working out the prompt for the scoring system of Y Combinator startups The challenge with non-deterministic LLMs is to have approximately consistent scoring across three runs of the analyzing algorithm #tweetgit