Unfortunately, the regular expression parser used by XDCR does not support negative lookahead (?!). This has come up before. The two suggestions in that thread were to enumerate all the prefixes in the “not a, b, or c” category, or to rename all the “not a, b, or c” documents so their keys start with a “d” prefix that would be easy to filter on (which would simplify the task of enumerating them).
If those solutions are not practical, you can try using the code from this StackOverflow post to generate a suitable regular expression.
In the long term, perhaps it would be nice to have an “Invert match” checkbox. I’ve filed MB-27941 as an enhancement request.
Thanks,
David