Reference

Metrics

Every performance metric TextToQuant reports, what it means, and how to read it.

Every backtest returns the same set of metrics so you can compare strategies on equal footing. This page is the definitive reference: what each number measures, and, just as important, how to read it without fooling yourself.

Metrics fall into four groups: headline performance, risk adjusted return, trade quality, and the overfitting aware statistics that make a result trustworthy.

Headline metrics#

These six appear on every results card. They answer "did it make money, and how?"

MetricWhat it measuresHow to read it
Total returnNet profit or loss over the test window, as a percentage of starting capitalThe top line result, but never judge it alone; a big return with a huge drawdown is fragile
Win rateShare of trades that closed profitableHigh win rate ≠ profitable. A 40% win rate can beat a 70% one if winners are bigger
Profit factorGross profit ÷ gross lossAbove 1.0 is profitable; 1.5+ is healthy; be suspicious of very high values on few trades
Sharpe ratioReturn per unit of total volatility>1 is good, >2 is strong, but see the overfitting section before trusting it
Max drawdownLargest peak to trough equity dropThe pain metric. Ask yourself: could you sit through this loss live?
Total tradesNumber of closed tradesSample size. Under ~30 trades, treat every other metric as noisy
Never read one metric alone

A strategy is a trade off between return, risk, and reliability. Total return tells you the reward, max drawdown tells you the risk, and total trades tells you how much to trust either.

Return over time & exposure#

Two runs of different lengths, or with very different time in the market, aren't directly comparable on raw return. These normalise for that.

MetricMeaning
Annualized return (CAGR)Total return expressed as a yearly rate, so a 6-month and a 3-year run compare on equal footing
Exposure / time in marketShare of bars you actually held a position. A strategy in the market 5% of the time carries very different risk from one always in
Avg / median bars heldTypical holding time per trade

Risk adjusted return#

Raw return ignores how much risk you took to earn it. These metrics divide reward by risk, so a calm strategy and a wild one become comparable.

MetricDivides return byUse it when
Sharpe ratioTotal volatility (up and down)The default risk adjusted score
Sortino ratioDownside volatility onlyYou don't want to be penalised for big winning months
Calmar ratioMax drawdownYou care most about surviving the worst stretch

Because Sharpe penalises upside and downside equally, a strategy with occasional large gains can look worse than it is. That's when Sortino tells the truer story. Calmar is the one to watch if your real constraint is "how deep a hole can I tolerate?"

More risk-adjusted & sizing scores#

MetricMeaning
Omega ratioProbability-weighted gains ÷ losses around a threshold, captures return skew the Sharpe misses
Information ratioReturn above a passive buy-&-hold, divided by the volatility of that excess. Differs from Sharpe, which compares to a flat zero
SQN (System Quality Number)Trade expectancy scaled by √trades (Van Tharp), one number for how tradable the system is
Kelly fractionThe bet size the edge implies. A ceiling to respect, not a target, full Kelly is famously wild

Trade quality#

Headline numbers hide how the money was made. These read the trade ledger directly.

MetricMeaning
Expectancy (R)Average profit per trade, expressed in units of risk (R). Positive expectancy is the whole game
Average win / lossMean profit of winners vs mean loss of losers
Payoff ratioAverage win ÷ average loss, how much bigger winners are than losers
R multipleEach trade's result as a multiple of the risk taken to enter it
MAE / MFEAverage worst drawdown (adverse) and best unrealised gain (favourable) reached inside a trade, how much heat you sat through before the exit
Success rateShare of trades that closed on one of your exit rules (stop, target, signal) rather than being force-closed at a horizon or the test's end
Win rate and payoff are two halves of one coin

A profitable strategy needs win rate × payoff to clear 1. A low win rate is fine if the payoff is high (trend following), and a low payoff is fine if the win rate is high (mean reversion).

Success rate ≠ win rate

Win rate asks whether a trade made money. Success rate asks whether it exited the way you designed (on a stop, target or signal) versus being closed because it ran out of time or the test ended. A trade can be a win that never hit your target, or a loss that cleanly hit its stop. Read the two together: a high win rate with a low success rate means your exits aren't doing the work.

Long vs short#

The ledger is also split by direction, so you can tell a genuinely two-sided edge from one that only works one way.

MetricMeaning
Long / short win rateWin rate computed separately for long and short trades
Long / short profit factor, expectancy, avg RThe same trade-quality read, per direction

A strategy that's only profitable long in a bull market is a very different bet from one that works both ways, this is where you catch it before it costs you.

Overfitting aware statistics#

A raw Sharpe is inflated by how many configurations you tried. Once a parameter search records the trials, these statistics appear on the metric cards, an honest read no other natural language tool ships.

MetricCorrects forHow to read it
Probabilistic Sharpe (PSR)Short samples & fat tailsHigher = more confident the Sharpe beats 0
Deflated Sharpe (DSR)How many configs you tried≥95% = survives the search; a big drop vs PSR = search luck
Overfit Probability (PBO)In sample best failing out of sample (grid / joint sweep only)Low is good; ≥50% = likely overfit
Haircut SharpeBonferroni correction for T trialsThe Sharpe you can still claim after the search
Min. backtest lengthSample too short for the searchWarns when history can't support that many trials
Reproducible by construction

Your query compiles deterministically to a fixed strategy spec, then a fixed engine scores it. The AI never sees, ranks, or tunes the numbers, so it can't steer toward a good looking result. Same query ⇒ same spec fingerprint ⇒ same test. Signals use only closed bar data (no look ahead). The results header shows a Reproducible chip and, when a hold out split exists, an always on out of sample verdict.

How grading uses these#

The letter grade you see on each run is not a single metric. It's a blend across four pillars, so a strategy can't earn an A by maximising return while ignoring risk.

PillarDriven mainly by
ProfitTotal return, profit factor, expectancy
RiskMax drawdown, Sharpe / Sortino
ConsistencyWin rate, equity curve smoothness, robustness checks
EdgeSample size and the overfitting aware statistics above
A grade is a starting point, not a verdict

Use the grade to triage which runs deserve a closer look, then validate the survivors with out of sample and Monte Carlo before you trust an edge.

Ready to see them live? Run a strategy and read its report end to end.

Buy ETH when the 50 EMA crosses above the 200 EMA, exit on the opposite cross, on the 1dRun in terminal