PENGU

The percentage counts comparable calibration cases that touched the target before the adverse boundary. All completed cases remain in the denominator, including timeouts. If both boundaries were first touched in the same minute, the order is unknown; the displayed frequency range keeps that ambiguity visible.

The confidence band describes sampling uncertainty and is approximate. “Time if reached” is conditional on known target-first cases; it does not promise that the target will be reached. Opening hits use the minute start (including zero minutes); other hits use minute-end upper bounds. Displayed time quantiles round upward. An adverse allowance is a scenario boundary, not a guarantee that waiting is safe.

Tick distances are exact price distances and may include fractional ticks. Reference boundary prices are conservatively rounded outward to the exchange tick grid. These are price observations, not execution or fill predictions.

A fixed, small decision tree identifies price states using past data. Probabilities and movement distributions come from a separate later calibration period. A still-later historical period checks those estimates against a frozen unconditional baseline. Maximum-horizon outcomes are kept out of earlier partitions.

The model does not use costs, P&L, account balances or quote-value activity. More historical observations, a larger adverse allowance, or a high event frequency do not by themselves establish an edge. Market conditions can change.

Historical periods in the wider source collection have been studied before. They are labelled retrospective; they are not an untouched live experiment. A fresh forward record is required before claiming live reliability.