Perhaps some stats could even be unknown or quite inaccurate before you have run weapons through actual tests ( like too hit chance vs certain speeds for example ).
Do you mean something like the following by this?
1) The hit chance when a system is originally designed is a "nominal" value.
2) The actual value would be NominalValue*RandomMultiplier where random multiplier runs from e.g. 0.5x - 2x (uniformly on a log scale).
3) The player would originally see the nominal value in dialogs, but as the system gets exercised/tested/used in combat the value would gradually change to actual value.
4) The value would be tracked on a researched component level. So e.g. an engine might have the power and fuel consumption adjusted, and would use the same parameters in all classes in which it was used.
On the one hand, I think that's really cool and realistic from the point of view of actual systems no matching design specs (e.g. early WWII US torpedoes). OTOH it's more micromanagement. OTGH, it puts a lot more fog of war into component design - minmax players wouldn't be guaranteed that subtly tweaking a design for an extra 5% would actually give them that 5%. It would make component design more like rolling characters in D&D, with the tweak that you have to do real-life testing of the components to tell if you got a lemon and want to reroll the same design.
John