I write this given the topic on this question is closed but I have potential info to save others’ time.
I just spent a day diagnosing and researching why this will not work. It is actually documented in a buried doc here: https://www.nxp.com/docs/en/data-sheet/PCA9515.pdf.
The bottom line is that given the wired-nor nature of I2C NXP (and TI btw) uses a tri-level I2C signal to explicitly stop the propagation of a remote signal (pull down) from being reflected and causing a lockup. AKA Static-offset. They (NXP) require their I2C signals to be 0.4V (max) to be propagated by the 9515 but it pulls down to a (only) min of 0.47 to 0.6V any signal they propagate over the dI2C side - meaning they will not propagate one of their propagated signals. This design is the basis of their automatic direction detection while avoid lockups as far as I can tell.
If this is already common knowledge, ok… If not, hope it helps others from wasted time.
Rich