The problem with all these font rendering discussions is that we are missing some ultra-high quality (offline) reference rendering system to compare against. Not only does that make these discussions pretty unproductive (just subjective back and forth), but also practically that drives font designers to build fonts that look good on their preferred platform(s), rather than something that is built to look good on spec. This drives then feedback loop where other platforms then need to start emulating the major popular platforms with their flaws instead of aiming for the highest quality; for example this stem-darkening is almost certainly just inspired by macos but doesn't really have justification outside that.