Point 4 is the one I'd build the launch around. For a voice matching tool, the proof people want is easy to describe: Starling learning someone else's writing and producing a draft that sounds like them. Ask five people who post on X a lot to let you run it on their old posts, then share a real post of theirs next to Starling's draft, with permission and their name on it. That's proof from someone who isn't you, and it makes a stronger first ten seconds for the 90 second demo than the setup screens. Their reactions, including the drafts they'd throw away, also tell you what to fix before a wider launch.