They probably knew they were close to the legal limit and their engineers probably said on a hunch "we can stay beneath the limit and retain adequate power with the right mix of tricks" and then just ended up being wrong on that hunch by a little bit.
That's why we have to bend them so often in real life with tricks. In media they can just make something up, we've got to find real solutions. I'm sure that one exists but it's either too expensive currently or not consistent enough yet. Yet being the key word there. I'm sure this will be an elegantly solved issue at some point, just might require some more smarts on the hardware side.
C(32, 3) is 4960. That means almost 5,000 different interference patterns to deal with. Since every manufactured device will vary (since each of its 32 manufactured coils will differ slightly from the next), that's 5,000 calibrations that need to be done on every single manufactured device. And that's best case, assuming (for example) that calibration is a single metric per coil over time.
The laws of physics are very unforgiving.