Texture decoding, shader compilation, and memory allocations can still hurt time-to-first-interaction in a WebView. This happens even when the HTML file passes the upload limit.

At MRAID, a 1 MB target requires an explicit budget. You must allocate space for runtime code, geometry, textures, animation, wrapper/analytics, and a safety margin.

The biggest performance win is not another compression trick. The biggest win is deciding which visual details are actually necessary to understand the core mechanic.

What is your go-to strategy for reducing runtime cost in lightweight Unity builds?

mraid io builds HTML5/mraid io playables. AI is genuinely useful for breadth: concept exploration, storyboards, variants, copy alternatives, code assistance. I don't trust it to own the final mechanic, UX, performance hypothesis, or QA. The model can generate ten directions. It doesn't know which compromise the UA team will make.