via a stable hash. I keep the row with the largest match_amount, drop the rest. For transactions, I dedup on (tx_hash, log_index) alone since the same op_hash can show up under two orders in proxy-wallet flows.

That was enough to get my 320-trade self-audit (+$34.31 DB → -$90.72 on-chain) to reconcile to within $0.02 of the wallet's actual USDC balance change. So I think it's right. But I built it solo, didn't have anything to cross-check against, and the random-wallet replication (0x1417…, 65 trades, +$32.36 DB → -$30.29 on-chain) is the only external data point I have.

Two questions

1. Anyone using a different dedup key? Specifically: do you trust transaction_hash + log_index as a primary key, or have you seen the CLOB emit the same (tx_hash, log_index) for distinct fills?
2. Edge case I'm unsure about: when a proxy-wallet executeTransaction bundles multiple maker fills, my current logic counts each maker fill independently. Should I be deduping at the bundle level instead? I haven't seen this affect totals but my N is small.

If you want to reproduce against your own wallet first:

pip install pnl-truthteller
pnl-truthteller --wallet 0xYourPolymarketProxy --output report.md

Read-only, no private key needed. Output is a markdown report — the "by exit reason" table is the one that surfaced my TIMEOUT-exit bleed.

If your slippage delta lands wildly different from mine (mine: ~5–7% on volume), I want to know — drop the number here.