Rather than just diving into pattern descriptions, you may find it useful to agree on a corpus of successful startups and then identify specific things they've

Rather than just diving into pattern descriptions, you may find it useful to agree on a corpus of successful startups and then identify specific things they've done well. Once you have a list of these things, you can start to cluster them together - pattern recognition! This is roughly the process I've used in building pattern languages on the past. I'd also suggest you have a good idea of what you mean by success, because this is the filter through which you'll identify what's worked well. In other words, patterns should be opinionated. And that's going to be non-trivial (but tractable) in a project with multiple contributors.