Build & AI

An Independent Backtest Audit of Your Code, Data and Fills

We check your backtest or trading bot for look-ahead, missing data, free fills and overfitting, then run it again with the fixes in. You get a written report.

Updated

A backtest that shows a profit is a claim about the past. Before money goes behind it, someone should try to break it. That is this job: we take your backtest, or the bot already running on it, and look for the ways it could be wrong.

You get a written report. Each finding names the line of code or the data that causes it, shows the effect on the result, and says how to fix it. Then we run the test again with the fixes in, so you see the number that survives.

Building a new bot is a separate job, on the algorithmic trading page.

What we check

The data. Gaps, duplicate rows, bad prints, prices from the wrong time zone. Markets or assets that stopped trading and are missing from the test, which makes the past look kinder than it was.

The clock. Every number the strategy reads has to have existed at that moment. A daily close used at noon. A revised figure used on the day the first estimate came out. An indicator computed over the whole series. Each one leaks the future into the test.

The fills. Which price the order filled at, and whether anyone was offering it. Fees, spread, slippage and partial fills. A strategy that trades often can live or die on these alone.

The number of tries. If many variants were tested and the best one kept, part of its result is luck. We count the trials and correct the Sharpe ratio for them, with the deflated Sharpe ratio that Bailey and López de Prado published in 2014. Our free calculator does the same sum in your browser.

The code that trades. In a running bot we also read the order path. What happens on a restart, a rejected order, a partial fill or a lost connection. Whether a limit can be switched off by the strategy it is meant to limit.

When to ask for one

  • Before you raise the size a bot trades.
  • When live results drift away from the test, and nobody can say why.
  • When you bought a strategy or a bot and want to know what it does.
  • Before you hand a developer’s code to your own team.

What you get

  • Your result reproduced on our side, from your code and your data, before anything is judged
  • A written report in which each finding names its line of code or its data, and shows its effect on the result
  • The number of trials behind the strategy you picked, and its deflated Sharpe ratio
  • For a live bot, a reading of the order path through restarts, rejected orders, partial fills and limits
  • The backtest run again with the fixes in, so you see the number that survives
  • A call to walk through the report with whoever wrote the code

How it runs

  1. NDA and read-only access

    We sign an NDA before you send anything. Then we take read-only access to the code and the data, or copies of them. We never need keys that can trade.

  2. Reproduce the result

    We run your backtest on our side and match your numbers. A result we cannot reproduce is the first finding.

  3. Look for the faults

    Data gaps and missing losers, look-ahead, fills nobody offered, costs left out, and the number of variants tried. In a live bot, the order path too.

  4. Fix and run again

    Each fault gets a fix, and the backtest runs again with all of them in. You see how much of the original result is left.

Questions we get asked

What do you need from us?

The code that produces the backtest, the data it runs on or a way to get it, and the result you want checked. For a live bot, read-only access to its logs and its order history as well.

Will you tell us whether to trade the strategy?

No. We tell you what the backtest supports once its faults are fixed, and how sure that number is. Whether to trade it, and at what size, is your decision.

What if the strategy came from someone else?

Then the audit matters more. We check what the code does, which can differ from what its seller described, and the report says where the two differ.

Can you fix what you find?

Every finding comes with its fix in the report. Your team can apply them, or we can, as a separate piece of work under the build service.

Is it confidential?

Yes. We sign an NDA before you send code or data, and at the end we return or delete our copies as the NDA sets out.

Further reading

Start a conversation

Let's scope the work.

Book a strategy call