My solution is to design policies that evaluate to real numbers rather than binary choices (i.e., rather than asking "did I break policy X," I ask "to what extent did I fulfill policy X"). I aim to have "policies" (though I call them "metrics") that range from 0 to 1, update in real time, and are easy to compose with other metrics.

I have a program that calculates these metrics in real time, and embed the output of that program into my desktop on my primary viewport. Since I'm usually working on my secondary viewport (a large external monitor) this allows me to have an eye on my metrics most of the time.

If I start feeling bad about the state of my metrics, I can usually do something to fix that. If I can't (if there isn't enough time in the day) I can just resign myself to having a bad day, and go back to whatever I was doing.

I prefer this because it allows me to see how fast I'm improving and how far I've improved. It also lets me 'fail' some days without much penalty.