There is no trade count that turns a percentage into a definitive answer. The useful questions are how much uncertainty you can tolerate, what population those trades represent and how you chose the rule. Wilson expresses that uncertainty without treating the observed percentage as exact.
What the Wilson interval says
The interval surrounds the estimated rate with a margin that shrinks as the sample grows. Wilson also moves its centre relative to the observed percentage, especially for small samples or extreme rates. It is not simply adding and subtracting the same margin from the published percentage.
Declared · The table uses a 95 % confidence level. Under a model of independent trades with a stable win probability, the procedure would cover that probability in approximately 95 % of many repeated samples. It does not assign that probability to a future outcome or replace a review of the selection process.
How the table was calculated
Declared · Every cell is calculated when generating the article with the same Wilson function used by Rigor's figure reader: 20, 45, 100, 300 and 1,000 trades, with declared rates of 55 %, 60 % and 71 %. These are mathematical examples, not measurements of a track record.
The reader uses the declared proportion without reconstructing an integer win count. Some table combinations do not correspond to a whole number of wins: they represent illustrative or rounded percentages. When analysing a file, preserve the original counts, trades and rules used to classify them.
Table: how the interval changes with sample size
| Trades | Declared · 55% | Declared · 60% | Declared · 71% |
|---|---|---|---|
| Declared · 20 | Declared · 34.2–74.2 % | Declared · 38.7–78.1 % | Declared · 49.1–86.1 % |
| Declared · 45 | Declared · 40.6–68.6 % | Declared · 45.5–73.0 % | Declared · 56.5–82.2 % |
| Declared · 100 | Declared · 45.2–64.4 % | Declared · 50.2–69.1 % | Declared · 61.5–79.0 % |
| Declared · 300 | Declared · 49.3–60.5 % | Declared · 54.4–65.4 % | Declared · 65.6–75.8 % |
| Declared · 1000 | Declared · 51.9–58.1 % | Declared · 56.9–63.0 % | Declared · 68.1–73.7 % |
How to read a row without turning it into a target
Declared · With 45 trades at 71 %, the calculated interval is 56.5–82.2 %. With 1,000 trades at the same percentage it is 68.1–73.7 %. The interval narrows under the same assumptions; it does not demonstrate that the strategy will retain that rate when conditions change.
Choose the precision you need before accumulating more data. Checking the interval after every trade and stopping when it looks favourable introduces another selection that this table does not correct. Decide in advance when to review, which rules stay fixed and what to do if the result is inconclusive.
The best of many configurations changes the reading
Declared · Suppose you publish the best of 100 configurations tried on the same history. Selection favours percentages that received a favourable random deviation. This is a search assumption, not a measurement of how many variants were actually tried.
Holding the rate and trade count fixed, Wilson returns the same endpoints: it does not automatically shift the interval because the best variant was selected. What changes is its interpretation. Nominal coverage for a rule fixed beforehand does not simply transfer to a rule chosen after comparing results.
Declare the search, keep discarded variants and reserve data that played no part in selection. The luck calculator explores selection through Sharpe; it does not correct the Wilson interval or turn a selected percentage into independent evidence.
More trades do not resolve every uncertainty
Overlapping trades, shared signals and concentrated market periods can reduce independent information. The table does not adjust for dependence or regime changes. Splitting one position into smaller entries is not equivalent to collecting new independent observations.
Win rate also does not measure the size of wins and losses, costs, drawdowns or exposure. Examine these separately. A narrow interval can precisely describe a figure that does not answer the economic question you want to study.
Reproduce the example and preserve the file
Declared · The example link opens the reader with 45 trades, a 71 % win rate, annual Sharpe of 1.8, 3 years and 100 configurations. Sharpe, years and trial count feed the luck reading; they do not change Wilson. These are illustrative inputs, not measured data.
The reader and calculator need no account. Change the figures and observe which assumption changes the reading. To examine original trades, export the full file following your platform's guide; the CSV guide gives the common schema. Your first full report is free with an account. Measured refers to calculations from files, Declared to declarations and Not measured to missing evidence.
FAQ
Does a large sample eliminate selection bias?
No. The table describes uncertainty under its assumptions. Trade dependence, variant selection and changing markets need their own review, even when the interval is narrow.