This paper presents a formal analysis of recent work by Breiss et al. (2021) enriching the quantitative understanding of a well-known case of paradigm uniformity in Japanese compounds. The phenomenon in question is Voiced Velar Nasalization (VVN), a process whereby a general allophony between [g] and [ŋ] gives rise to an alternation in the initial segment of underlyingly /g/-initial second-members of compounds (hence N2s). This paper focuses specifically on formalizing the central claim, made first by Ito & Mester (1996), that there is a link between the morphological status of the N2 (free vs. bound) and the obligatoriness of the change from free [g] to bound [ŋ] in its initial segment. The paper's novel contribution is a formal analysis of the frequency effect of the N2 demonstrated by Breiss et al. (2021), which is implemented in a Maximum Entropy grammar (Smolensky 1986, Goldwater & Johnson 2003) with scaled Output-Output Faithfulness constraints, inspired by the general approach of Coetzee & Kawahara (2013). The paper also provides evidence that the frequency effect feeds Rendaku, an inter-morphemic voicing process, which gives rise to an asymmetric distribution of frequency-conditioned optionality in N2s, which is argued to be due to its interaction with the broader phonotactic grammar of Japanese.