How every figure on this site is computed.

What counts as a signal

A signal is a strategy’s state changing on a given instrument at the close of a bar. It is recorded with the strategy’s reference code, the resulting state, a confidence score, the bar date, and a stated horizon in bars.

A strategy holding its existing state is not a new signal. Neither is a strategy standing aside — though that absence is shown on the symbol page, because “no position” is a result rather than missing data.

How horizons are measured

The horizon is fixed at publication and stated on the signal. Outcomes are measured from the close of the signal bar to the close of the bar the horizon elapses on — not to the best price in between, and not to whenever the trade happened to look good.

This matters more than it sounds. A record measured at a freely chosen exit is not a record of anything.

How returns are computed

Outcomes are simple percentage changes in the instrument’s price over the horizon, in the direction of the signal. A short call that gains is reported positive.

Outstanding: the fee, spread and slippage assumptions applied to published outcomes have not been decided. Until they are, treat every outcome on this site as gross of costs.

Live results versus backtests

Every equity curve is drawn as two distinct series with a labelled boundary: backtested equity dashed and muted, live equity solid. They are never blended into a single line, and the date live trading began is stated on each strategy.

Backtested results are hypothetical. They benefit from hindsight, they do not represent capital that was actually at risk, and they routinely overstate what the same rules would have achieved live. Read them as a description of the rules, not as a forecast.

Why published records are never recomputed

Performance is written to dated, immutable records at the time it happens, and the site reads those records. It is not derived from live positions at the moment you load the page.

A track record that silently changes when the underlying data is reprocessed is worthless as evidence, because no reader can tell whether what they are looking at is what was true at the time.

Statistical measures

Realised volatility is the standard deviation of bar-to-bar percentage returns over the window shown. Annualised volatility scales that by the square root of 252 trading days. The Sharpe figure shown on a symbol is the mean bar return divided by its standard deviation, annualised the same way, and assumes a zero risk-free rate — it is a shape statistic, not a portfolio Sharpe.

Return distributions bin bar-to-bar returns and overlay a normal curve fitted to the same mean and standard deviation. The overlay is there to show where the instrument departs from normal, particularly in the tails.

Outstanding: exact indicator parameters — EMA lengths, the compression lookback, the sigma windows — are not yet published here.

Data sources

Price and fundamental data come from third-party providers, as does news sentiment. Coverage is not uniform: news sentiment is unavailable for currency pairs and commodities, and those modules report themselves unavailable rather than estimating a value.

Outstanding: providers are not yet named on this page.

Corrections

If a figure here is wrong, we correct it and say so — we do not quietly restate. Published signals themselves are never edited; a correction is recorded alongside the original.

See also the disclosures, which cover risk and our conflicts of interest.