I must tell you that I still use compat, so tell me if any of these is already implemented in core.
1. I have a replacement task (replace.py + fixes.py, but as far as I remember, it is powered by textlib). I would like to exclude all the articles of countries. Now, it is a difficult task to identify countries by syntactical analysis of the text itself. Neither the word "country", nor categories give useful match. Infoboxes are the best, but still far from good. How great would be an exception based on Wikidata, which can state "this item is a country". If I also want to exclude biographies, I could expand my exception list with "this item is a person" statement.
2. Of course, if I can except countries, I also would like to require. Now we have only one kind of requirements, 'require-title'. We could have some Wikidata-based requirements, too.
3. At this point we should step over the long survived temporary solution, that 'require-title' is built among exceptions. A new 'requirements' dictionary should be invented with as many types as there are in exceptions. First step is to refactor the existing functionality, then implementing new ones.
4. Logical relations should be extended. Now exceptions exclude a page from text replacing if ANY of them matches, while reqiure-title requires ALL the conditions to match. Reqire + ANY is much more important than require + ALL; by taking out requirements from exceptions ot would by possible to implement ANY (either by keeping ALL, which is of course the better way, or instead). Combining except rules into ALL relations would also be useful sometimes.
5. And now I tell you my secret dream. I would like to create exception/requiring functions that use the title of the page. If the title occurs in a certain environment, except or incude the page into replacements, or use a function for replacement that uses title. I cannot pass the page title as a parameter. Tell me if it's only my dumbness.
6. Another dream: replaceexcept should return not only the new text, but number of replacements, too, as far as it can be defined and measured. I don't expect perfect result here. Or it may be measured outside replaceexcept. This would be good for statistical and other reasons, but more use cases should be described in the future. I think measuring effectiveness is one of them.
So this would ne the next generation of text replacements. And now I finally found the correct subject at the last sentence of my message. :-)