ACES2065-1 is a linear encoding. This data isn’t suited for a limited integer format. You will have to choose a different format to output to. ACEScct can be an option but I’m not fully sure if that remains accurate enough if you have really high dynamic range scenes like from CG. Another alternative could be an intermediate in the camera native log encoding of your project’s camera.
I’m not the best at explaining it… but when you have scene linear values these can reach far above 1.0 floating point. Whenever you store floating point information into an integer format it will only store the values within the range of 0 to 1.0 to it’s available bits. 1.0 means the maximum value for that channel. Like 1023 for 10bits or 255 for 8bit. So while that part of the transformation will look correct, any value that was above 1.0 will become clipped.
In theory you could scale all the scene information down to fit the 0-1 range for your 10bit file but that means that the brightest exposure stop would posses half of the available bits and the lower you go the less bits are dedicated to it. That’s why we have log encodings, to more evenly fit the captured stops so we can have equal quality and control across the captured dynamic range in these smaller formats.