How every number is computed

Trades are replayed in date order to give the net shares held in each ticker on every date, and each holding is valued at its most recent close on or before that date. A ticker with no close yet published carries its previous one forward rather than dropping to zero.

The series begins at your first trade date. Price bars published before it are still loaded, because they are what seeds the carry-forward value on the first day.

Simple rather than logarithmic returns, because the figures built on them — Sharpe, volatility — are conventionally quoted on simple returns and mixing the two produces numbers that cannot be compared against anything published elsewhere.

A period opening at zero or below produces no return at all rather than an infinite one. A portfolio that was fully liquidated and later rebuilt therefore yields a shorter return series than its value series, which is the honest outcome: the percentage change from nothing is undefined, not enormous.

A single pass carries the running peak and records the largest proportional fall below it. The result is reported as a positive fraction, together with the index of the peak it fell from and the trough it fell to, so the decline can be annotated on the chart rather than merely stated.

Recovery is measured separately: the number of trading days from the trough until the series first regains the old peak. A portfolio still below that peak reports no recovery length at all — 'still underwater' and 'recovered immediately' are opposite facts and must not render the same.

The annual risk-free rate is converted to a daily one geometrically, not by dividing by 252 — over a year those differ by enough to move the ratio's second decimal, and the second decimal is the one people compare.

The standard deviation uses the sample denominator (n − 1), matching the volatility figure reported beside it. Mixing sample and population estimators across one dashboard produces two numbers that quietly disagree.

Fewer than two returns, or a series with no volatility at all, yields zero rather than an error or an infinity. There is no ratio to report, and zero is the reading that does not overstate the case.

Volatility is the sample standard deviation of daily returns scaled by the square root of 252, the conventional count of trading days in a year.

The annualized return compounds rather than averaging. It reads only the first and last values, which makes it blind to the path between them — a smooth climb and a crash followed by a full recovery produce the same figure, which is exactly why it is shown beside the drawdown rather than instead of it.

Every pair is measured over the same dates. The intersection of all holdings' price histories is taken first, and returns are computed within it, because correlating observations from different days is a silent error that produces a completely plausible-looking number.

A ticker bought last month therefore truncates the window for everything it is compared against. Any holding that still has gaps after the intersection is excluded and named, rather than quietly omitted from the matrix.

Below twenty shared observations no matrix is shown at all. A correlation drawn from a handful of days is noise, and presenting it in a colored grid would lend it an authority it has not earned.

The ledger is replayed oldest first. Buys add to shares and to basis; sells reduce both proportionally and bank the difference as realized profit. Order matters — pricing a sell against purchases that had not happened yet would produce a figure that looks fine and is wrong.

A sell larger than the recorded position, which usually means an opening buy was never imported, is accounted for only up to the shares the ledger knows about. Realizing profit on shares with no recorded cost would invent money.