You need no a plugin, you need to figure out what is wrong with your settings. Normally optimization results should match any successive single test results.
Genetic optimization is not an approximation - this is just a method of selective enumeration of input parameters instead of brute force full enumeration.
No matter if genetic is enabled or not, a specific parameter set should always produce same results.
Unfortunately, we can't help more until you provide more details.
Once again, genetics have nothing to do with the issue. Genetic is just a matter of how enumeration of input parameters is done, not how your expert trading is performed during single test or optimization pass.
I can imagine an expert which trades using some external data (for example, by reading/writing some stuff in custom files or global variables), then it can indeed screw the optimization. Otherwise, an expert runs entirely inside environment provided by the tester, so it must show exact results for the same input parameters (if you absolutely sure that constant spread is used).
Please, provide more info, probably a screenshot with settings, optimization results, and single test results.
I have on a few occasion had Optimisations not be consistent and the solution was closing MetaTrader, deleting the cache files (in ".\tester\caches") and logs (because they can get rather large), and "all as well again in the world"!
However, I don't run Optimisations very often, so I don't know if this problem is a common occurrence or of it was just due to some other "snaffu" on my PC such as fragmented RAM or file corruption.
BTW, just a guess. Could you probably have a typo/omission or other kind of hidden bug in your source code, which makes at least one variable changing randomly inside the program (during startup or in the process)? It's so easy to overlook proper initialization, so a variable can contain garbage (at least under some circumstamces).
Try to optimize one of example expert advisers which are installed with the terminal. If this expert will show exact results during optimization and standalone tests, then the problem is most likely in your code.