Audit the denominator behind a chart-pattern success rate

From a visible candidate to an executed trade

Create a candidate record when the predeclared identification features first qualify. Preserve the timestamp, instrument, data source and boundaries known then. A candidate is not yet an entry. An additional rule might require a confirmed continuation breakout. Candidates that never satisfy it remain visible in the ledger; they do not become fictitious losing trades.

Opened positions need a second record: executable entry, stop, target, maximum follow-up and costs. Event order matters. A target reached after the stop describes the subsequent price path but cannot generate profit in a position already closed. When the data cannot establish event order, label the outcome unresolved and examine the supported range instead of assigning the favorable sequence.

Different percentages can all be arithmetically correct

The fraction of all candidates producing both an entry and an eventual target touch, the eventual-touch rate among entries, and the target-before-stop rate measure different events. None should be called a win probability without its definition. A conditional denominator is not inherently misleading: entries can be the appropriate population for a strategy that trades only confirmed breakouts. The problem is hiding the selection rule or changing it after seeing outcomes.

Do not count multiple screenshots or timeframes of the same trade as independent observations. A candidate identifier linked to its trade identifier helps expose duplicates. A large row count alone does not establish a large sample of independent market events.

One hundred candidates, sixty entries and twenty-four wins

In this invented example, 100 candidates are recorded before their outcomes. Sixty satisfy the entry rule; forty never open a trade. Among the sixty entries, twenty-four reach a +2R target before a −1R stop. The other thirty-six stop first. Twelve of those stopped trades nevertheless touch the old target later. All event sequences are known in this hypothetical ledger.

For the 36/100 measure, the counted event explicitly requires both a qualified entry and a later touch of that entry's target. A nonentry does not qualify for that joint event. This says nothing about the nonentry's other subsequent price movements. R denotes a fixed monetary amount of planned price risk before costs, and every completed trade incurs a hypothetical total cost of 0.1R. Net result is 24 × 2R − 36 × 1R − 60 × 0.1R = 6R: 0.1R per actual entry, or 0.06R per original candidate. Nonentries contribute no trade profit or loss. These numbers do not validate a real trading strategy.

A high target-touch rate cannot replace the trading ledger

Quoting only the 60% eventual-touch rate can suggest the corresponding trades won just as often. Their win rate under the specified policy is 40%. The toy net result is still positive because of its assumed win and loss sizes. Conversely, a higher win rate can lose money with larger losses and costs. Audit the event definition, denominator and net ledger together before transferring historical pattern statistics to another instrument.