Why LLMs struggle with COBOL

The reason why LLMs struggle with COBOL (and other mainframe langauges like JCL) is pretty interesting, one reason is in their training data

Training data is like a dictionary but for LLMs,and one of the reasons that AI struggles with COBOL is because of it
You see nobody cares about COBOL except finance companies,transportation companies,goverments and ATMs so why add more training data about it?Its weird logic as you would want more training data of everything, if you ask a LLM with a searching MCP and a documentation MCP theres a 99% chance there will be atleast one or two syntax errors,the problem is even worse with JCL

Another reason is because of us humans, most COBOL systems are propietary,only used for a specific thing and kept secret,there is hardly any open-source COBOL documentation and code

Third is syntax, LLMs cant really handle complex syntax like PICTURE, ALLOCATE and AMODE 31, and you may be saying "Oh SQL is all-caps" but no,all-caps isnt mandatory in SQL.

You may be confused and saying "What in the name of Mother Nature is COBOL and why am i saying what it is now", the reason im saying what it is now is because i wanted you to know reasoning before the question, so here is COBOL summarized:
COBOL is a programming language created by IBM and Dr Grace Hopper in 1960 and while many people say its dead, it used by almost everything finance,goverment and transport related, the bill you paid at your grocery store?Likely handled by COBOL.

I hope you learnt one or two things from this post, unfortunately i have to keep this post short so Goodbye,Alvida,Sayonara,Zai jian,Assalamu-alaikum and Adieu!