I often run into scenarios in which the same piece of information pops up in different contexts. Let's say that--while looking over lists of drugs that cause specific toxicities (e.g. all drugs that cause hepatitis, ones that cause diarrhea, etc.)--I learn that valproic acid is one of a handful of substances that can cause liver necrosis. Since this isn't a particularly intuitive fact, I encode it in my palace containing lists of various toxicities. Later, while I'm studying valproic acid on its own, "liver toxicity" comes up as a side effect.

Since I'm now using a separate set of loci specific to valproic acid, should I double-encode this fact? That is, should I encode Val Kilmer (valproic acid) in my liver necrosis loci and Will Ferrell (liver) in my valproic acid loci?

For me personally, the answer is generally no. I find that doing this forces me to think more broadly, so my palaces act in a more coordinated, inter-connected way. Instead of just listing off the various side effects of valproic acid, I'm forced to scan across my palaces and think more critically about how valproic acid operates and the various places it appears. Of course, if you're keenly aware at every instance in which you double-encode, that problem is likewise solved. Ideally, that should be the case. You should be constantly thinking critically about how new info meshes with the material you've already learned. Personally, however, I've found there's a slight tendency to miss these connections when encoding everything afresh as it appears, so I usually avoid doing it.