Delete all lines which don’t have n characters before delimiter

Here's an sh script that produces the results you need.

#!/bin/sh

grep -f /path/to/patterns.txt /path/to/*_856_2017* | sort -u > /path/to/foundFiles.txt 

while read -r LINE
do
    grep -F "$LINE" /path/to/foundFiles.txt
    if [ $? -eq 1 ]
    then
        echo "$LINE" not found
    fi
done < /path/to/patterns.txt

In this script, I assume you output the results of your grep to the file found.txt, and that you store your patterns in the file /path/to/foundFiles.txt.

As you can see, the grep in the loop will produce the same contents of the file found.txt while adding "$pattern" not found for the missing ones.

I also devised a second approach to your case:

#!/bin/sh

grep -f /path/to/patterns.txt /path/to/*_856_2017* |
    sort -u > /path/to/foundFiles.txt

comm -23 /path/to/patterns.txt /path/to/foundFiles.txt |
    xargs -L 1 -I {} echo {} not found > /path/to/notFoundFiles.txt

cat /path/to/foundFiles.txt /path/to/notFoundFiles.txt > /path/to/finalList.txt

In this case, patterns.txt needs to be already sorted for comm to work.

The comm command compares the two files returning the lines present only in patterns.txt (-23 parameter), which is the list of patterns not found by grep.

Then, xargs grabs every line (-L 1) and echoes the line ({}) with " not found" appended to it. The result of xargs is redirected to the notFoundFiles.txt file.

Finally, you simply concatenate foundFiles.txt and notFoundFiles.txt into finalList.txt.

@steeldriver already explained why your attempt didn't work (should work with GNU sed, though).

But why not keep it simple? Printing all lines with only non-numeric characters is the same as dropping all lines with numeric characters:

sed '/[0-9]/d' direcciones.csv

Easier to write and easier to read, isn't it?