- Why call internal-radix-constructor with hardcoded -1 for null-radix when make-radix handles this via undef? If you want null-radix to be a singleton rather than a constructor you can move it after make-radix.
- (radix->vector-iter result length length) in radix->vector-iter is simply result.
- There's a lot of (vector-length vector) recomputation in vector->radix.
- The leaf-length computation of vector->radix-iter duplicates work already done in vector-copy-32, which could return the vector and length as a cons.
- radix-copy-path calls 'proc' with the output of vector-position, which has a range of [0...31], so the 'unless' test of radix-append needs another look.