I played three pieces last night, and there were noticeably less eyeball jitter. It was just, I think, a matter of being easy at the keyboard.
I'm really excited to include aspects of music theory practice and training into the process, but I think it's going to be a one-step-at-a-time deal. I need to just focus on feeling easy in front of a new piece of music.
The problem is that there are lots and lots of moving pieces: the rhythm, the actual pitches, two hands, motif and repetition, dynamics, music theory and so on. Some of the 'literature' I've read approaches the problem by breaking the issues into individual efforts (i.e. practice one hand at time, just the rhythm, or arhythmically), and while there may be merit to that approach I think sight reading is a whole-greater-than-it's-parts endeavor and that while some of the parts may be DeEmphasized by specialized practice, one should try to play the music without reducing it for pedagogical effect. What I'm specifically saying here is that it's OK to, say, practice the piece REALLY SLOWLY or at tempo and error-ignorant, but that time spent practicing one hand at a time or just the rhythm and no pitches is time wasted. I think it takes longer to learn isolated skills and integrate than to simultaneously integrate and learn.
I am personally vulnerable to 'skills practicing'. I'll spend HOURS practicing some mundane skill in the hope that 'mastering the basics' will vaguely contribute to my overall playing. I've wasted time doing this in lots of places: scales, Liszt Technical Exercises (a truly non-navigable forest of pieces), guard returning in Soul Calibur II, wave dashing or shield reflecting in Smash Bros. Melee. While these atomic and repetitive practices are easy to do and uniquely educational they are dangerously dry: such activity becomes unsupportably boring after a week and a half, and that boredom risks abandonment of what was a fascinating/worthwhile task a week and a half ago.
I return to my Current Activity Darling (racquetball) for an alternative learning model. I've been fortunate to have enough friends interested in playing with me and one who does so on a very regular basis (James!) that whenever I WANT to play, I can play with somebody. When playing with an opponent, you have NO OPTION but to immerse yourself in every aspect of the game, and you have to deal with every mistake you might make, every technical detail, every possible maneuver, in real-time. There's no space to isolate variables- you have to develop each skill (hitting balls very close to the wall, judging expected distances, backhands, aiming, using your feet and setting up position etc. etc.) while in hot competition. And, with time and exposure, you just get better at these things, all at once, as a whole. But it takes a while. After fucking up a corner shot fifteen times or not getting to the middle quick enough, you sense that you're doing things wrong in specific places and adapt.
I think that the same model can be applied to sight reading, that on can work on individual weaknesses in SR best within the context of the entire piece, and not a sliced up, boring version of it. Things like reading motifs, or learning to correlate physical distances to notational ones, or proper timing and rhythm, all feel the same way at issues on the court do. But of course all this is still very early in development (PSR:2!) and time will tell how accurate the rqtbl model really is.