I am not a statistician so cannot tell if your sample size is sufficient. In general, the closer your system is to random, the larger the sample size you need to determine utility. As an example, a coin toss that is slightly bias toward head, say 50.1/49.9 vs a random one at 50/50 needs a huge number to determine its validity. On the other hand one that is biased 80/20 will need far less, perhaps less than 50 samples. Most trading systems only provide a small advantage vs randomness and it is hard to prove. So my gut says you don't have the sample size. I know because I am struggling with the same issue.
Very much could be, only time will tell (if I don't ditch it sooner). When I create system I have a cutoff of 100 trades in each training and verification period. I assumed that's enough, but there are more metrics to make systems pass. Good thing is that there is a strong correlation between results on in-sample data and out-of-sample. Very rarely (like maybe 1 in 100) a system passes in-sample test but fails on out-of sample.
Thursday's trade: System 230904_56931_B sold cri Yesterday: System 230904_41122_B bought 85 shares bcs at 23.40
November update: Account value at 102k, so about 2% profit in 4 months. I've noticed that my systems (those created by computer) pick up a lot of rubbish. Some of the stock I own are suspended (about 15k worth) and some of those will certainly go bust. If I account for this then there is no profit. I'm considering suspending this experiment and going back to the drawing board. Especially because my other systems (created by hand) are more profitable in the last 1, 2 and 5 years. I would love to have more systems to spread my bets (mainly because of pathetic liquidity in the Polish stock market) so I will probably be back with something similar. I have a few more tests to run to save my current implementation, then I'll decide.