Factors given by DoE can experimentally not be reached We are running a DoE in a CCD manner at the moment.
One factor is quite difficult to reach and to reproduce: Humidity because it has to be set by chemical reactions.
To be more concise: The factor 56 % relative humidity can only be reached with 44 % relative humidity and some +/- % to reproduce it.
How to deal with that? My suggestion is to ignore that and kind of cheat the measurements by using the value 56 % for all the values which should be 56 %. I'd do this because in my understanding of the DoE it is less the value itself but more the factor that will be used. Sure, they are connected but in real experiments you will never be able to achieve all factors without any deviation.
Any advice?
The values are given here:

A = H2 concentration
B = CO concentration
C = C2H2 concentration
D = Environment temperature
E = relative humidity

I can also ask for the actual DoE plan with all single measurement points? At the moment, I only receive the measured data on a dripping periodicity, so the first data points are not revealing the DoE. But I can ask for the plan.
 A: You are finally analyzing the results of the experiment using some regression models. The situation seems to be that for the variable humidity, you cannot set the defined factor level exactly, but when set, you can measure it exactly?
If that is right, I can see no reason not to use it, in the analysis, as a numerical variable, using the measured value.
EDIT
Thinking a bit more about this, can there be some possible problems with this solution?  Write a stylized model as
$$ Y_i = \beta_0 + \beta_1 x_i + \epsilon_i $$  where $x_i$ is the chosen value in the design, which it in practice is difficult to reach. Then the realized model is
$$ Y_i = \beta_0 + \beta_1 (x_i+\delta_i) + \epsilon_i $$
Now, if the chemical processes in the lab which makes it difficult to realize $x_i$, and in its place leads to $x_i+\delta_i$, impurities or whatever, is related to the experimental error $\epsilon_i$ (like impurities influencing both), so $\epsilon_i$  and $\delta_i$ is correlated? Then there is problems --- but since $\delta_i$ is known, this can be investigated, maybe by plotting $\delta_i$ and model residuals.
