P&L, account equity and returns
Net P&L is recorded SPX results, net of recorded fees; other trading in the accounts is not counted. Account equity is the trading accounts’ cash plus open positions at their entry cost, the same basis the position sizing uses. It is not a marked-to-market balance, and it does include all trading. Money put in is the starting balance plus deposits.
Returns are time-weighted: each day’s results are divided by the capital in the account that day (the previous close plus same-day deposits), and the days are chained. Deposits and withdrawals therefore never count as performance, and a strategy’s return is its results measured on the whole account’s capital. Results dated before the account’s capital record begins count toward P&L but not toward returns. If the capital record is unavailable, returns fall back to a fixed starting base, and the figures say so.
Account page: all trading, money put in and two kinds of return
The account page covers the whole account. Its figures come from the account balance (cash plus open positions at cost), which includes all trading, not only the recorded SPX results; figures marked SPX ledger use recorded results instead. Money put in is the starting balance plus deposits, and gain on money put in is equity minus money put in. Date and strategy filters do not apply to this page.
The time-weighted return chains each day’s equity divided by that day’s base (the previous close plus same-day deposits). It is the growth of money kept in from the first day, so it measures the trading, not the deposits; growth of $1,000 shows the same index in dollars, and monthly returns chain it within each calendar month. The money-weighted return is the internal rate of return: the single daily rate that turns the starting balance and each deposit, from the day it went in, into today’s equity, expressed over the whole period. It measures how the money actually put in has done, so it is lower than the time-weighted return when gains came on a small early balance or deposits arrived before losses. Neither is annualized. Drawdown here is the time-weighted index’s fall from its running peak.
Each deposit’s earnings are the index’s change from the close before it went in; the amounts’ worth adds up to today’s equity. The SPX ledger time-weighted return divides each day’s recorded results by the same daily base and matches the overview. The live-vs-model comparison uses its own research index (SPX flows with other account activity removed), so its return can differ from both.
Dates and missing days
Results are grouped by ledger result date. Broker-derived trades use expiration, not entry date. Missing days are not zero-return trades. Calendar dates and range filters use date-only ledger values; a filtered period rebases cumulative P&L and drawdown to zero.
Settlement chart
Step-after curves hold the last recorded value until the next result date, then change vertically. The displayed resolution is daily, not the actual intraday settlement timestamp. The final value extends to the selected end date. Cumulative P&L starts at zero for the selected period and is drawn in the loss colour below zero. The account-equity line is the actual balance, so it also steps up on deposit days.
Provisional results
Estimated expired-position settlements remain in the ledger until broker settlement replaces them. They can change and are marked in blue throughout the report. They are not open-position unrealized returns.
Drawdown and recovery
Daily drawdown combines all results on a date before comparing with the previous daily peak. Trade sequence also includes within-day realized peaks. Neither is intraday marked-to-market risk. Percent drawdown is the time-weighted fall from the previous peak, so deposits never hide or create a drawdown.
Recovery tables always use daily-close peaks. Recorded days count only observations below peak; calendar duration runs from the prior peak (or first result) through recovery or the last recorded result. A trailing + marks an unrecovered episode.
Risk and volatility
Everything on the risk view is read from recorded result days; nothing is fitted, simulated or annualized, and none of it is a forecast. A day’s return is its combined net divided by the account’s capital that day (time-weighted, deposits excluded), or the fixed base when account data is missing. Volatility is the sample standard deviation of those daily returns over the selection or a rolling window of complete result days. Worst weeks chain the daily returns of Monday-start weeks.
Observed percentiles interpolate between ranked days: P5 is the level about 1 in 20 recorded days fell to or below. It is shown only from 20 result days, P10 from 10. With a short record the worst outcomes are barely sampled, and a future day can be worse than any shown. Concentration removes the best days from the net, a counterfactual check, not a simulation. Time under water counts result days below the prior peak of the time-weighted index.
Strategies moving together
For each pair of named strategies, the correlation of their daily net P&L uses only the days both recorded a result. It is shown from 5 shared days. The shaded band is the range a correlation reaches 95% of the time by chance alone with that many independent days (Fisher approximation); a value inside it is not evidence of any relationship. Days both lost counts shared days where each strategy lost money. Exploratory and unreviewed trades are left out.
Strategy attribution and categories
Strategy contributions are dollar P&L, ranked by total net contribution, with exploratory and unreviewed trades last. They are not normalized for capital at risk or position size, and strategy drawdowns are independent histories that do not sum to portfolio drawdown.
Each automated trade is assigned to its strategy by a private rulebook kept off this site, so strategy names and rules are not published; the labels here are anonymous. Trades placed before the automated strategies started, and any manual trades, are shown as Exploratory: they count toward total P&L but not toward any strategy. A trade that does not match exactly one strategy is shown as Needs review until it is checked. Categories never change P&L.
Strategy profiles and share of return
A profile shows one strategy’s recorded trades under its code name. Its share of return is measured in percentage points of the account’s time-weighted return: each result day, the strategy’s net divided by that day’s capital, multiplied by how much the account had grown before that day. Added across every strategy, the shares equal the account’s return exactly, so a strategy can contribute more than the whole when others lost. Win-rate ranges are 95% Wilson intervals over trades; streaks count consecutive trades in result-date and ledger order. The live-vs-model section always covers the model’s full run, whatever dates are selected.
