A percentage can hide almost everything
A percentage is a ratio, not a measure of certainty. If a pattern worked once in one attempt, its recorded success rate is 100%. That sounds perfect, yet another attempt could change the rate to 50% immediately. The headline number has not had much chance to prove that it is stable.
Always ask how many qualifying observations produced a rate and how those observations were selected. A rate based on 200 relevant examples usually gives us more to work with than the same rate based on two. It still does not guarantee what happens next.
More examples help, but they do not fix bad examples
A larger sample tends to reduce the amount a result jumps around through ordinary chance. That is useful only when the observations are relevant and the measurement is sound. Two thousand examples drawn from one unusual period can still mislead. So can a sample chosen after someone already knows which cases produced the nicest result.
Observations may also be connected. Ten companies reacting to the same market shock are not necessarily ten completely independent pieces of evidence. Sample size is important, but it belongs beside selection, timing, data quality and the question being asked.
What this changes in Tradour
Where Tradour shows a rate, the qualifying count should be close by. A strong-looking percentage is not allowed to float free of its sample, comparator, historical period and limitations. Indicator count is a research dimension, not a substitute for the number of observations.
One legacy historical cell recorded 20 displayed-direction successes from 26 observations. That is 76.9%, but it emerged from 47 populated indicator-by-strength cells, uses a non-canonical outcome and has no unseen validation. It is useful as a question for a future frozen test. It is not proof of a 77% edge.