Guide › Reference

The numbers in Luxis

Every figure Luxis shows is here, under the name the screen gives it: what it measures, how Luxis computes it, and the line or colour that says whether it is good. When you ask the AI about a number, this is the definition it starts from.

Rules for every figure

  • A trade is a position opened from flat and closed back to flat on one symbol. A position still held is not a trade: it is in no trade statistic.
  • After commissions. Every trade's P&L is counted after the commissions of its fills, unless a figure says "before".
  • A live's commissions. A live records no commission on stock and option trades, only a crypto fee. Its P&L is therefore before those commissions; the Broker check says how much they were.
  • Two ways to measure. A backtest measures Sharpe, drawdown and the other daily figures on the account's value at each session close; its cards say "account at each close". A live, and a backtest made before those values were recorded, measure them on the closed trades, each trade's P&L on the day it closed. See Backtests.
  • Days. Daily figures count every market session of the window. A day with nothing closed counts as 0%. A daily figure scaled to a year counts 252 sessions a year, with no risk-free rate.
  • "—" means the figure cannot be computed: too few trades or days, nothing to divide by (no losing trade, no drawdown), no starting capital, or a price the broker did not give.
  • Colours. Green and red are Luxis's own thresholds, given under each figure.
  • Older backtests. A backtest's cards are computed when it runs and kept as they were. One run before a change to a figure shows that figure as it was then, so it can differ from what this page says. A live's figures are computed each time they are shown.

Backtest: totals

Initial Capital

The money the backtest started with. Total Return and Cost Drag on Return are measured against it.

Realized P&L

Also shown as: P&L.

What the closed trades made, after commissions. Positions still held at the end are not in it: they are in Open at End.

In a fold's row of the Walk-Forward panel, the P&L is the account's value at the end minus the starting capital, so a position still held at the end is included. It equals Realized P&L only when the run ended with nothing held.

P&L Before Costs

What the closed trades made before commissions. The fill prices already include the simulated slippage, so it is after slippage. Realized P&L is this minus Commissions.

Commissions

The commissions paid on the closed trades' fills, shown as a negative number. Those of positions still held at the end are inside Open at End.

Open at End

What the positions still held when the backtest ended were up or down. Each is valued at the close of its last bar in the final week, or at its cost if it had no bar in that week, minus everything paid for it, commissions included. Shown only when something was still held.

Equity

What the account was worth at the end: Initial Capital + Realized P&L + Open at End.

Total Return

(Realized P&L + Open at End) ÷ Initial Capital, in percent. A simple total, not compounded and not per year.

Return on Deployed

Realized P&L ÷ the money the trades put to work, in percent. Each trade's entry value (entry price × quantity) is added up, so capital reused by many trades counts each time: this is turnover, not the most money ever in use. Idle cash does not lower it.

Commissions Share

Commissions ÷ P&L Before Costs, in percent: how much of the profit went to commissions. "—" when P&L Before Costs is zero or a loss.

Avg P&L

Realized P&L ÷ the number of closed trades: the average trade, after commissions. It is the same number as Expectancy.

  • On a live, the line under it gives the average winning trade and the average losing trade.
  • In By Direction and By Symbol, it is the same, on that group's trades.

Total Trades

Also shown as: Trades.

The number of closed trades. Every ratio built on trades rests on this many.

  • In By Direction and By Symbol, "Trades" counts that group's trades.
  • In a fold's row, the count is the run's own.

Win Rate

Also shown as: Win.

The share of closed trades that made money after commissions. A break-even trade counts in the total, not as a win. Green above 50%, red below 40%. A low win rate is not bad by itself: read it with Risk/Reward.

  • On a live, the line under it gives the wins and the losses ("12W / 8L"); a break-even trade is in neither.
  • In By Direction and By Symbol, it is the same, on that group's trades.
  • In the Edge Report, it covers every out-of-sample trade of every fold, the open fold included.

