CPHA2$() Weaknesses

a few months ago, I added a function called CPHA2 to my DEFY Code Library, but I'm sure there are weaknesses in it. I don't know how to solve them. See, I know it doesn't achieve the avalanche effect quite well enough. The best solution I can come up with is making CPHA3 use CPHA2 multiple times on different parts of the input, and with different seeds, then combine them in the end with another CPHA2. I believe that would give a lot more satisfactory results when it comes to the avalanche effect, but then I think it's still possible to carefully manipulate the bits to create hash collisions. That's because of me using XORSTR$ to get the processed data to a constant length. I don't know a better way though. So I'd like if you guys could test it and examine it to tell me how I could improve on it. I wont update it for a while though. I'm still holding my breath for SmileBoom to fix the gold membership.