Is it normal that Crowdin marks approximately 60% of a page as “not translated” when only small changes are made to a couple of places on the page? Maybe 15-20 words in total and 2-3 sentences.
This is really frustrating on long pages and then takes a long time when clicking the already translated terms back to their places from translation memory.
MD files have no defined (key-value) structure. Once you change anything in the strings/files, the translations will be lost as the strings are considered to be absolutely new.
Still, in case the string remains the same and only its place (x-path) was changed, you will see a 100% TM match for the strings, so they can be easily fulfilled with some suggestions:
You can use advanced workflow for your project which will automatically translate all the untranslated strings with the TM suggestions every time the file is updated: