I was reminded of this idea by some scalping EAs I've seen with lists of different entries set up to be enabled or disabled according to backtesting. Not a bad idea. But ...
How hard is it to execute a strategy or two while collecting results of other strategies but not actually trading them? Teach your EA to paper-trade.
If your scalper watched results from ten or twenty entry scenarios and then put real money behind the winners it could adapt automatically to market conditions. Continually switching strategies as necessary.
Or you could apply this to a multi-pair trader and turn off loser pairs until they start to perform well again. All without risking funds during their down-times.
You could take a page from Google's book and save surrounding data during winning and losing times (indicator values, etc.) and build a rudimentary artificial intelligence system that could grow and improve over time.
This could easily start as a simple system and grow into something amazing.
Just an idea.