You did only 4 years of backtesting, but maybe you should've done 14? Yep thats right, strategies need to work on the ancient data as well, and good ones do. On the other hand, some of the top systems have very bad months where they look broken, and one month later they start working again. This december i've seen more automated strats (futures) fail than other months historically. Not mine, but other people's. Maybe the market is really weird for some of the bots right now. Who knows. There's many ways to judge the robustness though. You can look at where the trades are falling within the distribution of the backtest and forward test and if the mean center of gravity starts going weird, or stddev is way off, it's a different problem and a warning.
Switch it off and paper trade it for 20 days, see what it does.