Max Win

The best closed trade, in dollars, after commissions. Green when it made money; red when even the best trade lost.

Max Loss

The worst closed trade, in dollars, after commissions. Red when it lost money; green when even the worst trade made money.

Backtest: return and risk

Sharpe Ratio

Also shown as: Sharpe.

The average daily return divided by how much daily returns swing, scaled to a year. The higher, the more return for each unit of swing. Green above 1, red below 0. The line under it, "n N sessions", is the number of days it rests on.

  • What it is measured on. The account at each close, or the closed trades by exit day (see the rules above).
  • In a fold's row of the Walk-Forward panel, it is this same card's number.
  • For the Edge Report's Sharpe, see OOS Sharpe (mean) and Live Out-of-Sample.

Sortino Ratio

Like Sharpe, but only the falling days count as risk: the average daily return divided by the swing of the days below zero, scaled to a year. Green above 1.5, red below 0. "—" when no day fell.

Volatility

How much the account's value swings from day to day, scaled to a year, in percent. Green at 20% or less, red above 40%.

It reads the same days as Sharpe: the account at each close, or the closed trades by exit day over every session of the window. The line under it gives the number of sessions. "—" without a starting capital or a window.

CAGR

The steady yearly growth that turns the starting capital into the ending value, in percent.

  • The span. The run's window, the one Sharpe uses: a backtest's dates, a live's first run to today.
  • Short spans. Under three months it is not turned into a yearly figure: it is the period's own return, and the line under it says "period return".
  • Longer spans. From three months up it is a yearly rate, so a short window extrapolates: four good months read as a large yearly rate.
  • "—" without a starting capital.

Max Drawdown

Also shown as: Max DD, DD.

The largest fall from a high to a later low: in dollars, unless the figure shows a percentage.

  • Backtest card. On the account at each close ("peak to trough, account at each close"). A dip inside a day is not seen. On closed trades, it is the running total of their P&L from $0, checked after each trade.
  • A live's card ("peak to trough, realized"). The running total of the live's closed trades from $0. Open positions are not in it.
  • A fold's row in the Walk-Forward panel ("DD"). The run's card, in dollars.
  • The Live Out-of-Sample tile ("MAX DD"). The running total of the window's closed trades from $0.
  • The Edge Report gate is a percentage of a fold's capital: see The eight gates.

Calmar Ratio

CAGR divided by the worst fall in percent. Green above 2, red below 0.5.

  • The yearly return is the CAGR card's, over the same span: under three months, the period's own return.
  • The drawdown is measured in percent below the running high.
  • "—" when the run never fell: there is no drawdown to divide by.

Recovery Factor

The total profit in dollars divided by the Max Drawdown in dollars. It is not per year, so a longer run scores higher. Green above 3, red below 1. "—" when the run never fell: there is no drawdown to divide by.

Profit Factor

Also shown as: PF.

The dollars won by winning trades for each dollar lost by losing trades, both after commissions. Green above 1.5, red below 1.

  • No losing trade. It shows "—", with "no losing trade" under it: there is nothing to divide by.
  • In the Edge Report it covers every out-of-sample trade of every fold, the open fold included.

Expectancy

The average dollars per closed trade: win rate × average win − loss rate × average loss. It works out to the same number as Avg P&L.

Risk/Reward

The average winning trade divided by the average losing trade. A break-even trade counts as a losing one. Green above 1.5, red below 1. It does not use where the stops were. "—" when there is no winning trade or no losing trade.

Max Consec. Wins

The longest run of winning trades in a row, in exit order. A break-even trade ends a run.

Max Consec. Losses

The longest run of losing trades in a row, in exit order. A break-even trade counts as a loss. Compare it with what chance alone would give at this win rate.

VaR (95%)

A bad-trade threshold: the trade at the edge of the worst 5% of closed trades, in dollars. Five percent of the trades are counted, at least one.

  • Under 40 trades 5% of the trades is one trade or less, so it is the single worst trade: the same number as Max Loss.
  • Under 5 trades it is "—", as CVaR: too few trades to name a threshold.

