In a concrete example, the patent application describes a particular method and, as part of the enabling disclosure, reports certain detector readings as confirmation of the expected outcome i.e. confirmation that the method works.
However, one year later, the inventors publish a journal paper in which they summarize this work for a scientific audience. Here, the inventors describe the results as ambiguous and uncertain. They describe the same data referenced in the patent disclosure as indistinguishable from noise and call into question whether the expected outcome actually occurred or not. The inventors say they are not sure what was going on - the effect could have been there, or not. In any case, they have no confidence to argue either way based on the instrumentation used at the time.
So, at this point, nobody actually knows whether the invention works or not as described in the patent application (and the experiment is too elaborate that anyone would try or has since tried again without confidence that it does work). Again: maybe the effect does occur based on the proposed setup, if only a better detector were used - but there is just no way of telling based on the public information to date.
I would think this questions the requirement of enablement "without undue experimentation."
Does such a patent have a chance of surviving or can it be easily litigated? Would it survive if it could later be shown that the effect was in fact present even if it was drowned out by noise at the time and in the described embodiment? If the latter were the case, then the patent author would have made "the right guess" at the time I suppose.