Signal vs. noise · since your first backtest lied to you

Is your backtest real — or a mirage?

Upload your results. We tell you whether the edge is real or an artifact of overfitting — before you trade a single euro.

We only ever see your curve — never your rules.
Two ways in: a Freqtrade file or a plain CSVsee what we need (2 min read).
Equity curve · in-sample → out-of-sample
Real edge Mirage (overfit) Out-of-sample
Works with any strategy that has a track record
Equities Futures Crypto FX Options Multi-asset

Any market, any timeframe, any holding period — because NeuralTEST reads your returns, not your assets. If it produces an equity curve, a returns series or a trade log, we can test it.

A backtest, tortured long enough, will confess to anything.

The more parameter sets you try, the more likely your best curve is luck wearing the mask of skill. That's not a character flaw — it's mathematics, and it fools professionals every day.

The catch: an overfit strategy looks indistinguishable from a great one on the very chart you used to build it. The tell only shows up out-of-sample — when it's your money on the line.

NeuralTEST runs the statistical gauntlet that separates a durable edge from a beautiful accident. It sells rigor, not riches.

The Backtest Gauntlet

What we actually check

Five tests from the academic literature on backtest overfitting — the ones quants use to keep themselves honest, in plain language.

/ 01

Probability of Backtest Overfitting

Splits your configurations every which way and asks: how often does the in-sample winner underperform out-of-sample? A high PBO means your "best" is a coin flip.

CSCV · Bailey, Borwein, López de Prado, Zhu
/ 02

Deflated Sharpe Ratio

Discounts your Sharpe for how many strategies you tried and how skewed and fat-tailed the returns are. The Sharpe that survives the deflation is the one to trust.

López de Prado & Bailey
/ 03

Reality Check

A bootstrap that asks whether your best rule beats the benchmark once data-snooping is priced in. Non-parametric, so fat tails don't sneak a false positive past it.

White's Reality Check · bootstrap
/ 04

Minimum Track Record Length

How long a live record you'd need before the Sharpe is statistically believable. If it's longer than your backtest, you don't have evidence yet — you have a hunch.

Bailey & López de Prado
/ 05

The modeling red flags

Costs, look-ahead, survivorship, capacity, point-in-time data and more — the modeling shortcuts that inflate a curve. You declare them; we weight the verdict accordingly.

Honest attestation · you answer
/ ✓

One unified verdict

No wall of statistics to interpret. Every test folds into a single traffic light — real, fragile, or mirage — with the specific reasons behind it.

Red · Amber · Green
How it works

Three steps, ninety seconds

Bring your curve

Drop a CSV — a matrix of configurations, an equity/returns series, or a trade log. We auto-detect the format. No integration, no code.

We run the gauntlet

PBO, Deflated Sharpe, Reality Check and the rest run in seconds, right here — your file is processed in memory and deleted, never stored.

Get the verdict

A traffic-light call plus the reasons: which test flagged, by how much, and what to fix before you risk real capital.

The output

A verdict you can act on, not a stats dump

The example on the left is a strategy that looked spectacular in-sample. Once you account for how many configurations were tried, the edge evaporates — and NeuralTEST says so, plainly.

See exactly which test failed and by how much
Know whether to fix, keep testing, or walk away
No promise of profit — a measure of evidence

Your rules stay yours

NeuralTEST analyzes the return series — the shape of the curve — not the logic behind it. We never ask for your signals, your code, or your parameters. Files are processed and discarded, nothing is stored. Bring your own results; keep your own edge.

Your first test is free

Find out before the market does.

Upload a curve and get an honest verdict in under two minutes. No signup for your first check.

Run the test First, how to prepare your file
The science

Built on peer-reviewed research

NeuralTEST doesn't invent the statistics — it implements the field's established, published methods. Here is exactly what runs under the hood, and where each test comes from.

View the 6 peer-reviewed sources & what we use from each+
  • Probability of Backtest Overfitting (PBO), via CSCV
    We use the Combinatorially Symmetric Cross-Validation procedure to estimate how often the best in-sample configuration underperforms out-of-sample — our core overfitting metric.
    Bailey, Borwein, López de Prado & Zhu (2017). “The Probability of Backtest Overfitting.” Journal of Computational Finance 20(4).
  • Deflated Sharpe Ratio
    We deflate the observed Sharpe for the number of trials attempted and for non-normality (skewness, kurtosis) and sample length — the probability the edge is real, not the luck of the search.
    Bailey & López de Prado (2014). “The Deflated Sharpe Ratio.” Journal of Portfolio Management 40(5).
  • Probabilistic Sharpe Ratio & Minimum Track Record Length
    We compute the probability the true Sharpe exceeds a threshold (PSR) and the minimum sample length needed to trust it (MinTRL) — the honest answer to “do I have enough data yet?”
    Bailey & López de Prado (2012). “The Sharpe Ratio Efficient Frontier.” Journal of Risk 15(2).
  • Reality Check for data snooping
    A non-parametric bootstrap test of whether the best rule truly beats the benchmark once data-snooping is priced in — robust to the fat tails that fool parametric tests.
    White (2000). “A Reality Check for Data Snooping.” Econometrica 68(5). Non-parametric variant popularized by Aronson (2006), “Evidence-Based Technical Analysis,” Wiley.
  • Backtest-overfitting framework & the deadly sins
    The framing of how overfitting destroys out-of-sample performance, and the checklist of modeling shortcuts (costs, look-ahead, survivorship, capacity) we ask you to attest.
    Bailey, Borwein, López de Prado & Zhu (2014). “Pseudo-Mathematics and Financial Charlatanism.” Notices of the AMS 61(5).
  • Reference methods & conventions
    General reference for the implementations, cross-validation conventions and financial-ML machinery behind the engine.
    López de Prado (2018). “Advances in Financial Machine Learning.” Wiley.

What is ours: the implementation, the unification of every test into one verdict, the honest-attestation layer and the plain-language explanations. What is the field's: the statistical methods above — and that is the point. The rigor comes from published, peer-reviewed science, not a black box you have to take on faith.

NeuralTEST is an independent implementation of these published methods. It is not affiliated with, sponsored by, or endorsed by the cited authors or publishers.