CVaR (95%)

The average P&L of the worst 5% of closed trades, at least one, in dollars. Under 40 trades it is the single worst trade. "—" under 5 trades. It is not Expected Shortfall 95%, which averages days, not trades.

Backtest: how much to trust it

Sharpe Std. Error

How uncertain the Sharpe is, given how many days it rests on: a Sharpe of 1.5 with an error of 0.5 could reasonably be anywhere from about 1 to 2.

Probabilistic Sharpe

The probability, from 0 to 1, that the true Sharpe is above zero. It allows for the number of days, the lean of the returns (Skewness) and their extremes (Kurtosis). Needs 30 days. 0.95 is the bar: green at 0.95 or more, red below 0.5.

It does not allow for how many backtests were tried before this one.

Skewness

Whether the daily returns lean towards big losses (negative) or big gains (positive); 0 is symmetric. Needs 30 days. Green at 0.5 or more, red at −0.5 or less. Days with nothing closed count as 0%.

Kurtosis

How heavy the extreme days are. A normal spread of returns reads 3; red at 6 or more. Needs 30 days.

Days with nothing closed count as 0%. A strategy whose P&L lands on few days therefore reads high by construction.

Top-5 Concentration

The share of all winning dollars that came from the five best trades. Needs 20 trades. Red at 50% or more: the result rests on a few trades. Green at 25% or less.

Sharpe ex Top-5

The Sharpe again on the closed trades, after removing the five best. Needs 20 trades.

  • Always on trades. It is measured on the closed trades even when the Sharpe above it is measured on the account, so the two do not compare one to one.
  • "account wiped". It reads this when the remaining trades lose the whole capital.

Expected Shortfall 95%

The average return of the worst 5% of days, at least one, in percent of the account per day. Needs 20 days. With 20 to 39 days it is the single worst day. Unlike CVaR it averages days, not trades.

Ulcer Index

How deep and how long the account stayed below its last high, in one number: the square root of the average squared percent below the high, over every day. Green at 5 or less, red above 15.

Drawdown Duration

The longest stretch of sessions the account stayed below a previous high. A stretch still open at the end counts. Green at 21 or less (about a month), red at 126 or more (about half a year).

Time in Market

The share of the window's sessions on which at least one position was held at some moment ("N of M sessions"). It counts presence, not size: one share counts like the whole account. A live's open positions are not counted.

Backtest: costs and slippage

Slippage is measured only on market orders, against the bar's open or, for options, the mid between bid and ask. Stop and limit fills are not measured. A live records no slippage.

Avg Entry Slippage

The average price given up on entry, in percent of the price without costs. Red above 0.1%.

It averages the trades whose entry was measured; the line under it says how many. A trade whose entry filled at a price the strategy named has no measure and is left out, not counted as 0. "—" when no entry was measured.

Avg Exit Slippage

The same as Avg Entry Slippage, on the exit.

Avg Round-Trip Slippage

Avg Entry Slippage + Avg Exit Slippage: what a trade gives up, in and out. Red above 0.2%.

Gross P&L

What the closed trades would have made with no slippage and no commissions: Net P&L + the slippage in dollars + Commissions.

The slippage in dollars is each side's percentage applied to that side's value: the entry's to entry price × quantity, the exit's to exit price × quantity. A price better than the costless one lowers it.

Net P&L

What the closed trades made after slippage and commissions. It is the same number as Realized P&L.

Total Cost

The slippage in dollars plus the commissions. The line under it splits the two.

Cost % of Gross

Total Cost ÷ the size of Gross P&L, in percent. Red above 25%. "—" when Gross P&L is 0.

Cost Drag on Return

Total Cost ÷ Initial Capital, in percent: how many points of return the costs took.

Backtest: by direction and by symbol

By Direction splits the closed trades into LONG and SHORT. By Symbol splits them by symbol. A backtest groups options by their underlying, a live by the contract.

