Trading Consistency Is a Number, Not a Virtue

The Word Is Doing Two Jobs

When a mentor tells you to be more consistent, they mean your behaviour: take the same setups, at the same size, in the same session, by the same rules, whether or not the last trade won. Call that process consistency. It is a statement about inputs.

When a metric tells you that you are consistent, it means your results: how tightly your trade outcomes cluster around their own average. Call that outcome consistency. It is a statement about outputs, and it is what every number sold as a consistency score actually measures - standard deviation of returns, SQN, Sharpe, and the rest.

The two are connected, loosely. Good process tends to produce tighter outcomes. But the connection runs one way at best, and it is weak enough that you can sit in any quadrant: disciplined process with wild outcomes is the normal state of a breakout trader, and sloppy process with tight outcomes is the normal state of anyone quietly cutting winners early. No widely used metric measures process consistency. If that is the thing you are trying to fix, the number on your dashboard is not going to tell you about it.

The rest of this post is about the measurable one, because the measurable one is the one that gets quoted at you, misread, and occasionally optimised in exactly the wrong direction.

The Arithmetic: Consistency Is the Spread of Your R-Multiples

To measure the spread of your results you first need results that are comparable to one another, which dollars are not. A $400 win on a trade where you risked $200 and a $400 win on a trade where you risked $2,000 are not the same event. Normalise each outcome by the risk you had on that specific trade and you get an R-multiple: the first is +2R, the second +0.2R.

Now the series is comparable, and consistency is simply its standard deviation. Low dispersion means your trades resemble each other. High dispersion means a handful of outcomes are doing most of the work in both directions. That figure is meaningless alone, though, because it says nothing about whether the cluster sits above or below zero. It only becomes interpretable next to the mean of the same series, which is your expectancy.

Two traders, forty trades each, identical expectancy:

They finish in exactly the same place. The difference is entirely in the path: A gets there through a sequence of large gains and large losses, which means deeper drawdowns along the way, which means a higher chance of breaching a limit or losing confidence before the arithmetic pays out. B gets there almost boringly. If you are trading a funded account with a fixed loss limit, the path is not a detail - it is the thing that decides whether you are still trading at trade forty.

That is the real case for caring about dispersion, and it is narrower than the usual sermon. Lower dispersion does not make you more money. It makes the same money survivable.

SQN Is a t-Statistic, Not a Grade

The number in that last row is the System Quality Number, and it is the standard way of combining the mean and the spread into one figure:

It is worth recognising that formula, because it is not a bespoke trading metric. Mean divided by standard deviation, scaled by the square root of the sample size, is the one-sample t-statistic on the usual sample-deviation convention: the standard test of whether a sample mean differs from zero. SQN is the t-statistic of your R series.

That reframes the famous threshold. An SQN above 2.0 does not mean your system is good. It means your average R sits roughly two standard errors above zero, which is approximately the conventional threshold for saying an effect is distinguishable from noise at all. It is a statement of confidence that you have any edge, not a measurement of how large that edge is. A tiny, reliable edge and a large, erratic one can produce the same SQN.

It also explains why the number is so easily gamed by accident. Three separate quantities feed it - edge size, dispersion, and sample count - and the output is one scalar. When your SQN moves you cannot tell from the figure alone which of the three moved, and one of them moves on its own just by continuing to trade.

A Tight Cluster of Losses Scores Best

Here is the degenerate case, and it is not a hypothetical corner - it is the natural consequence of treating dispersion as a virtue in itself.

Take a trader who enters badly and gets stopped out at almost exactly the same distance every time. Fourteen trades, fourteen losses, average −0.92R, standard deviation 0.11R. That is a trader whose results are astonishingly consistent - far more consistent than anyone profitable. Their SQN is about −31. The magnitude dwarfs anything a real edge produces, and it is entirely an artefact of the denominator: as cross-trade variance approaches zero, mean divided by standard deviation grows without bound, at any sample size and regardless of sign.

This is not an exotic failure. Any approach that exits nearly every position at the same fixed stop produces it structurally. It is common enough that SignalDeck's own Sharpe calculation carries an explicit guard against it: the figure is withheld entirely once the raw mean-to-sigma ratio passes a bound, and withheld again below ten closed trades, because past that bound the number is describing near-determinism rather than risk-adjusted return.

The lesson generalises past the arithmetic. Consistency is a second-order property. It is only worth measuring on a system that has already demonstrated positive expectancy, because on any other system it is measuring how reliably you lose. Check the sign before you admire the spread - and if you want the more common version of this mistake, where a single flattering ratio hides what is underneath it, that is the subject of why win rate lies.

Sample Size Is Inside the Metric, Not a Caveat On It

Because √n is a factor in the formula, an unchanged system scores higher the longer you trade it. Nothing about the edge improves. The confidence that it exists does. Take a distribution that averages 0.20R with a standard deviation of 1.0R and never changes:

Two practical rules fall out of this. Never compare your SQN to another trader's unless both were computed over a similar number of trades, because otherwise you are comparing sample sizes with extra steps. And never read a high SQN on a short record as evidence of a good system: at low n the estimate of the standard deviation is itself unstable, so a run of similar outcomes produces a flattering figure that regresses as soon as the distribution shows its tails. Thirty closed trades per setup is the usual floor for looking at any of this, and that is a floor rather than a target.

The corollary is the useful one. Because the score drifts upward on its own, a falling SQN on a growing record is a genuine signal - the mean or the spread must have deteriorated enough to overcome the √n tailwind. That is one of the cleaner edge decay warnings available to you.

