Only alphabet values in secret string is a bit misleading.
When trying to develop my own spectre attack I keep running into cache line issues at different byte values.
It seems spectre attack is not very reliable for byte ranges from 0 to 255.
Doing a spectre attack for ascii values from A to Z and a to z is probably much easier and more reliable.
Thus I find this technique misleading, it's not very "generic".
I could be wrong about this, but you/anybody could prove me wrong, by changing the secret string to be more of a secret byte array (where the entire byte value range of 0 to 255 is used as secret values) and then trying to recover that instead.
For now I am willing to bet the spectre attack will fail under these conditions. Since some variables used will cause "false" cache line hits and thus through off results.