Each group shows Trades, Win Rate, Total P&L (the group's closed trades, after commissions) and Avg P&L, computed as above on that group alone. There is no minimum number of trades per group.

Backtests side by side

When the AI compares backtests, a table gives each run's Total P&L, Win Rate, trades, Avg P&L, Max Win and Max Loss as the run recorded them, and a curve per run: its closed trades one after the other, after commissions. A run that ended holding a position adds a last point at its Total P&L, so the curve ends on the table's figure.

The Walk-Forward panel

winner

The run chosen for the fold, with which the fold's out-of-sample window runs. It is the fold's recorded winner, the star given to the active fold, or the run on the live.

best

The training run of the fold with the highest Sharpe. It is shown only when it is not the winner.

Folds closed

Folds closed so far ÷ folds in the plan. Live folds count too, so a plan that goes on live can read "14 / 12".

Last OOS ends

The end of the last fold's out-of-sample window. On a plan that goes on live, it is the open fold's end.

A fold's row shows its run's own Sharpe, P&L, PF, Win, DD and trades: see Sharpe Ratio, Realized P&L, Profit Factor, Win Rate, Max Drawdown and Total Trades.

The Edge Report

The Edge Report measures only the out-of-sample windows. See Reading the Edge Report for how to use it.

  • The open fold. The fold that is running live counts in the figures built on trades: OOS trades, Win Rate, Profit Factor, Return (Σ %), the drawdown gate, Monte Carlo and the daily heatmap. It counts in the verdict, WFE and OOS Sharpe (mean) only once it closes.
  • Commissions. A backtest fold is counted after commissions, a live fold after the crypto fee recorded on each trade. A live records no commission on stocks and options; the Broker check says how much they were.

Validation

Also shown as: Verdict.

Whether the edge held on unseen data, in one word: Robust, Fragile, Overfit, Failed or Inconclusive. The server computes it from two numbers:

  • Robust: WFE 50% or more.
  • Fragile: WFE 30% to 50%.
  • Overfit: WFE under 30%.
  • Failed: OOS Sharpe (mean) at or under zero.
  • Inconclusive: no closed fold has an out-of-sample Sharpe, or no closed fold's winning training run has a Sharpe, or their average is at or under zero, so WFE cannot be computed.

"provisional · 8/12" means fewer folds have closed than the plan has: the verdict can still change.

EdgeFound Readiness

How many of the eight gates the strategy passes, "n / 8": four on performance (Perf) and four on robustness (Robust). A gate still pending counts as not passed.

"Integrity pending" is a group of checks not built yet: it counts nothing. The bars under the number are the eight gates: green pass, red fail, grey pending.

WFE (OOS/IS)

Out-of-sample Sharpe as a share of in-sample Sharpe: how much of the training Sharpe survived on unseen data. It is computed in two steps:

  1. Take the average out-of-sample Sharpe of the closed folds.
  2. Divide it by the average Sharpe of those folds' winning training runs.
  • The thresholds. 50% or more is Robust, 30% to 50% Fragile, under 30% Overfit.
  • "—" when no winning training run has a Sharpe, or their average is at or under zero.

Folds

The number of closed folds in the report, with the plan's training and out-of-sample months ("12 · 12m IS / 3m OOS").

OOS window

From the first fold's out-of-sample start to the last fold's out-of-sample end. The open fold is included, so the end can be in the future.

OOS trades

Also shown as: Sample.

All out-of-sample trades of every fold, the open fold included. The Sample gate asks for 100 or more.

OOS Sharpe (mean)

The plain average of each closed fold's out-of-sample Sharpe; the open fold is not in it. It is the number the Sharpe gate checks (1.0 or more).

A backtest fold's Sharpe is its run's own card. A live fold's is measured on its closed trades, by exit day, on the fold's capital base.

Live Out-of-Sample

The fold running live now, from the start of its window to today, on the live's closed trades whose exit falls in the window, across all its runs.

