@jwakely: The proposed change would seem like an improvement. Is "sentry object" somehow a term of art at the moment?

It's not. Some non-code-font uses are just inconsistent/wrong, e.g. [istream.unformatted] p1 where we're specifically talking about the sentry class, so "the sentry constructor" should obviously be code font too. It's the constructor of the sentry object introduced earlier in the paragraph.

Other cases just seem like inconsistencies. [istream.extractors] p12 says "After a sentry object is constructed" and p14 says "After a sentry object is constructed". There is no difference intended here, they should be consistent. Both are referring to the actions of constructing a sentry as specified by [istream.unformatted] p1.

We could introduce wording that says "The object of type sentry is referred to as the sentry object" in two places (one for istream and one for ostream) but it seems simpler to just use code font everywhere. I checked every case in Clause 29 [input.output] and they should all be code font.

[string.io] p1 has:

Effects: Behaves as a formatted input function (29.7.4.3.1). After constructing a sentry object, if the
sentry converts to true, ...

[string.io] p6 is similar. It should be code font, but it also seems a little informal/imprecise. Converts to true spontaneously, by radioactive decay? In [istream.formatted.reqmts] we say "if the sentry object returns true, when converted to a value of type bool" which is more verbose but more precise. We don't have a name for the object so we can't say "if static_cast<bool>(cerberus) is true" instead.

Those are the only uses of "sentry" (in any font) outside Clause 29, except for a code example in [locale.general] (which is obviously already in code font).

Great, thank you! Let's do this then.