This seems like something where 'cheap talk' is a serious problem. You're not even surveying or polling users, you're hearing from self-selected ones. And even

This seems like something where 'cheap talk' is a serious problem. You're not even surveying or polling users, you're hearing from self-selected ones. And even if you get a representative sample, what people say is different from what they do: users may not even know if they genuinely prefer it. Lots of design is subconscious, and there are measurable effects like additional milliseconds in loadtime where I doubt anyone would ordinarily be able to notice without sitting down with a stopwatch - it just feels worse. If it is really impossible to support more than one theme in the long run, this seems like a perfect use-case for an A/B test: randomize half your users into dark-mode and half into light-mode, and track total attrition and activity over the next few months (and not some proxy variable, switching to dark mode is too important to use some unreliable intermediate measurement).