It seems like you're just trying things at random and
not understanding any of it. Until you either have a
"gut" understanding of how BJTs behave, or can do
the analysis, this will probably not bear fruit.
I much prefer the "gut" approach and it's been fine
across a 30+ year career that's been about 50% bipolar
technology. Though I knew other designers who
much preferred a math approach (like one guy who
really knew his stuff and taught us an internal course
where he derived bipolar operation from the hydrogen
atom and basic physics, over the course of some
months, and was notorious for his 200-transistor
op amp product designs (which met some pretty
good specs, but still)). H-parameters spoke to him.
I just nodded my way through it and kept the notes.
It's often best to work backward from the outcome
desired, through the output drivers and back to the
control-point, establishing robust-to-PVTXYZ "on",
"off" or linear bias for each stage along the way.
You can make use of "switch" primitives and vcvs
sources if you are having trouble with "too many
active devices to understand" and build up from
small sections to larger. Maybe along the way your
gut will get its degree.