I’ve tried most of the suggestions people have given here, especially the trick with intentionally entering the wrong answer. It still isn’t working for me.
One thing that has helped is to make note of what answers I give and think about why I may have given them. Often, the wrong answer is related to visually or conceptually similar item, and looking at both of them helps clarify and prevents the mistake from infecting other items.
I think this is the core of the issue for me. In addition, the way the website works right now, you are actually given negative feedback on the right answer after giving the wrong answer.
That is to say, after giving the wrong answer, the “level down” notification is shown when you give the correct answer in the same review session.
Given the current setup of Wanikani, I think the best solution is probably yours, @Radish8, intentionally drilling wrong answers after a review session using BISHBASHBOSH.
Speaking more generally about the site’s design, I think giving negative feedback at all, even when you first get the item wrong, is probably a mistake, because all it does is make the mistake more memorable. Quizzing the item again in the same session is probably also a mistake, because it’s possible for people to think they know the answer and end up reinforcing the falsely memorized answer. Rather, doing a re-review of all the missed items at the end of a session might be more effective, similar to BISHBASHBOSH, but just repeat the lessons for those items.
Additionally, since Wanikani is a memory network, missing an item might be a sign that some parts of the network are broken. For a vocabulary word, it might mean you forgot the word, because you forgot a kanji, because you forgot a radical. If you give the wrong answer but it’s a right answer for ANOTHER item, it might also mean that your memory of that OTHER item is starting to get broken or mis-associated. This is why I try to think about why I answered incorrectly and what other thing I was thinking about, but I haven’t tried digging deeper into the network for an item when I get wrong answers.
An intelligent userscript might be able to produce really good feedback on the whole network when a wrong answer is given to make this easier, perhaps creating custom review sessions based on the type of mistakes made.