Why does the trade count matter more than the win rate?
The trade count decides how wide the uncertainty around every other number is, so a win rate or profit factor quoted without it has no stated precision at all.
A strategy's expectancy is an average over its trades, and the average of a small sample wanders a long way from the true value. Thirty trades from a rule with no edge will show a positive expectancy roughly half the time, and some of those runs will look excellent. The result is not lying; it is simply a sample too small to say anything.
The intuition that fails here is that a clean-looking equity curve is evidence. Over thirty trades, a random rule produces a smooth rising curve often enough that seeing one should not move your belief much. What moves it is the same shape persisting over hundreds of trades, because that is where chance runs out of room.
This is why the trade count belongs beside every headline figure. A profit factor of two on forty trades and a profit factor of one and a half on six hundred are not comparable, and the second is the far stronger result.
How many trades is enough?
Enough is the number at which the standard error of the average trade is small compared with the average itself, and for realistic forex edges that is typically several hundred trades.
The arithmetic is simple. The standard error of the mean falls with the square root of the trade count. If a strategy's trades have a standard deviation of about one R and its true edge is a tenth of an R per trade, then distinguishing that edge from zero at two standard errors needs the standard error below a twentieth of an R, which the square-root rule puts at four hundred trades. A larger edge needs fewer; a noisier strategy needs more.
Two things follow. First, the number is not a constant; it depends on how large the edge is relative to the scatter of the trades, so a strategy with small consistent wins needs fewer trades than one with occasional large ones. Second, the number is almost always larger than what a manually run test contains, which is the practical reason hand-testing over a few months of charts cannot settle the question.
The honest answer for a specific strategy comes from its own numbers: compute the standard deviation of its R-multiples, decide the smallest edge worth trading, and the square-root rule gives the count.
What inflates the apparent trade count?
The trade count is inflated by trades that are not independent: overlapping positions, several entries from the same setup, and long runs from a single regime that all reflect one condition rather than many.
Overlap is the obvious case. A rule that can hold several positions at once, all opened on the same signal in the same direction, produces trades whose outcomes are nearly identical. Five such trades count as one for the purpose of uncertainty, and a test that reports five is overstating its evidence by that factor.
Regime concentration is the quieter version. Two hundred trades from one strong trending year are two hundred samples of one market condition. They say a great deal about that condition and very little about the strategy, and a count that looks adequate can still rest on a single period. Spreading the sample across years and conditions is what makes the count mean what it appears to mean.
Correlated instruments compound both effects. Adding a second pair that moves with the first adds trades without adding much information. It is still worth doing, because the correlation is imperfect, but the effective sample grows more slowly than the raw one.
How do you get more trades without inventing them?
More trades come from more history, more instruments and a shorter timeframe, in that order of preference, because each step down that list adds trades at a rising cost in realism.
More history is the cleanest source. A decade contains regimes that a two-year window does not, so the extra trades add both count and diversity. The constraint is data quality; older history at fine resolution is harder to obtain and more likely to contain gaps, so the archive has to be checked rather than assumed.
More instruments help provided the rule is genuinely meant to work across them. A structure rule tested on thirty markets accumulates trades quickly, and the consistency across markets is a finding in its own right. A rule that only works on one pair is not helped by adding the other twenty-nine; it is exposed by them, which is also useful.
A shorter timeframe multiplies trades fastest and costs the most, because the spread is a larger fraction of each trade's range and the edge has to clear it. Dropping from hourly to five-minute bars for the sake of the count often converts a marginal strategy into a losing one, and the larger sample then proves the loss with great confidence.
What does QuantParadox report about the count?
QuantParadox states the number of trades every result rests on next to the result, grades across a decade of minute history on thirty instruments so the count can be reached honestly, and marks a result as not established rather than passing or failing it when the count is too small.
The not-established state is deliberate. A rule that fires twenty times in ten years has not failed; it has produced too little evidence to be judged, and reporting a profit factor for it would be reporting noise with two decimal places. The platform cannot manufacture trades for a rule that rarely triggers, and it does not pretend to.
Where the platform can help is in reaching the count without lowering the timeframe. The same rule can be run across the instrument set and the whole archive in one pass, which is the version of more trades that adds information rather than cost.
The count reported is the raw count. Overlap and regime concentration still have to be read by the person looking at the result, and the per-year and per-instrument breakdowns are there so that reading is possible.