I'm rusty but doesn't the increase depend differently, the second adjust lets you decide to assign a value for the variable increase much later, whereas the first adjust fixes the value of increase at the point of evaluating that very assignment statement. There's a name for this, dynamic scoping or something.
Yes, my argument relies on immutability of all values (including `increase`). Also, under lazy evaluation we can call `adjust2` immediately, since we know it's a `lambda`; yet attempting to call `adjust1` will force evaluation of `increase`; not good if `increase = provable(collatz_conjecture)`!