01

Confirmation must be observable

“It looks strong” cannot be checked consistently. A useful condition names an event: a candle closes beyond a specified boundary, price revisits that boundary and holds, or a defined swing is broken. It also names the timeframe and the deadline. Two readers following the same rule should be able to agree whether the event occurred, even if they disagree about what it means.

02

Choose evidence that matches the idea

If the idea depends on a close outside a range, an intrabar wick is not the same evidence. If it depends on a level holding on a retest, entering before the retest skips the condition. Use the least complicated observation that tests the actual claim. Adding several unrelated conditions can make a setup so rare that a few hand-picked examples appear convincing while the rule has no meaningful sample.

03

Waiting changes the trade-off

Confirmation often means a later entry, a wider distance to invalidation or no trade at all. That is the price paid for additional evidence, and it can reduce the remaining potential reward. A missed move is not automatically proof that the rule was bad; a late entry is not automatically safer. Measure the full distribution of outcomes after costs rather than judging the process from one chart.

04

Give every setup an expiry and a retry rule

A condition that appears hours or days after the original context may no longer test the same idea. Set a time or bar-count expiry. Decide whether a failed attempt can be taken again, what must reset first, and how many attempts the plan permits. Keep these rules fixed in replay so you do not turn a losing setup into a second, unrecorded strategy.

05

Worked example: make “rejection” testable

Replace “price rejected resistance” with a rule such as: price trades into the defined zone, a 15-minute candle closes back below its lower boundary, and the next two completed candles do not close above the zone. This is only a sample definition; its value must be tested on your instrument and data. Record cases where the condition never completes as “no trade,” not as missing data.