Wow, that is a shocker!
I'm wondering if maybe when you get back in good #s you should hold the dose for longer before reducing. When I look back at the last round of 2.6u and he was getting good results, I'm thinking maybe the 2.4u dose reduction was premature (not meaning to criticize! just brainstorming). With Bix, I found if he was zoomy and I reduced it was a disaster. Sometimes it was just a few more cycles before it was crystal-clear a reduction was needed, but somehow waiting made all the difference. I guess because the zoom proved it was reduced, vs. me hoping it was reduced. Not sure if that is the case with Blackie, but could be.
[ETA: Looking at his last round at 2.6, my instinct is not to reduce unless there is a nadir < 50, or a PS a little lower, maybe <130 or something. I'd count the PS at 147 or so as ideal, suggesting it was the right dose, rather than too high. That is just my non-expert 2c though, and I know you are WAY more familiar with his #s and patterns than I am. Disclaimer, disclaimer, disclaimer.

]
Hope things get steadier soon. If it were me, I'd probably be toying with trying a different insulin too. Seems like it has been a lot of stuggle for you lately on PZI. Not to say that it can't work out successfully, but at some point the convenience factor has to weigh in, if it's a constant struggle to keep the #s good.
There's definitely a benefit in my mind to being able to shoot on lower PSs without having to worry so much about the drop. I think sometimes that's a block for PZI. I know with Bix we are in this place where he's a little high, but not high enough to feel comfortable shooting - sometimes I think with a different insulin it might be easier. But he's doing well enough that I'm not going that my route myself unless & until my PZI vial runs dry, but if he were in Blackie's shoes...? Hard to know what to do.
I have found it helps to set some timeframes - if I can't get this working in, for example, 6 weeks (whether it's the insulin, or user error), I will move on. That approach has helped me take some of the agony out of trying to decide what's best, and just put it in a results framework instead.