Thanks for spotting this. A few betas ago, I had to make changes to the routines that determine if a parameter is a number. There are were several different routines that did this, in slightly different ways and called from different features, each added over many years. These were all replaced by a single optimized routine for consistency. There were also a number of features that did not handle floating point values when they should have and others that allowed floating point values when they shouldn't have. All of these were changed to handle the correct number type. The above issue will be fixed in the next beta so that it matches the old behaviour.