Any figure computed from a trade history, the expectancy, the win rate, the average winner, is an estimate of what the strategy does, taken from a sample of what it did. The estimate has an error, and the error shrinks with the square root of the number of trades. Four times as many trades halves it. That is a slow rate, and it is why small records say so little.
The arithmetic
Suppose a strategy's true expectancy is 0.2R per trade and the spread of individual trade results has a standard deviation of 1.2R, both ordinary figures. The standard error of the average over n trades is 1.2 divided by the square root of n. With 30 trades that is 0.22R, larger than the expectancy itself: the strategy's observed average could easily have been zero or negative by chance, or double. With 100 trades the error is 0.12R. With 400 it is 0.06R, and the expectancy is finally distinguishable from zero with some confidence.
The usual rule of thumb is that the average needs to be at least twice its standard error before it is taken seriously. For the strategy above that needs about 150 trades. For a strategy with a smaller edge or more variable results, several hundred.
What it means for records
A demo run of six weeks and 25 trades has told you nothing that a coin could not. A year of a strategy trading twice a week, about 100 trades, is the beginning of evidence. A multi-year record with several hundred trades is where the numbers start to hold. And a strategy trading once a month, however long its record in years, may never accumulate enough trades to prove itself statistically, which is a real limit of low-frequency systems and should be acknowledged rather than argued away.
Two strategies each show an expectancy of 0.3R. The first has 40 trades with a standard deviation of 1.5R: standard error 0.24R, and the true expectancy could plausibly be anywhere between minus 0.2 and plus 0.8. The second has 500 trades with the same deviation: standard error 0.07R, and the true figure is very likely between 0.15 and 0.45. The first is a hope. The second is a strategy.
Count the trades in an EA's record before reading anything else from it, and discount everything in proportion. A record of a few dozen trades supports no claim, however good it looks. A few hundred supports a cautious one. When a vendor cites months of history, convert to trades: a scalper produces hundreds in a month, a weekly system produces a handful, and the second needs years before its record means what the first's means in a season.