Verdict D

We remove the best periods and trades from what you uploaded and measure what is left. If the total falls to zero or below, it rests on a few events that may not repeat. This is not a forecast.

7 of 8 scenarios end at zero or below.

On the curve (compounded total return)

On the closed trades (net result after commission and swap)

A total and a maximum drawdown do not say what the history was like to live through: how long it went without a new high, how long the worst fall took to come back, and what the worst day and month were. These are the numbers that make people switch a system off.

The worst fall is not regained by the file's last date.

The deepest falls Measured

Each fall runs from the last point at a high to its lowest point and ends on the first date back at that high. One not back by the file's last date is still open.

The curve is rebuilt from closed trades: open losses do not show, so the real falls lasted and measured at least this much.

Calendar days from the uploaded equity curve; months from each month's last point.

Your trades grouped by entry day and time. If nearly all the result comes from one day or one session, a change of server time, holidays or news can erase it.

Entry times as the file states them (platform or server time); net result after the fees the file itemises per trade.

We look for the moment the curve's average return changed most and measure whether that change is larger than the normal swing of its returns (a CUSUM test that allows for one return influencing the next). It does not change the class.

Changed The average return changed around 18 Mar 2026 (probably between 12 Nov 2025 and 22 Jul 2026): +149.4% a year before and -99.2% a year after. With p = 0.038, chance alone is unlikely to explain a difference like this.

CUSUM of the returns in time order (Ploberger and Kramer); cautious long-run variance; p-value from the Brownian bridge.

Each return in the file is placed by the VIX (how much the options market expects the S&P 500 to move over the next month) at the close of the market day before it starts: a calm market below 20, a turbulent one from 20. Since 1990 the VIX has closed at 20 or more on about one day in three. Period: 2025-09-19 to 2026-09-16. Measured

The gap in mean return between the two columns (0.80 standard errors) is not enough to say it behaves differently depending on the market. Measured

VIX: public data from FRED (series VIXCLS, from CBOE) read when the report was made. It measures US equities: if the strategy trades another market, read it as a general gauge of fear in markets. It does not change the class.

The curve's levels, in dollars, converted at each day's exchange rate (the Federal Reserve's New York noon buying rate), from 2025-09-19 to 2026-09-16. If you live in another currency, this is what the account was worth in it. The difference from the dollar row comes from the exchange rate, not the strategy: when the dollar rises against your currency the result in it rises, and when it falls, it falls. Measured

The file does not say which currency the account is in, so it is read as dollars. If it is not, this section does not apply.

US inflation over those dates was +3.1% in total. Measured

The rows "after its own inflation" divide by each country's official consumer price index of each month, or that of the latest month published; a currency without a current official index shows only its row before inflation. The return a year is shown from one year of history. Exchange rates and US prices from FRED, read when the report was made. It does not change the class. Consumer prices: Mexican peso, Source: INEGI, Índice Nacional de Precios al Consumidor (INPC), used here to take inflation out of the balances; real, Banco Central do Brasil (IBGE's IPCA); euro, Eurostat (through FRED); pound, Office for National Statistics, licensed under the Open Government Licence v3.0; yen, created by editing Japan's Consumer Price Index (Statistics Bureau, Ministry of Internal Affairs and Communications), through e-Stat; Canadian dollar, Bank of Canada (Statistics Canada's CPI, available free of charge at bankofcanada.ca); Swiss franc, Eurostat's harmonised index.

What a trading journal would tell you: whether losses are held longer than gains, whether a new trade follows a loss quickly, and how trades do after a losing streak. It does not change the class: these are questions to ask.

Closed trades by entry and exit time; net result after the fees the file itemises.

When a robot or a signal trades several markets, the total can come from one of them while the others lose. It does not change the class: these are questions to ask.

Closed trades by the instrument the file names; net result after the fees the file itemises.

At the same share of losing trades and in random order, the longest losing run is typically 4 in a row, and 1 history in 20 reaches 6. This history had 12. Measured

The losses came closer together than chance explains: a run like this shows up in fewer than 1 in 20 random orders. It usually points to losses that depend on the kind of market, or to positions open at the same time.

How much of this could be chance?

With 344 trades, every figure has a margin. 95 % range: the underlying values consistent with these trades, if each is independent of the others and the system did not change. It is not a prediction.

The range of the average per trade or of the profit factor includes break-even (0 and 1): with these trades the system cannot be told apart from one that neither wins nor loses per trade.

Long

Short

The more configurations are tried, the higher the best one comes out even when none has an edge. Here the file's Sharpe sits next to what pure luck would give with the configurations counted, using the published math of Bailey and López de Prado and of Harvey and Liu. It is the same calculation that decides the "Number of settings tried" dimension, in numbers.

