We were getting off-by-one for the total amount that the change is for, since it rounds the fee *down*, independent of the total weight of the entire tx. We fix this by using the diff btw the fee of the total weight (w/ and w/o the change output)
param_psbt