It really depends on everything from mood, how "in the zone" I am, how much time I have, and how good I am as it is. When I go for a high score, WR or PB, as long as it's a short run (<30 mins), I generally look at the current best run and the TAS run (assuming one exists). I make a realistic path for myself, often split it into three paths labeled "easy/safe" "realistic" and "hard/risky". Written down or in my head, most often the latter. I try to execute said path, and play. Often for a long while. Sometimes I practice individual bits or segments I feel particularly weak at, or that could easily kill a run. I also generally like to finish a run here and there, as that keeps me out of reset central.
For my 10 M run on SM for instance and some others, I went deep into the analytics. I analyzed every single jump of OVER FIVE RUNS to find out hypothetical base scores, potential improvement, percentage chances etc. I took that very seriously. Most of the time I do not make a document or spreadsheet such as that one, but I generally do analyze... a lot. After I have analyzed, found the theoretical aspect, I have to find the practical aspect. That means... I have to actually pull off parts of my path. I prefer having a path consisting of only things I can do 50% or more of the time for "realistic" paths until I am really trying to optimize. (Metro for instance, I have a few tactics where I have a higher fail ratio than 50%)
In terms of what I spend my time doing when going for high scores, most to least:
Play the game relaxedly > play the game focused > analyze path > analyze own performance > practice specific things.
I find that it works. My only real crutch is that, while I am a very mild-mannered person and very rarely get genuinely angry, I have a bit of a raging problem in games I am good at. I have a habit of getting frustrated with mistakes that I deem "easy", which sometimes helps me get motivated, but most of the time it just fucks me.