One of the most tedious tasks working on rules is finding real life examples and adding those.
Using LT as a local server, the api and an enormous text file with sentences, a PHP program writes the api output; a different one ‘flattens’ this to records, a third one transforms these into prototype examples and adds those to the grammar.xml as comment in the appropriate rule.
The amount of example is limited to 10% of the found errors, 10 when possible, not more then 100 anyway. Shortest examples get the priority, selecting different marked areas too.
It is far from perfect. There are some assumptions in it, and the way I change the xml is certainly not professional programming. But if anyone is interested, feel free to contact me.