Multiple passes

Based on more information about the problem and its structure, I would suggest the following steps:

• Split each line by two, right before the second pattern.
• Take the part you want from each line.
• Combine the lines so that the matches are on the original line.

This means something like this:

• Replace ^(\d*\s*)(\d*\s*)$
  
  with $1\r\n$2
  
  . Just leave \r
  
  if you are not at the windows and I doubt it. Perhaps you should consider adding a macro to the end of the line. It should be something that is not included in the rest of the document (for example #
  
  ). $1
  
  means to replace the first captured group (stuff inside the brackets). So replace it with $1#\r\n$2
  
  .
• Now take the desired length of each line: (^.{n}).*(#?)
  
  and replace it with $1$2
  
  . This will grab the first characters n
  
  and insert a macro if found.
• Remove macros after macros #\r\n
  
  . Delete them or replace them with \0
  
  .

Notes

• First, you need to filter the rows that match first (^\d*\s*)
  
  .
• If you want another macro, enter occurrences #
  
  in the answer above. It shouldn't appear in the rest of the file, at least not at the end of the line.
• This answer uses backlinks which should be no problem .

Single pass

One pass may be possible here.

^(\d[\d\s]{n-1})[^\d]*(\d[\d\s]{n-1}).*$

      

        
        
        
      

    

Matches these lines, if you extract one and two groups, this will filter the desired result from the file. Just replace it with $1$2

.