  • The tiles:
    • RETURN: P&L ÷ the capital base.
    • P&L: after the crypto fee recorded on each trade.
    • SHARPE: daily, over every session since the start.
    • WIN RATE, PROFIT FACTOR, TRADES: as in their own paragraphs.
    • MAX DD: dollars, from $0.
  • The capital base is the account's value when the run active at the window's start went live, plus what that run had realized before the start, after each trade's fee.
  • "day X of Y" counts calendar days.

Return (Σ %)

Also shown as: OOS ret %.

The out-of-sample return of every fold added up, the open fold included. Each fold's P&L is divided by that fold's own capital. Folds are added, not compounded, because each runs on its own capital. It is the last point of the Validated Performance curve. "—" when a fold has no capital base.

In the Fold Ledger, "OOS ret %" is one fold's own. A backtest fold that ended holding a position counts it, valued as in Open at End; the curve and Return (Σ %) count it too, at the fold's end. It is not a trade, so it is in no trade figure: OOS trades, Win Rate, Profit Factor, the drawdown gate, Monte Carlo and the heatmap.

Fold hit rate

The closed folds whose out-of-sample P&L was positive, out of the closed folds ("7 / 10 pos.").

Alpha vs SPY

The yearly return left after the part explained by SPY. The daily account value is rebuilt from each closed fold's trades (capital, closed P&L, open trades at the day's close), each trade's recorded fee taken off on the day it closed, and compared with SPY's close-to-close change on the same sessions. Pooled over the closed folds that have daily values; no risk-free rate.

Beta vs SPY

How much the folds' daily returns move with SPY's: 1.0 moves with the market, 0 is independent of it. Same days and same rebuild as Alpha vs SPY.

The eight gates

Also shown as: Perf, Robust, Param stability, Monte Carlo, Live OOS.

Each gate is Pass, Fail or Pending (no value yet).

A · Performance (on the out-of-sample trades):

  • Sharpe: OOS Sharpe (mean) 1.0 or more.
  • Profit factor: on all out-of-sample trades together, 1.3 or more. Pending with no losing trade.
  • Max drawdown: for each fold, the account starts at the fold's capital and moves trade by trade. The worst fold's fall below its high, in percent, must be 15% or less. This is not the backtest card's Max Drawdown, which is in dollars.
  • Sample: OOS trades, 100 or more.

B · Robustness:

  • Walk-Forward: WFE 50% or more. It is the only gate that moves the verdict; the other way is an OOS Sharpe (mean) at or under zero, which makes it Failed.
  • Live OOS: the Sharpe of the last live fold is above zero. While a fold runs it is that fold's Sharpe so far; Pending with no live fold.
  • Param stability:
    • What passes: across the closed folds' winners, how much each numeric parameter moved (standard deviation ÷ median) is 0.30 or less for every one.
    • What is skipped: a parameter whose median is 0.
    • Pending with fewer than 2 closed folds.
    • Clicking shows each parameter's values.
  • Monte Carlo:
    • The draws: 1,000 paths. Each draws, with replacement, as many days as the out-of-sample has days with a trade exit, and adds their daily returns.
    • What passes: the total at the 5th percentile (the 51st worst of the 1,000) is above 0.
    • Needs: 30 such days.
    • What it ignores: the draws are independent, so runs of bad days in a row are ignored (a finer version is pending).
    • Not on the sheet: the median and the 95th-percentile drawdown are computed but not shown.

Fold Ledger

The walk-forward fold by fold, one row per fold, newest first. Its columns:

  • Winner: the fold's winning run.
  • Code: the version, or "a → b" when one was published during the fold.
  • Drift vs prev: the parameters that changed from the previous fold.
  • OOS window.
  • Trades.
  • IS → OOS Sharpe: the winner's training Sharpe, then the out-of-sample one.
  • OOS ret %: see Return (Σ %).
  • Beta: the fold's Beta vs SPY.

