I worked briefly on GDS/ARS protocol in modern times (for reservation system on Linux servers that could talk directly to the mainframe network, rather than using a middleware wrapper around your own mainframe)

The protocols are heavily documented in many ways, but we also had an on-site pair of experts on this particular mainframe network, as an information resource, and we needed them. And I still had to reverse-engineer some semantics or format from real-world protocol captures, and freeze that knowledge in unit tests.

There was one opcode that initially sounded simple. IIRC, linguistically, it turned out be closer to an eval than an echo.

This kind of work, carefully interoperating with critical legacy systems, can be more interesting and positive than serving cat pictures and running surveillance trackers in exactly the architecture memorized for a Design Interview. But if you do anything involving mainframes, and then want to go back to startups or Big Tech, I wouldn't put the toxic keyword "mainframe" on your techbro resume; use euphemisms like "global financial system" instead. Also, you should say that you "disrupted" it; though disrupting a critical system is not usually considered a positive achievement in other circles.