The seed charter would contain a finite set of normative principles as data points describing the shape of an implicit normative framework. However, the authors of the seed charter might unwillingly miss certain important parts when trying to paint the picture of their values. Those slippages might be caused by the authors not being consciously aware of some of their normative principles. Alternatively, they might be caused by the authors underestimating the importance of certain principles (e.g. by taking them as obvious to advanced cognitive systems).
In a sense, counterfactual cross-validation followed by targeted red teaming resembles interpolation of values rather than extrapolation, because the charter would be automatically extended to patch up the seed principles, rather than posit qualitatively novel ones. In contrast, memetic colonies in their vanilla form target an extrapolation mechanism. Those two themes might cross-polinate to account for this seeming flaw of deontic arrays.
That said, a sufficiently broad seed charter might still capture the vast majority of the implicit normative framework, requiring only interpolation to refine. One somewhat brittle argument here relies on human law being relatively effective in orchestrating society, while in large part relying on expanding and applying previous ideals to new situations.