Introduction

Kurzharr is a validation platform for systematic traders. It does not try to find you a winning strategy. It takes the one you already have and runs it through a fixed battery of statistical checks whose only job is to catch the ways a backtest can look better than the strategy really is.

Most backtesting tools report a Sharpe ratio, an equity curve and a win rate, and leave the interpretation to you. The trouble is that those numbers are the easiest thing in trading to inflate by accident — a lookahead bug, a sample that happens to span one kind of market, costs that were never subtracted, or fifty quiet re-runs of the same idea until one of them clears. A strategy that dies in live trading almost always passed its backtest first.

What a run produces

Every strategy you submit goes through the same eight-check validation stack, in a fixed order, with the bar it must clear locked in before the run starts. You do not get a score out of 100. You get a scorecard profile: which checks passed, which failed, and — when the shape of the failure matches a pattern the platform has already characterized — a plain-language account of what the evidence says went wrong.

The verdict at the top of the scorecard is one of four: Cleared (every check passed), Cleared with one gap (one non-veto check fell short), Has not cleared (two or more did, or a veto), or Not enough trades to grade (the run traded too little for the graded checks to run at all). “Cleared” is a deliberately narrow claim: not “this looks good”, but “this survived every check we could think of to make it look bad.”

What it will not do

The platform tells you what the evidence says about why a check failed. It does not tell you what to try next. It will not rank your strategies by potential, predict which style suits the current market, or suggest parameter values. That line — diagnosis, not prescription — is the whole point. A tool that helps you tune a strategy until it passes is a tool for fooling yourself, and the market is already well supplied with those.

“Cleared” also means only that the edge is statistically real, not that it is large enough to be worth trading or that it beat simply holding the asset. The result panel shows an annualized return and a buy-and-hold line next to the strategy’s curve so you can judge that, and you can add a minimum-profit-factor bar to your own criteria if you want it enforced.

Where the checks come from

The stack is a mechanized version of the process used in the research project behind the platform. In that project it rejected five of five candidate strategies before any of them reached real capital, and it was calibrated against a known, real market effect first, to confirm it can tell genuine signal from noise rather than simply failing everything. One of those five had a full-sample Sharpe of 1.16 that the temporal-halves check showed had already decayed to roughly nothing — the kind of result that ships as a winner almost everywhere else.

Who it is for

• —Systematic and algorithmic traders who have had at least one strategy pass a backtest and then lose money live. That experience is what makes “truncated-rebuild leak audit” read as a solution rather than jargon.
• —Traders already backtesting in Python or pandas who want the methodology done properly and the same way every time.

It is a weaker fit if you are looking for trade signals or market forecasts — the platform produces neither — or if your strategy is discretionary and not yet expressible as explicit rules.

What v1 does not include

• —Live execution or broker order routing. Validation only.
• —Arbitrary Python strategies. You compose a declarative spec and, on the Research tier, reference contract-checked custom indicators.
• —Uploading your own historical data, or non-standard bars — Renko, volume, range, tick.

From here: Getting started walks through building and running your first strategy; the validation stack explains each check; the scorecard & verdicts covers how to read the result.