That's an awesome solution. I will try to analyze it, Chubler please correct if I mess up along the way
Let's look at the sprintf call:

Code:

sprintf("%-*s", t-f+1, v)

The format flags of sprintf mean the following:
So (t-f+1) is the width of the string v to be printed. Now the repl() calls:

Code:

function repl(s,f,t,v)
{ return substr(s,1,f-1) sprintf("%-*s", t-f+1, v) substr(s,t+1) }

takes a string s, position "from" f, position "to" t, and what to insert "v". It will only modify the string from f to t, inserting v properly padded.

Code:

a=repl($0,96,98,"ABC")

will take the whole line ($0) and replace the characters 96-98 with "ABC".

Code:

a=repl(a,99,113,"DEF")

will take a and replace chars 99-113 with "DEF" left aligned and padded.

Code:

a=repl(a,190,198,"XYZ")

will replace chars 190-198 of a with "XYZ" left-aligned and padded as needed.
So it is modifying the variable a, one step at a time, and taking the output of previous call as input for next one. The same 3 calls could be written nested as:

Code:

 a=repl(repl(repl($0,96,98,"ABC"),99,113,"DEF"),190,198,"XYZ")

But we can all agree, that the former is much more readable, debuggable and commentable.