The present paper addresses doubtful cases concerning the use of umlaut in the adjectival comparison of contemporary German: bang 'anxious' -banger/bänger -am bangsten/bängsten. It aims to shed light on the concrete distribution of this variation, i.e. the preference for one of the variants. Corpus-based analyses will show that the adjectives under discussion are not equally affected by umlaut variation: some are (surprisingly) stable (e.g., gesund 'healthy'), whereas many others have a clear preference (i.e. > 70%) for non-umlauting forms (e.g., blass 'pale', nass 'wet'). Interestingly, a few of the supposedly stable cases appear to have at least some non-umlauting forms (e.g., krank 'ill', nah 'near', grob 'rough'). Even more interesting (but still comparatively rare) is the use of umlaut in conceptual orality contexts with adjectives that exhibit no umlaut comparison in Standard German, e.g., klar 'clear', falsch 'wrong', doof 'stupid'. As will be demonstrated, these doubtful cases reflect a centuries-old and still ongoing reorganization process within umlaut comparison. It will turn out that a complex network of interacting factors such as token frequency, phonological schemas, and morphological complexity is at work.