It is built as a weighted sum of three pieces: the variance of overnight close-to-open returns, the variance of open-to-close returns, and a Rogers-Satchell range term that is drift-independent. The weighting parameter is chosen to minimise total estimator variance.
This is the most complete of the common estimators and generally the best default for equities, where a meaningful share of total variance arrives between sessions. For a typical large-cap stock, somewhere around a third of daily variance is overnight, so estimators that ignore it are structurally low.
The trade-off is that it needs clean open prices, which are the least reliable field in most data feeds due to auctions and late prints. A bad open corrupts two of the three components at once, so pair it with a data-quality-check.
Related: garman-klass-volatility, parkinson-volatility, close-to-close-volatility, data-quality-check