GNU sort produces unexpected results when there are an uneven number of spaces between the columns of each line

Question

def  24  word
abc  2   word

I write the above data to a file called tmp and sort by the first column. As expected, the second line becomes the first:

  /tmp/sort: cat tmp
def  24  word
abc  2   word
  /tmp/sort: sort -k1,1 tmp
abc  2   word
def  24  word

But when I sort using the third column as the primary key and the first column as the secondary key, the lines are not reordered:

  /tmp/sort: sort -k3,3 -k1,1 tmp
def  24  word
abc  2   word

# ^^ WRONG

I suspect that this has something to do with the fact that there is 2 spaces between the second and third columns of the first line, and 3 spaces for the second line. Indeed, once I replace one of the 3 spaces of the second line with a non-space character, the sort works correctly:

  /tmp/sort: cat tmp
def  24  word
abc  2x  word
  /tmp/sort: sort -k3,3 -k1,1 tmp
abc  2x  word
def  24  word

# ^^ CORRECT

Does anyone have an explanation for this behavior?

EDIT: even more bizarrely, the sort works correctly when I specify -n (numeric) for the first column, even though it's not actually a number

  /tmp/sort: cat tmp
def  24  word
abc  2   word
  /tmp/sort: sort -k3,3 -k1,1 tmp
def  24  word
abc  2   word

# ^^ WRONG

  /tmp/sort: sort -k3,3 -nk1,1 tmp
abc  2   word
def  24  word

# ^^ CORRECT

Answer 1

I've figured out the solution: use -b

  /tmp/sort: sort -b -k3,3 -k1,1 tmp
abc  2   word       # expected: abc < def
def  24  word

from the man page:

       -b, --ignore-leading-blanks
              ignore leading blanks

I suspect that without -b, sort was treating the excess spaces after the space after the second field as being part of the third field. Hence, the third fields for the two lines were word and word, which are not the same. With -b, the third fields for both are identical: word, and so the first field is the determinant, as desired.