But to get beyond that - and getting beyond playtesting, data-collection and other brute force feedback mechanisms entails doing so - one has to systematically devise original feedback signals - pass-fail checklists, writers' rubrics, dependency graphs, workout logs and so on. Once you have the framework in hand and have validated its philosophical premise to some degree, you can feed your work and ideas through it and it automatically corrects you and guides the ideas away from major fallacies.
Of course, some amount of experimental data helps, but the huge amounts we have grown accustomed to have a habit of producing a lot of busywork - a product tuned and tweaked and prodded to improve this or that short-term KPI or audience appeal metric - without striking at the critical issues that one can probe with a bit of followthrough on self-feedback.