I had been using poke(0x5f00, 128 ..etc) to set transparency and setting it to 128 works for spr,
Then I tried the same thing for tline and it wouldn't work unless I set it to 16 or had bit5 flipped.

This test code alternates between the 2 values, 16 works for both 128 only works for spr.

t,f = 0
memset(0x2000,1,0x1000)
function _draw()
 cls"12"

 -- palette
 t+=1
 if t%20==0 then
  poke(0x5f00,f and 16 or 128)
  f = not f
 end

 -- tline
 for i = 31,95 do
  tline(47,i,111,i,0,i/8)
 end

 -- spr
 spr(0,16,60)

end

Is this intentional? A bug? What's the difference between those higher-bits? I was confused when I stumbled into this.

It's probably a misfeature.

Only @zep would know, obviously, but as far as I know, it's only supposed to be bit 4, but I wouldn't be surprised if there's rasterization code that simply checks to see if the unsigned byte value is >15 as an optimization vs. masking off bit 4 and testing it.

Or, since you said something about using bit 5 as well, perhaps there's some broken code that does the test just plain wrong, or bit 5 is meant to do something special in some circumstance we haven't discovered yet, since zep likes to hide functionality.

Regardless, you're right to call it out as inconsistent.