Methodology

Four Alpaca paper accounts run a fixed daily cycle: after the close the desk computes signals and nets them into an order plan; at the next open it submits that plan; after that close it sweeps late fills, marks positions and snapshots account equity; then it archives everything with an internal hash chain. A separate publisher reads that archive — never the live database — and writes the public repository.

Book equity is read from the broker. Each session’s net asset value is Alpaca’s account equity taken at the after-close mark. It is not modelled or reconstructed from our own fill records.

Daily return is NAV today ÷ NAV yesterday − 1. Returns are time-weighted; with a single opening deposit and no later cash flows that reduces exactly to compounding those daily returns, so there is nothing for a money-weighted variant to disagree about.

The curve starts at funded capital. The desk’s first equity snapshot is taken after the first trading day’s close, so it already contains that day’s profit and loss — starting the curve there would silently delete the opening session. Each book is instead anchored to the broker’s own equity on the last day before it traded, with the account fully in cash.

Every metric is computed by one function in the firm’s metrics module and by nothing else — not in the publisher, and not in your browser. That is the only way “our calculation source is open” can be a fact rather than a claim.

Sharpe, Sortino and Calmar are excess of the risk-free rate. Interest on cash is not alpha. The rate is the 3-month Treasury constant-maturity yield, averaged over the window the ratio covers rather than taken as today’s print, and the exact rate used is published beside every number so it can be reproduced.

Annualised statistics are withheld until 60 sessions. Sharpe, CAGR, Calmar, volatility, maximum drawdown, VaR and win rate are suppressed below that threshold. On a handful of sessions they are not imprecise estimates, they are meaningless ones. Cumulative return and the equity curve appear from day one, because those are statements of what happened rather than estimates of anything.

These are not equally hard numbers and are never presented as though they were. Book level is exact — broker equity, broker fills. Per strategy is an attributed model: the broker nets our orders, so a single net fill is attributed back to the strategies whose intents contributed to it, pro-rata by requested size. It is internally consistent and sums to the book, but a different rule would give different per-strategy numbers from the same fills.

SPY total return — split- and dividend-adjusted — on the same dates, plus a cash line accrued at the risk-free rate.

These books are not SPY-like. They carry shorts and multi-asset legs. The benchmark answers “versus just holding the index?” and should not be read as a like-for-like comparison.

Two broker endpoints disagree slightly. Alpaca’s account equity (our published NAV) and its daily portfolio-history series do not share a timing basis; on 12 August 2026 they differed by about 17 basis points for one book. Both are broker figures. Every record publishes ours, theirs, and the difference — neither is adjusted to match the other.

Gaps are gaps. If the box was down, the series has a hole. Nothing is interpolated across it, the chart line breaks, and no value is carried forward to hide it.

Net asset value, daily returns, metrics and benchmarks are published with no lag. Orders, fills and positions are held back for 0 days.

The day count is a floor; the binding rule is stricter. A cycle’s detail is released only once that cycle has actually executed. The desk stages a plan after the close for the next open, and a stage can also sit unexecuted for days if something failed — a pure date rule would eventually publish an order plan that had never been sent.

Residual limitation, stated plainly: these books hold positions for more than one day, so lagged holdings still approximate current ones. The lag closes the window on new orders; it does not hide the portfolio.