Before you trust a backtest, find out what it is hiding — then size it like a risk manager would
Paste the strategy — its description or the code that ran it — and the equity curve, return series or trade log it produced. The browser computes every metric and lints the code for free. The audit lane says whether the result can be believed and by how much to discount it; the risk memo lane turns the real drawdowns and tails into limits, sizing and a monitoring plan.
Both examples ship with saved model runs for both lanes, so you can see a whole audit and a whole risk memo without signing in and without spending a credit.
What this does, and what it does not
The free lane is a real table reader and a real metrics engine, not a keyword search. It finds the date column and the value column, works out whether you pasted an equity curve, a return series or a trade log, whether returns are in percent or decimals, and what the sampling frequency is from the gaps between dates. It then computes, over every row, the numbers a risk manager would ask for: CAGR, annualised volatility, Sharpe and Sortino against the risk-free rate you set, the maximum drawdown with its dates and the longest time under water, Calmar, historical VaR and CVaR at 95 and 99 percent, skew and excess kurtosis, hit rate, profit factor, tail ratio, the probabilistic Sharpe ratio and the minimum track record length that would make the Sharpe statistically distinguishable from zero. Every one of those numbers is handed to the model as a fact, and the risk memo's own figures are checked against them afterwards - if the memo quotes a Sharpe the browser did not compute, this page says so next to it.
The code lint looks for the shapes that make a backtest lie: a negative shift, a
back-fill, a centred rolling window, a scaler fitted on the whole sample, a signal multiplied by
the same bar's return, Pine's lookahead_on, backtrader's cheat-on-open, a universe
built from today's index members, a dropna that removes every delisted ticker,
costs set to zero or absent, a grid search with no out-of-sample split, unseeded randomness.
The claims you paste - a Sharpe, a drawdown, a return - are compared with what the series
actually computes. All of that is sent to the audit as flags it must confirm or set aside, and
the page reports which it did.
It reads; it never runs your code, never fetches a price, never reaches an exchange or a broker, and never tells you to buy or sell anything. It is a review of one backtest and a risk reading of one series, not investment advice. A credential in the paste is named and flagged for rotation, never repeated.
Past performance does not predict future results. A backtest is a simulation of the past; even one that survives every check here can lose money live. The limits, sizing and verdicts are a reading of the evidence in the series you pasted, not a forecast and not a promise of any return.
Nothing to hand? Load the , a pandas cross-sectional momentum strategy with a negative shift, a current-constituent universe, no costs and a Sharpe of 2.4 that is too good to be true, or the , a trade log from a futures carry book with real costs, a walk-forward split and a 2022 drawdown to reckon with. Both replay saved runs in both lanes for free.