Rolling results, concentration and break-even rate
Rolling means use complete 5, 10 or 20-observation windows; overlapping windows are dependent. Result-day windows combine same-day trades and exclude dates without a result. Same-date trade ordering is ledger ordering, not verified fill time.
The average-trade decomposition divides gross wins and signed gross losses by all selected trades, including flats. Concentration subtracts the best 1, 3 or 5 positive outcomes without replacing trades or removing losses. Payoff ratio is average win divided by absolute average loss. Break-even non-flat win rate is absolute mean loss divided by mean win plus absolute mean loss. No estimate of future edge is implied.
Intervals and percentiles
The 95% positive-day interval uses the Wilson score method. It assumes independent days with stable probability; regime clustering and serial dependence can invalidate that assumption. Percentiles linearly interpolate ranks in the observed trade sample, not in an equity time series. They are not forecasts of tail risk.
Bootstrap ranges (average trade, win rate, and each strategy’s average on the scorecard) resample the selected trades with replacement 4,000 times using a fixed seed, so they do not change on reload, and report the middle 95% of the resampled values. The band on the settling-average chart is the mean plus or minus 1.96 standard errors of the trades so far, from the fifth trade. Both assume independent trades from one unchanging distribution; streaks, regime changes and changes in sizing are not captured, and strategies under 10 trades get no range. “Trades to tell the average from zero” is the count at which that interval’s half-width would equal the observed average at today’s spread: an order of magnitude, not a target.
Outcome distribution
Histogram bars use equal-width dollar bins on a dollar axis (each bar spans its bin, ticks on the edges, lower bound included) and exclude flat results; dashed and dotted lines mark the average and median. Percentage bars divide by all selected trades, so they sum below 100% when flats exist. The empirical cumulative distribution includes flats and ties, uses actual dollar values and ends at 100%; it is a sample distribution, not a future loss probability.
Sample statistics and size
Return standard deviation divides each trade by the account’s capital on its own day (previous close plus same-day deposits) and takes the sample standard deviation of those returns, so deposits never shrink earlier trades’ returns; without account data it uses the fixed base. Standard deviation uses the sample formula, with no annualization. Weekday comparisons use result weekdays and combine same-day trades; missing days are omitted, not treated as zero. These are descriptive, not evidence of a persistent trading advantage. No Sharpe, alpha or annualized figure is inferred. Selections under 20 trades, and strategies under 10, are flagged as small samples: one trade can move their averages and rates a long way.
Period reports and comparisons
Report cards group recorded results by result date into Monday-start weeks (labelled Monday to Friday) or calendar months, in New York dates. A period that has not ended is tagged To date; its totals run through the report date. Return is the period’s time-weighted return: each day’s net divided by that day’s account capital, chained over the period’s result days, so deposits never count. Best and worst days combine same-day trades. Weeks or months without a result are listed, not skipped, and “Partial” marks a period the date filter cuts.
The comparison takes its two periods from its own pickers and applies only the strategy filter. Change is A minus B; return and rate changes are in percentage points. Max drawdown is measured from each period’s own start at daily closes. Periods of different length, or one still in progress, are not like for like: compare net per result day, or pick a “same days” period. Periods under 10 trades are flagged as small samples. Everything is descriptive, not a forecast.
Live vs model
The account’s equity is compared with backtest replays of the same strategies. The replays start a week apart through history; each begins with the account’s first balance, adds its deposits on the same days, switches each strategy on the day it started live, and uses the current sizing rules and live-like fills. Two cases are shown: edge intact (every start since May 2022) and edge gone (every start from 2018 to 2022, when most of the strategies had about no edge and roughly broke even). Before settlements post overnight, the latest day is valued at the SPX close and marked provisional.
Percentile is the share of windows below the account on that day. The likelihood ratio compares how likely that day’s balance is under each case; days overlap, so ratios are not multiplied. The alert line is the edge-intact 5th percentile; the room above it is the account minus that line, which rises with the replays and with each deposit. Strategy ranges use the same windows. Average trade percentile ranks the account’s average result per position among equal-sized samples drawn from the backtest.
The forward view restarts the same windows from today’s balance with no deposits, includes execution costs that grow with size and stops each strategy at its estimated capacity. It is a replay of history under current rules, not a forecast; the edge-intact case assumes the backtest edge continues. The checkpoint reads the same from-today bands at one date; it is measured from today’s balance, unlike the alert line, which is measured from the start with deposits. All figures come from the owner’s research engine and are refreshed after each trading day.
Refresh and data source
The page reads the shared ledger and checks for changes every minute while it is visible; the header shows when the ledger last changed. The model comparison is regenerated after each session (in the evening at the SPX close, then again once settlements post) and the page checks for a new run every ten minutes; on that view the header shows when the comparison was last run. Refreshing never logs into a broker: a separate server-side sync worker retrieves account history.