Best Practices

Key metrics: EV, Sortino, drawdown, R-chart

Quantprove scores ten core metrics across its modes: EV per trade (caps at 0.50 in scoring), Sortino (caps at 1.80), CVaR 95, max drawdown duration in trades, profit factor, the confidence multiplier (sqrt(n)/sqrt(500)), EV retention, monthly stability, win rate, and edge velocity. Three tools read them: the R-chart, benchmark compare, and projection.

Last updated: 2026-06-10

What metrics does Quantprove score?

Quantprove reads a CSV trade log and reduces it to a small set of metrics, each tied to a scoring bucket. You interpret three things from any metric: its raw value, the direction that is favorable, and how much weight it carries in the score. The list below covers every metric the engine surfaces, grouped by what it tells you about an edge.

Two metrics carry their own glossary entries because they roll up many inputs: the Edge Score and the Stability Score. Read the Edge Score glossary entry and the Stability Score glossary entry for the full bucket breakdown. The per-metric definitions here explain the components those scores aggregate.

How do you read EV per trade, edge velocity, and profit factor?

EV per trade is the mean profit-and-loss per trade, in your CSV’s unit (R-multiples or dollars). It is the cleanest single read on whether a system has a positive expectancy. In Backtest scoring, EV per trade earns up to 15 points and caps at an EV of 0.50; above that you have a strong edge, and more EV does not buy more points.

Edge velocity multiplies EV per trade by trades per month. A 0.20R edge fired 40 times a month compounds faster than a 0.40R edge fired 4 times. Profit factor divides gross winning P&L by gross losing P&L; a value above 1.0 indicates a net-positive system, and the further above 1.0, the more cushion each losing run has.

  • EV per trade: mean P&L per trade. Favorable = higher. Scoring caps at 0.50.
  • Edge velocity: EV x trades/month. Favorable = higher. Rewards frequency, not just size.
  • Profit factor: gross wins / gross losses. Above 1.0 = net positive; below 1.0 = losing.

EV per trade caps at 0.50 in Backtest scoring. Past that point you have a strong edge, and the model stops paying for more.

How do you read Sortino, CVaR 95, and drawdown duration?

Sortino measures return per unit of downside volatility, ignoring upside swings that should not be penalized (Sortino and Price, 1994). Quantprove computes it per trade with no annualization, and the score caps Sortino at 1.80. CVaR 95 (conditional value at risk) is the mean of your worst 5% of trades, which answers "when it goes wrong, how wrong" better than a single max-loss figure.

Max drawdown duration is measured in trades, not days, because a date-based count distorts on active versus quiet periods. It tells you how many trades the system spent below a prior equity peak. Loss streak distribution looks at how consecutive losses cluster; a system with the same win rate but tighter loss clustering recovers faster and scores higher.

  • Sortino: return per unit of downside risk. Favorable = higher. Scoring caps at 1.80.
  • CVaR 95: mean of the worst 5% of trades. Favorable = closer to zero (smaller loss).
  • Max drawdown duration: trades spent below a prior peak. Favorable = fewer trades.
  • Loss streak distribution: how consecutive losses cluster. Favorable = shorter, rarer streaks.

What are the confidence multiplier, retention, and monthly stability?

The confidence multiplier discounts a score for small samples. Backtest uses max(0.2, min(1.0, sqrt(n)/sqrt(500))), so a 500-trade log scores at full weight and a 20-trade log lands on the 0.2 floor. Validation uses min(1.0, sqrt(n_live)/sqrt(500)), with no floor, and it depends only on the live sample size. The multiplier is why a backtest with 18 trades and a 0.6R edge will not read as exceptional.

Retention (EV retention and Sortino retention) compares your live numbers to the backtest as a percentage: live EV divided by backtest EV. A retention near 100% indicates the live system tracks the backtest; a drop toward 50% indicates the edge faded in real conditions. Monthly stability uses the coefficient of variation across months; lower variation in monthly results scores higher, because a steady edge is easier to trade than a lumpy one. In simpler words... retention near 100% means live kept the promise.

A 20-trade backtest hits the 0.2 confidence floor in Backtest. The same edge at 500 trades scores at full weight.

Why does the R-chart use trade number on the X-axis instead of dates?

The R-chart is a cumulative R-multiple equity curve. Each point adds the next trade’s R to the running total, so the line shows how the edge accumulated. Quantprove plots trade number on the X-axis, never dates, because a date axis compresses active days into a sliver and stretches quiet weeks into wide gaps, which distorts the curve’s shape even when every Y value is correct.

