Stax (packed), 43 bytes

å3o╞╝&∞╝7►JôyG♦◄╨s│*T→╢φY'┘ò☼≤⌠░▼e╓Δ█•Aφ/│.

Try it online!

Doesn't work right now for some reason ... when I use the JS version it looks good ... maybe the TIO interpreter is an old version.

For this version, the interpreter needs to be forced to output in latin1. Stax Encoding is used. There is an extra trailing newline, but this appears to be accepted here (judging from many other answers).

Explanation:

The unpacked source is:

"r{32-m2+c+95|EVB|EB128+s+"r{32-m2+c+95|EVB|EB128+s+

• The first half simply pushes the string which is also the second half.

• The second half builds the full unpacked source out of it and packs it:

  r{32-m2+c+95|EVB|EB128+s+ Second half
  r{32-m                    Reverse and subtract 32 from each character (for packing)
        2+                  Append 2 (double quote - 32)
          c+                Concatenate with self
            95|E            Decode as a base 95 integer (for packing)
                VB|E        Encode as a base 256 interager (for packing)
                    B128+s+ Add 128 to the first byte (for packing)
                            Implicit output
  

Stax (packed), 103 bytes (50 characters)

üö╖╞╖┘û■Å╣ß$æi7⌐ê↔T)ç¢┤,I_º>┐ó♫Z╪Æ≤◄▐0σ▓☻E.α╬TŶ7É

Run and debug it

Source and output are UTF-8 here, so they are counted as such. Much longer, but UTF-8 is a little nicer to look at.

Stax (packed), 43 bytes

å3o╞╝&∞╝7►JôyG♦◄╨s│*T→╢φY'┘ò☼≤⌠░▼e╓Δ█•Aφ/│.

Try it online!

Doesn't work because for some reason latin1 does not accept the C1 control codes.

For this version, the interpreter needs to be forced to output in latin1. Stax Encoding is used. There is an extra trailing newline, but this appears to be accepted here (judging from many other answers).

Explanation:

The unpacked source is:

"r{32-m2+c+95|EVB|EB128+s+"r{32-m2+c+95|EVB|EB128+s+

• The first half simply pushes the string which is also the second half.

• The second half builds the full unpacked source out of it and packs it:

  r{32-m2+c+95|EVB|EB128+s+ Second half
  r{32-m                    Reverse and subtract 32 from each character (for packing)
        2+                  Append 2 (double quote - 32)
          c+                Concatenate with self
            95|E            Decode as a base 95 integer (for packing)
                VB|E        Encode as a base 256 interager (for packing)
                    B128+s+ Add 128 to the first byte (for packing)
                            Implicit output
  

Stax (packed), 103 bytes (50 characters)

üö╖╞╖┘û■Å╣ß$æi7⌐ê↔T)ç¢┤,I_º>┐ó♫Z╪Æ≤◄▐0σ▓☻E.α╬TŶ7É

Run and debug it

Source and output are UTF-8 here, so they are counted as such. Much longer, but UTF-8 is a little nicer to look at.

Stax (packed), 43 bytes

å3o╞╝&∞╝7►JôyG♦◄╨s│*T→╢φY'┘ò☼≤⌠░▼e╓Δ█•Aφ/│.

Try it online!

For this version, the interpreter needs to be forced to output in latin1. Stax Encoding is used. There is an extra trailing newline, but this appears to be accepted here (judging from many other answers).

Explanation:

The unpacked source is:

"r{32-m2+c+95|EVB|EB128+s+"r{32-m2+c+95|EVB|EB128+s+

• The first half simply pushes the string which is also the second half.

• The second half builds the full unpacked source out of it and packs it:

  r{32-m2+c+95|EVB|EB128+s+ Second half
  r{32-m                    Reverse and subtract 32 from each character (for packing)
        2+                  Append 2 (double quote - 32)
          c+                Concatenate with self
            95|E            Decode as a base 95 integer (for packing)
                VB|E        Encode as a base 256 interager (for packing)
                    B128+s+ Add 128 to the first byte (for packing)
                            Implicit output
  

Stax (packed), 103 bytes (50 characters)

üö╖╞╖┘û■Å╣ß$æi7⌐ê↔T)ç¢┤,I_º>┐ó♫Z╪Æ≤◄▐0σ▓☻E.α╬TŶ7É

Run and debug it

Source and output are UTF-8 here, so they are counted as such. Much longer, but UTF-8 is a little nicer to look at.

Stax (packed), 43 bytes

å3o╞╝&∞╝7►JôyG♦◄╨s│*T→╢φY'┘ò☼≤⌠░▼e╓Δ█•Aφ/│.

Try it online!

Doesn't work right now for some reason ... when I use the JS version it looks good ... maybe the TIO interpreter is an old version.

For this version, the interpreter needs to be forced to output in latin1. Stax Encoding is used. There is an extra trailing newline, but this appears to be accepted here (judging from many other answers).

Explanation:

The unpacked source is:

"r{32-m2+c+95|EVB|EB128+s+"r{32-m2+c+95|EVB|EB128+s+

• The first half simply pushes the string which is also the second half.

• The second half builds the full unpacked source out of it and packs it:

  r{32-m2+c+95|EVB|EB128+s+ Second half
  r{32-m                    Reverse and subtract 32 from each character (for packing)
        2+                  Append 2 (double quote - 32)
          c+                Concatenate with self
            95|E            Decode as a base 95 integer (for packing)
                VB|E        Encode as a base 256 interager (for packing)
                    B128+s+ Add 128 to the first byte (for packing)
                            Implicit output
  

Stax (packed), 103 bytes (50 characters)

üö╖╞╖┘û■Å╣ß$æi7⌐ê↔T)ç¢┤,I_º>┐ó♫Z╪Æ≤◄▐0σ▓☻E.α╬TŶ7É

Run and debug it

Source and output are UTF-8 here, so they are counted as such. Much longer, but UTF-8 is a little nicer to look at.