The Inconsistency the Metric Cannot See

Now the part that matters most, and the reason a trader can post a respectable SQN while running an account that is one bad week from zero.

R-multiples are normalised by the risk on each individual trade. That is what makes them comparable, and it is also what makes them blind. If you risk 0.5% of the account on Monday and 4% on Thursday, and both trades hit a two-to-one target, both are recorded as +2R. The series shows two identical outcomes. Your equity curve shows something else entirely.

So position-size drift - the single most destructive form of inconsistency there is, the one that actually ends accounts - is invisible to the standard deviation of R by construction. It does not show up as noise in the metric. It does not show up at all.

Measuring the second one is not difficult, and it does not need a new metric. Pull the planned risk on every trade, express it as a percentage of account equity at the time, and look at the range. If your stated rule is 1% and the record shows a band from 0.4% to 3.5%, you have found the problem, and it is not a strategy problem. The usual causes are recognisable: sizing up after a loss to make it back, sizing up on the setup that feels obvious, or sizing by contract count out of habit and letting the instrument's volatility set your real exposure.

The fix is mechanical rather than psychological, which is the good news: fixed-R position sizing makes the risk per trade a calculation rather than a decision. That removes the discretion at the exact moment discretion is least reliable, and it is the only intervention in this post that reliably changes a number.

Setup drift is the second form, and it is the easier one to catch: trades taken outside your defined criteria. It needs only that each trade carries a strategy label applied at entry, after which the off-plan trades can be grouped and scored as their own population. In most records they are a small share of the trades and a large share of the losses.

What a Prop Firm Means by Consistency (Something Else Entirely)

If you trade a funded account you will meet the word again as a hard rule, and it does not mean dispersion. A prop consistency rule caps the share of your total profit that may come from your single best day.

Topstep documents a Consistency Target of 50% in the Combine and a 40% figure on the Express Funded Account, where it governs payouts rather than passing. The5ers publishes a 50% per-day consistency figure on some plans, while Hyper Growth is documented as having no consistency restriction. These figures vary by programme and by promotion, so read the specification table for the exact plan you are buying rather than assuming - the full comparison is in the Topstep and The5ers rules breakdown.

What the rule is doing is obvious once you see it from the firm's side: it exists to stop an account passing on one lucky day and then being handed real capital. It is a profit-concentration limit, an underwriting control. It is not a measure of the spread of your returns, and a trader can satisfy it with a terrible SQN or violate it with an excellent one.

The practical consequence is that it constrains your best days, which is an unusual thing to have to plan around. A strategy whose returns are concentrated in a few outlier trades - most breakout systems, structurally - can be genuinely profitable and still fail a consistency check, and the tracking it demands is a running ratio of largest day to total profit, which is not a number any generic analytics page shows you.

What SignalDeck Computes Here, and What It Does Not

Specifics, because a post arguing for precise measurement should be precise about its own tooling. Verified against the product as of September 2026.

SQN is computed on the dashboard from the R-multiples of your closed trades, using the formula above. Two details worth knowing if you reproduce it in a spreadsheet: it uses the population standard deviation, so a sheet built on the sample convention will return a slightly lower figure, and it is a paid-tier analytic - on the free plan the field comes back empty rather than zero.

There is no minimum-sample floor on SQN. Sharpe, which is computed for the strategy-performance view rather than for this dashboard, is withheld entirely below ten closed trades and withheld again when the mean-to-sigma ratio goes degenerate - in both cases the figure is suppressed rather than clamped. SQN has neither guard, and will render on a handful of trades. Read the trade count next to it every time, and apply the sample-size table above yourself.

There is no rolling SQN, and no rolling-versus-lifetime divergence flag. What exists is a twenty-trade rolling window on two other things: expectancy momentum, which is your recent average R minus your lifetime average R, and win-rate velocity, which does the same for win rate. Those catch a deteriorating mean. Neither catches a widening spread, so the dispersion half of the story currently has to be read off the equity and drawdown curves rather than off a consistency number.

Risk dispersion is not computed at all. The fields needed for it are there - planned risk in money, the original stop, and a risk-as-a-percentage-of-equity figure snapshotted at entry - but no adherence or risk-variance metric is derived from any of them. That percentage carries a deliberate gap of its own: it is recorded when a trade is entered in the app and is intentionally left empty on imported rows, because filling it in afterwards would divide by today's balance and store a number that was never true. Across a mixed history, the check described in the previous section is currently a manual export and a sort. Given that this post argues it is the more important of the two dispersions, that is a real gap rather than a rounding error, and it is stated here rather than left for you to discover.

What does exist alongside SQN: average R grouped by your own setup-quality rating, which is the closest thing to a process-consistency cut in the product, plus expectancy and profit factor per strategy label, and a rolling win-rate curve. If any of the gaps above ship, this section is the part of the post that should change.

What to Actually Do With This

Four things, in order, none of which require you to be a better person.

Check the sign before the spread: establish positive expectancy over a real sample first, because consistency measured on a losing system is a description of the losing. Then read SQN with its trade count attached and never without it. Then measure the dispersion of your planned risk separately, since the metric everyone quotes cannot see it. And if you are funded, track your largest day as a share of total profit, because that is the only definition of consistency that can actually cost you an account.

"Be consistent" survives as advice because it is unfalsifiable. Once it has a number attached it stops being a virtue and starts being a measurement, which is less inspiring and considerably more useful.