Validated Performance

The curve of Return (Σ %) over time: within a fold the return grows trade by trade on the fold's capital, and the folds' totals are added one after the other. A backtest fold that ended holding a position adds its value at the fold's last point, so each fold ends on its Fold Ledger figure. Purple folds ran live.

OOS Daily Heatmap

Also shown as: OOS days, Negative days, Worst day streak, Negative months, Best day, Worst day.

One cell per day with a trade exit. A day's value is the sum of its trades' P&L, each divided by the fold's account just before it, in percent.

  • Worst day streak: the longest run of negative trade days in a row. Days without trades are skipped.
  • Negative months: months whose daily values add up below zero.
  • Monthly Summary: adds the days; it does not compound them.

Live Fidelity — Shadow

Also shown as: Match rate, P&L gap.

How closely the live matched its shadow backtest of the same days.

  • Match rate: matched trades ÷ all trades of both.
  • P&L gap: the difference between the two realized P&Ls ÷ the starting capital, without its sign.
  • The verdict: TIGHT at 90% matched or more and a gap of 1% or less, DRIFT otherwise.

A live

Total P&L

What the closed trades made: after commissions in a backtest, after the fee recorded on each trade in a live.

On a live's P&L tab, it counts every trade closed since the live's first run started ("since Mar 3"). Positions still open are not in it: Today P&L shows them.

A live records no commission on stocks and options, so those are not taken out; the Broker check says how much they were.

In By Direction and By Symbol, it is that group's closed trades.

In Backtests side by side, a run's Total P&L is the account's value at the end minus the starting capital, so a position still held at the end is in it.

Today P&L

What the trades closed today (New York date) made, after the fee recorded on each trade, as Total P&L. "open +$…" is what the positions open now are up or down since each was opened, not since this morning.

Unrealized P&L

Also shown as: Unrealized %, Market value.

What the live's open positions are up or down now, at the broker's current price. For each open position:

  • Market value: the broker's current price × quantity.
  • Unrealized P&L: the market value minus what this live paid.
  • Unrealized %: that difference as a share of the cost.

The positions are this live's own fills, never more than the account holds. When the broker gives no current price for a position, its three figures are "—", and so is any total that would include it: a total that leaves a position out is not the total.

P&L %

In the list of a live's trades, the price move ÷ the average entry price: per share or contract, not a share of the account. The P&L next to it is before the crypto fee, which has its own column.

Hold

In the list of a live's trades, how long the trade was open: from its first entry fill to its last exit fill.

Broker check

Whether Luxis and the broker agree on this live's trades, and what the broker charged that Luxis doesn't count. The card says:

  • "agree" means every Luxis trade has a broker round trip of this live's orders on the same symbol, with the same quantity and the same P&L to the cent, exiting within 15 minutes, and none is left over on either side.
  • "N trades" is Luxis's count.
  • "commissions $X, not in P&L" is what the broker charged on this live's own orders since its first run: the fees it ties to an order, such as an option's clearing fee. The crypto fee is not in it: Total P&L already takes it out.
  • "account fees $Y, whole account" is the fees the broker writes once a day for the whole account, on no order (regulatory fees on sales). Every live and every order placed by hand on the account shares them, so they can't be split by live.

Not shown yet

These cards exist in Luxis but no screen shows them today: they wait for a market series of the platform's own. The Edge Report's Alpha vs SPY and Beta vs SPY are computed another way and are shown.

Beta

How strongly the strategy's daily returns, from its closed trades, move with the market's: 1.0 moves with it, 0 is independent of it.

Alpha (ann.)

The average daily return left after the market's part, × 252, in percent; no risk-free rate.

Information Ratio

The average daily return above the market's, divided by how much that difference swings, scaled to a year. 0.5 or more is good.

Regimes Covered

How many of four market moods (calm, normal, stressed, crisis, by the market's volatility over 20 days) the window contains with at least 15 sessions each.

Trades Off-Calendar

Closed trades whose exit day has no market session in the series.