Just to keep this sadly degraded newsgroups head a bit above the murky waters of solution manuals and questionable conferences.... No response expected, just nice to tell my story to a few people, in addition to my work colleagues, that might appreciate it.
Working on a setup that controls an exothermic reactor, the temperature, amongst other things, is a bit critical. The reagents mix with water in a fairly typical CSTR, main handles being the addition rates. In order to help the temperature control, there are two water additions, hot and cold, from different parts of the plant. The system's been running, sort of, for years, the ops put parts of it in manual regularly when it goes unstable.
Typical for this sort of control configuration, which resembles spaghetti, it's not easy to see immediately where the problem is. But it looked more or less sensible, with some efforts to decouple the main interactions, and the main dependencies being dealt with by the right handles. The process guy finally sussed it - the 'hot' water is sometimes colder than the 'cold' water. There's been a suggestion to just switch the signals round when that happens, I'm digging my heels in.