A smooth, steadily rising R-chart is consistent with a repeatable process. A jagged curve with sharp jumps suggests results lean on a few outlier trades or variance. An optional secondary MMM ’YY date label can sit below the trade-number tick, so you keep a calendar anchor without surrendering the trade-number scale.

What does the Benchmark compare tell you?

Use the Benchmark tool to test whether an edge reflects skill or exposure to a rising market. It compares a strategy against an asset class, the S&P 500, EUR/USD, or another reference, and shows the correlation, beta, and alpha-style relationship alongside an equity overlay.

A strategy whose equity curve tracks the benchmark closely with high correlation may owe much of its apparent edge to a rising market rather than skill. A strategy that produces positive returns with low correlation to its benchmark is more likely to hold up when the underlying asset stalls or reverses. Read the overlay for where the strategy diverged from the market and whether those divergences were favorable.

How does System Projection model the future?

System Projection is a forward Monte Carlo simulator on the Projection tab. You set starting capital, monthly deposit, risk percent per trade, and a horizon in years; a prop-account mode adds challenge and funded targets plus drawdown limits. It returns a base-case compound equity curve and a Monte Carlo cloud of alternate paths from your historical R distribution.

The projection uses per-trade metrics and does not annualize anything; it resamples your actual trade outcomes forward rather than scaling a yearly return. The personal-account view reports final capital, total return, and risk of ruin; the prop-firm path reports the probability of passing a challenge and getting funded at the risk levels you set. Treat the cloud as a range of plausible outcomes, not a forecast.

What metric mistakes distort the read most?

Four mistakes recur, covered in full on common mistakes that distort a backtest. Insufficient sample size overstates an edge until the confidence multiplier discounts it. Overfitting produces a high Edge Score on a tuned backtest that collapses live, which Validation retention and the KS/AD test surface. Deposit and risk distortion break comparability when you mix R-multiples and dollars or change account size mid-stream. The fourth is the R-multiple annualization fallacy.

R-multiples are not percentage returns, so you never annualize them with a sqrt(252) scaling, which would produce figures like "36R per year" that mean nothing. Report Total R or EV per trade instead. Keep one unit across a comparison, and rescale R if your per-trade risk changes. The Best Practices hub links the deeper guides.

  • Small samples overstate edge; the sqrt(n)/sqrt(500) multiplier discounts them.
  • Overfitting inflates Edge Score; Validation retention and the KS/AD test expose it.
  • Mixed units or mid-stream account changes break backtest-vs-live comparability.
  • Never annualize R-multiples; use Total R or EV/trade, never "36R/year".

Frequently asked questions

EV per trade is the mean P&L across all trades, in R or dollars; it tells you the average outcome. Profit factor divides gross winning P&L by gross losing P&L, so it reads above 1.0 for a net-positive system. EV reports size; profit factor reports the ratio of total wins to total losses.
Quantprove counts drawdown duration in trades because a calendar count distorts on active versus quiet periods. Twenty trades in one busy day and twenty trades over a slow month represent the same recovery distance for the system, so trade number is the comparable unit. Dates can appear as a secondary label, never as the scale.
The confidence multiplier scales a raw score by sample size. Backtest uses max(0.2, min(1.0, sqrt(n)/sqrt(500))), so 500 trades score at full weight and 20 trades land on the 0.2 floor. Validation uses min(1.0, sqrt(n_live)/sqrt(500)) with no floor, keyed to the live sample size only.
No. R-multiples are not percentage returns, so scaling them with sqrt(252) produces meaningless figures like "36R per year." Report Total R or EV per trade instead. If your account size or per-trade risk changes, rescale the R values so the series stays internally consistent before you compare anything.
The Benchmark tool checks whether a strategy’s returns come from skill or from market exposure. It compares the strategy against an asset class such as the S&P 500 or EUR/USD and reports correlation, beta, and an alpha-style relationship with an equity overlay. High correlation to a rising benchmark suggests the edge may be riding beta.
No. System Projection resamples your historical per-trade R distribution forward and does not annualize. You set starting capital, monthly deposit, risk percent, and a horizon; it returns a base-case equity curve plus a Monte Carlo cloud. The prop mode reports the probability of passing a challenge and getting funded at your chosen risk levels.

References

Validate your strategy in under a minute.

Upload your trade log and read your Edge Score, free. No credit card required.

Start for freeHow it works

No credit card required·Swiss Made