Some texts are printed lightly on dark paper/with bleed-though, for example this one: https://upload.wikimedia.org/wikipedia/commons/thumb/9/93/Hamlet%2C_Second_Quarto%2C_1603_%28Folger_STC_22278%29.djvu/page63-1024px-Hamlet%2C_Second_Quarto%2C_1603_%28Folger_STC_22278%29.djvu.jpg

This means OCR struggled if you set 50% as a threshold level, because that leaves lots of junk in the image:

convert /tmp/input.jpg -threshold 45% t45.png

50%:

45%:

The different in OCR is major: using the same trained model (eng)

50%:

Now what my Lord s proofe ha(l_) madeyouknow,

45%:

Now what my Lord is proofe hath made you know,