The files do not say how many configurations were tried before this one was picked. The table shows how much history each search size would need: ask the vendor.

The file's Sharpe: 0.57. History: 12 months. Measured

With under a year of history, an annualised Sharpe moves a lot on little data: read it as an order of magnitude.

E[max Sharpe] of unskilled trials (Bailey & Lopez de Prado); minimum backtest length (Bailey, Borwein, Lopez de Prado & Zhu); Bonferroni haircut (Harvey & Liu).

Rules simulated: Generic · Two-step evaluation, phase 1. Generic reference rules, not any one firm's terms.

Open losses The file shows the balance only, and the red flags found open losses the balance hides (Hidden floating drawdown). They are not counted here, so these figures come out optimistic: do not decide with them without the equity curve with floating results.

How much does it change with what this report found?

The same program (Generic · Two-step evaluation, phase 1), resampled the same way, on different stretches of the history or with what the report discounts. These are scenarios of the same history, not predictions: if the figure drops sharply out of sample or with costs, the full-history figure is optimistic. Measured

The same optimistic figures as above apply to this table: the balance hides open losses.

At what size? The challenge at 0.5x, 1x, 1.5x and 2x

The same program as the ladder (Generic · Two-step evaluation, phase 1), at another size. In each row, reaching the target, breaking a loss limit and not reaching the target within the cap share out all the simulations, counting every phase. The table shows what changes with the size; it advises none. “Reaches the target” counts only what gets there within that cap, which the simulation sets and the rules do not. At a smaller size the target takes longer: the simulations that move to “does not reach” ran out of days; they did not break a loss limit, which has its own columns. Measured

1x is the size of the history you uploaded: each simulated day gains or loses the same share of the balance as a day of the file. 0.5x is half that size and 2x is double. The shares at 1x are measured on the file's starting balance (1,000). Declared

Lot or risk per trade at 1x: Not measured the audit keeps neither the lot nor the stop loss of each trade, so the lot or risk per trade at 1x is not known

The simulated rules fix no account size: they are shares (of the starting balance or of the day's), so the table does not depend on the account size.

Method and assumption: the ladder's full-history row with every daily return multiplied by the size; it assumes that changing the size scales every daily return in the same proportion, as linear leverage does when the costs grow in proportion to the size (the same cost per lot) and the execution does not worsen with more volume.

The same optimistic figures as above apply to this table: the balance hides open losses.

• Reference rules typical of two-step evaluations; not any one firm's terms.

Assumptions:

• Resampled estimate from the supplied history: it is not a prediction.
• Daily data cannot see intraday floating drawdown, so the estimate is optimistic against the daily and total limits.
• It assumes the future resembles the history and that every day with a non-zero return counts as a trading day.

Which firm's rules does your history fit?

The same history, resampled the same way, under each firm's published rules, from most to least likely to pass every phase of the program within the best-day rule, where the firm has one; ties go by name. It compares rules; it does not recommend buying any challenge. Measured

Figures from the full history without the reference cost; the ladder above shows how much they change.

The same optimistic figures as above apply to this table: the balance hides open losses.

1. Is this the only account running this strategy? Ask for the accounts that were closed or restarted too: showing only the one that went well is common.
2. Ask for the backtest of the same robot with the same settings: uploaded together with this account, the report compares the two trade by trade.
3. Losing trades last longer than winners: how does the system decide to close a loss?
4. What does the system do after several losses in a row: change size, pause, or enter again straight away?
5. Ask for the (floating) equity curve, not only the balance: the balance hides open losses.
6. Does the robot increase size after a loss? What is the largest size it can open?
7. Does the robot add positions against the move when price moves away? How many at most?
8. Ask for the full history with every deposit and withdrawal: how much money was deposited in total, when, and how much was withdrawn?
9. Which positions are still open, since when, and with what floating loss?
10. The history has jumps, gaps or repeated values: where does the data come from and how was it cleaned?

Sharpe corrected for autocorrelation (Lo, 2002): 0.49, against 0.57 from the plain calculation. Returns of nearby periods tend to move together: the plain Sharpe comes out inflated. Measured

When the returns are not taken as independent of each other, the Sharpe's variance grows 1.1 times: the probability that the true Sharpe is above zero goes from 68.75% to 67.70%. Not even ten times the 258 returns it has would take it to 95%. It is informational: the class uses the plain count. Measured

Trials used in the deflated Sharpe: 1 Not measured not declared; computed with 1, the most favourable case

Variance used: the larger of the one observed across the variants you uploaded and the one sampling error produces.

Stationary block bootstrap, resamples: 500 · block 20

Not measured no out-of-sample start declared

Not measured no benchmark uploaded

Not measured no variants uploaded