The measurements first.
The opinions later.

The curve you were shown

Consider what an equity curve actually is. It is the cumulative sum of a set of outcomes, plotted in the order those outcomes happened to occur. Two facts follow immediately and neither is obvious from looking at the picture.

First, the endpoint depends only on the outcomes, not on their order. Added in R, the same numbers in any sequence give the same total; compounded under fixed-fractional sizing, each trade multiplies the balance by one factor and multiplication commutes, so the final balance is identical too. If you are being shown a curve to demonstrate profitability, the shape of the line is decoration — the final value is the entire claim, and it could have been stated as one number.

Second, and much more importantly, everything else about the curve depends almost entirely on the order. The deepest decline, the longest period spent below a previous high, the worst run of consecutive losses, the point at which an operator would have stopped trading — all of these are properties of the sequence, and the sequence is the one thing about a historical record that will certainly not repeat.

This is why a curve is not evidence in the way it is usually offered. It answers a question about the past that nobody is really asking, and it is silent on the question everybody is asking, which is what the range of plausible futures looks like.

One curve is one draw

The useful mental model is sampling. A strategy, if it has an edge at all, is a process that generates outcomes from some underlying distribution. A track record is one finite sample from that process, arranged in one particular order. Showing it is roughly equivalent to flipping a biased coin two hundred times, drawing the running total, and presenting the drawing as proof of the bias.

The drawing is real. The bias may well be real too. But the drawing tells you very little about the bias, and it tells you nothing at all about how bad the running total might have looked under a different arrangement of exactly the same flips.

The correct response is not scepticism about the record. It is to ask for the other draws — and because they do not exist, to generate them. That is the entire purpose of resequencing: take the outcome set you actually have, reorder it thousands of times, and look at the distribution of the numbers that depend on order.

What ordering alone does to the worst number on the page

The figure below does exactly that. A single configuration is evaluated once, producing a fixed set of trade outcomes. That same outcome set is then re-drawn across a hundred and fifty seeded orderings, and the deepest decline from peak is recorded for each. Nothing about the strategy changes between runs. No parameter moves. The trades are the same trades.

The spread is the point. If the record you were shown happens to sit at the favourable end of that distribution — and records that get shown to people tend to — then the drawdown figure quoted alongside it understates the drawdown the same strategy would routinely produce. Not because anyone is being dishonest, but because one ordering was observed and the rest were not.

The three curves below make the same argument in a form that is harder to dismiss. They are the gentlest, the middlemost and the harshest of those hundred and fifty orderings, drawn from the identical set of trades. They finish at precisely the same value, to the decimal — that is arithmetic, not coincidence. Almost nobody would have held all three to the end.

A worked example

Suppose a record contains 240 trades. Ninety-six of them were profitable, giving a hit rate of 40 percent. The average win was 2.4 R and the average loss 1.0 R, and total costs came to roughly 0.06 R per trade. Expectancy is then straightforward.

Now measure the uncertainty rather than the result. The dispersion of individual outcomes around that 0.30 R mean is about 1.67 R, which is typical for a distribution containing 2.4 R wins and 1.0 R losses. The standard error of the mean is that dispersion divided by the square root of the sample.

Two conclusions follow that most operators would not draw from the curve alone. The record is strong enough to justify believing there is an edge, and far too weak to justify sizing as though the edge were 0.30 R. And a plan built on the assumption of 72 R over 240 trades is a plan built on the upper half of a range that includes 21 R.

Note what happened here. Nothing was added except the dispersion — a number that every trade log already contains and that almost no trading report ever quotes. The curve had the information the whole time; drawing it as a line simply threw the information away.

Three things a single curve cannot tell you

Whether the result is repeatable. A curve has no confidence interval attached. Two records with identical endpoints and identical trade counts can differ enormously in how much evidence they contain, and the difference is entirely in the dispersion of the individual outcomes.

How bad it gets. The observed maximum decline is one draw from the distribution in Figure 2. Quoting it as the drawdown of a strategy is the single most common overstatement in retail trading material, and it is usually made in good faith.

Whether you would have held on. The curve shows what happened to the capital. It says nothing about what happened at trade 114, four months underwater, when the operator was deciding whether the system still worked. Every historical curve was, by construction, held to the end.

What to do about it

None of this requires new data. Everything below can be done with a trade log you already have.

Convert every outcome to R before anything else. Currency amounts encode position size, which encodes decisions you made about confidence — and those decisions contaminate every statistic downstream. R strips them out, and it is the unit every measurement on this site uses for exactly that reason.

Report dispersion beside every mean. An expectancy figure without the standard deviation of the outcomes behind it is not a measurement, it is a summary statistic pretending to be one. The two numbers cost the same effort to produce.

Resequence before you trust a drawdown. Shuffle your own outcome set a few thousand times and record the deepest decline each time. The number you should be planning around is somewhere in the upper region of that distribution, not the one your history happened to deliver.

Ask what the sample can support. Before concluding that a change improved anything, work out how many trades would be needed to detect an improvement of that size given your dispersion. The answer is very often larger than the entire record, which is a useful and deflating thing to know early.

Treat the curve as an illustration. It is a perfectly good way to communicate the shape of an experience. It is not a way to establish that the experience will repeat, and it should never be the artefact on which capital is committed.

The reframing is small and it changes almost everything: stop asking whether the curve is good, and start asking what range of curves the same process could have produced. One of them is the one you were shown. The others are the ones you will have to live through.