You can definitely increase correctness of the final product with more and more up front specs. There's an understanding/level-of-detail hurdle that I've run i

You can definitely increase correctness of the final product with more and more up front specs. There's an understanding/level-of-detail hurdle that I've run into a lot: people have a desire to over-pseudocode and under-specify in usual design decisions. If you really want to get a good spec that will result in both good estimates and avoid surprises in terms of "whoops, this won't work actually" then you need to get really detailed here. There's also a business one. Most businesses don't have quite as exact requirements as "fly to this point in space" or such. And often, the requirements iterate with user feedback. Or you try five things and see which users like the most. So the extra time delays really add up, when there's already a bunch of stuff going on behind the scenes than just the final feature that launches to the general public. If you're in this discovery phase, perfect is the enemy of "learning," and learning is what you need.