Methodology: reconciling DB-side P&L to on-chain USDC delta within $0.02 across 320 trades

[manja316]

Posting the full methodology in case anyone is trying to reconcile their own Polymarket trade history to wallet USDC delta and wants something to crib from. This is the actual procedure I used on my 320-trade self-audit that surfaced the +$34.31 DB → -$90.72 on-chain divergence (the one pnl-truthteller was built to find).

Inputs

Three sources, all pulled read-only:

1. CLOB /data/trades for the wallet (paginated, taker_only=true).
2. CLOB /data/transactions for the same wallet (every USDC in/out, including fees and rebates).
3. On-chain USDC balance delta on the proxy wallet between min(trade_ts) and max(trade_ts) — single number from Polygonscan.

If (1) + (2) don't reconcile to (3) within a few cents, the bot's reported P&L is fiction.

Reconciliation steps

1. Normalize side: CLOB returns side: BUY/SELL on the maker, but taker_side is what mattered for me. I flip everything to taker frame so BUY = USDC out, SELL = USDC in.
2. Dedup trades on stable hash of (order_id, maker_order_id, transaction_hash, log_index). Keep row with largest match_amount. Without this, partial fills double-count by ~12% in my dataset.
3. Dedup transactions on (tx_hash, log_index) alone — op_hash is not a primary key in proxy-wallet flows.
4. Compute realized P&L per market: Σ(sell_proceeds) − Σ(buy_cost) − Σ(fees), then sum across markets where redeemed_at IS NOT NULL or current_position == 0.
5. Compute unrealized: Σ(current_position × mid_price) from /clob/price snapshot.
6. The check: realized + unrealized − initial_USDC == current_USDC (on-chain). In my data this landed at $0.02 — within the rounding noise from the 6-decimal USDC representation.

What I expected vs what I found

Metric	DB said	Reality (on-chain)	Delta
Realized P&L	+$34.31	−$90.72	−$125.03
Fee burn (320 tx)	$7.20	$58.41	−$51.21
Slippage on volume	0.4%	6.8%	−6.4 p.p.

The fee gap was the single biggest contributor. Bot was logging fee = 0.4% on every trade per its config; chain said the realized fill price was also hit with maker-spread crossing on retry. Two different leaks, both invisible from inside the bot's own logs.

Edge case I'm still unsure about

When a proxy-wallet executeTransaction bundles multiple maker fills, I count each maker fill independently in step 2. I have not confirmed whether deduping at bundle level would change totals — it didn't on my N=320 set but I'd take a data point from anyone running ≥1k trades.

Repro

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

Read-only. Outputs a markdown report with the by-exit-reason table that's the actually-useful one. If your numbers reconcile cleanly, great — your bot is honest. If they don't, the breakdown shows you where the leak is.

Drop your numbers below if you run it. Trying to collect n≥10 reconciled wallets to publish a slippage index (see Discussion #3).