We rebuild your bot’s PnL from the only sources that can’t flatter it — official market settlement, exchange fills and on-chain cash flow — then try to break every edge it claims before you put more money behind it.
Polymarket · Binance spot & perps · Deribit options · custom Python bots. Read-only exports only — we never ask for keys.
Before offering this to anyone, we ran it on ourselves: 6,452 settled trades reconciled against official outcomes, exchange data and the wallet. Almost none of the error was exotic. It was bookkeeping.
Every audit scores your bot on the same eight dimensions, so you can compare versions, strategies and vendors on one sheet.
Duplicate rows, fees applied to winners only, the wrong settlement source, positions restored across restarts.
Unclosed candles in features, timestamps off by a bar, survivorship in the symbol universe, dropped days in the backtest.
Fills at mid or last price, maker fills that never happened, ghost fills that matched off-chain but failed on-chain, adverse selection, and no-fills near expiry.
Out-of-sample split, day-clustered bootstrap intervals, multiple-testing correction, and how many trades a verdict actually needs.
Kill switch, risk limits, order idempotency, orphaned positions after a crash, what happens when an order’s fate is unknown.
Your ledger against exchange history or on-chain cash flow, market by market, until every dollar has a named cause.
Stale feeds treated as live, crash loops, silent restarts, heartbeats that report alive while nothing trades.
The numbers in your README, pitch deck or vendor sales page, reproduced exactly — then re-run the way the bot really trades.
All from our own fleet, anonymized. None of these bots crashed or threw errors. Each one simply reported an edge that wasn’t there.
A liquidity filter dropped 26 of 104 Fridays — the ones that moved most. Restored at the same Deribit costs, t fell from 1.83 to 0.34 and the worst week doubled to −25.5% of notional.
335 entries opened on the same bar as another trade. Removing the top 10 trades cut the mean from +57.9 bp to +12.5 bp.
Meanwhile the bots’ own records were $13.62 too optimistic: 16 markets traded but never logged, and 7 logged at the wrong value.
The order matters: we never argue with a number we couldn’t reproduce, and we register what would count as failure before we look at the results.
You send code, ledgers and read-only exports. We agree scope and the claims under test in writing.
We rebuild your headline numbers exactly. If we can’t, that is the first finding.
Every trade re-settled against the official source, exchange fills or on-chain cash flow, with fees applied to every fill.
Pre-registered tests: look-ahead, holdout, day-clustered bootstrap, cost stress, a bot-faithful replay.
Crash, restart and unknown-order scenarios simulated offline against your code — never against your account.
A scored report, a ranked list of fixes with file and line, and a clear call on what to do with the bot.
| Dimension | Score |
|---|---|
| Code correctness | 4 / 5 |
| Data integrity & look-ahead | 2 / 5 |
| Accounting truth | 3 / 5 |
| Execution realism | 4 / 5 |
| Live safety & state | 4 / 5 |
| Operational health | 4 / 5 |
| Statistical evidence | 1 / 5 |
| Claims vs reality | 1 / 5 |
Priced per bot. Larger fleets and firms with multiple strategies get a written quote after intake.
Is the PnL real?
Is the edge real, and is it safe to run?
Keep it honest after launch.
Payment in USDC on Base, Polygon, Ethereum or Solana, or by bank transfer. 50% on signing, 50% on delivery. Invoices state the exact network and amount.
Audits are technical reviews of software and data, not investment, legal or tax advice, and never a recommendation to trade any asset. A clean audit is not a guarantee of future results.
Case-file figures are historical results of paper, demo and small live bots, some simulated. Simulated results have inherent limits: they do not reflect real execution and may under- or over-state the effect of market factors such as liquidity. Trading involves substantial risk of loss.
Tell us the venue, the strategy in one sentence and the number you want verified. We reply within two business days with scope and a fixed quote.