It takes a lot of fiddling to get Schrödinger clues to work. A lot of the strategies for making them are similar to those you might use for making &lits, or getting good surfaces -- but here, you need to make excessive use of these methods to get the clue to work, not just to get it to sound nice.
Generally, my strategy is to construct from one place outwards. (Usually the start or end, but you can start from the middle and extend both ways.) I find it helpful to have an idea of how to end the clue -- or even better, several -- but not setting it in stone. At least for me, it's much harder to join up two half-clues than to finish one clue.
Your goal is often to desynchronize your clues: you don't want something to be used the same way in both parsings, because that means you have to get the same result. Here are some tricks I often use to do this:
- Vary your usage of words: the same word could be used as an indicator, as fodder, or as something to synonymize, depending on what you need. (You might use "second part of puzzle" to mean
_U_ in one reading but
S+P_ in another, playing off both meanings of "second".)
- Use abbreviations and first/last/middle letter extractions. A constraint "this next word has to start with T" is easy to satisfy. And you can often alternate which reading is more constrained this way, and that lets you avoid running into impossible situations as you construct.
- Leave options open. If you know you need an anagram indicator, you can select a letter from it in the other reading based on what you need then.
- Think about using "null words". Words like "and" and "with" can be simple concatenation indicators in one reading, but important parts of the clue in another. Or perhaps you can use a phrase like "X placed into Y", where in one reading "placed into" is an indicator and in another "placed" is fodder and "into" is an indicator. This gives the same type of wordplay in both readings, but you've got an extra word in one reading that can now be used in a different way.
- Try "bidirectional" indicators. Words like "initially" can be placed on either side of the word they're modifying, and so in your two readings they can apply to either side. This gives you a single-letter constraint for the previous/next part of your clue, and you can build off of that.
- Don't be afraid to use more words than you usually would. There's nothing wrong with having a definition that takes 5 or even 10 words, as long as it's still accurate -- and those extra words can be good places to fit wordplay that you couldn't otherwise.
You can see a few of these strategies in the clue hexomino shared in a comment:
"Sort of race" describes knight perhaps following east of hard pullers' second member (5½±½)
Here's how it breaks down:
This is roughly what my text document looked like while I was writing the clue -- I find this notation useful to see what words are doing in both parts, and how I can continue.
At the start, I used a definition for HUMAN that I probably wouldn't normally use. It's perfectly fine, but allowed me to get several letters of DANCER, and anagrammed too. I also used "sort of", which can mean either "rearrangement" or "type". (I actually only stumbled upon this start after about half an hour of experimentation, while I was working from the middle and going backwards. I ended up discarding most of my original version.)
"describes" is another word with different roles: in one it's a linking word, and in the other it's a container indicator. Looking through lists of indicators can be helpful for finding these (though it's not perfect); I also write down any potential ambiguities I notice in other cryptic clues.
"knight perhaps" has the same role in both, but it gives a different result. I got lucky finding this; if I couldn't do that, my next strategy would be to make a longer linking phrase for the HUMAN parsing, and have the DANCER clue parse it as "
[around] [last letter>]". Finding an acceptable phrase for this might be difficult though, and it could instead be worth moving the last letter indicator to the other side. (And at that point, there are two last letter indicators separated by a reverse-concatenation: it might be a good idea to flip the entire clue around so I could remove the concatenation indicator entirely, and use a bidirectional last letter indicator to get both in one shot.)
"following east of" is another example of a 'null word' (or, well, a 'null phrase'). "X following Y" and "X following east of Y" both mean the same thing: that X goes on the right side of Y. But the two different parsings allowed me to 'desynchronize' the clues: if I needed more letters, I could've used an indicator in the place where the word "hard" is, and that would allow that word to be fodder in the DANCER clue but an indicator in the HUMAN clue.
"hard" is an abbreviation: I set up the previous part of the clue so I just needed to satisfy "something that abbreviates to H and has last letter D", assuming D was common enough that I would find something. A quick search through the Big List of Questionable Abbreviations later, and I found something that I deemed acceptable.
"pullers' second member" is a pretty strained definition. It works, but it's not the first thing that would come to mind (and it almost certainly wouldn't be used in a non-Schrödinger clue). But it is accurate, and it works as a definition. (I could've also used something along the lines of "pulling [???]'s second member", if I could find a good word meaning "team" with second letter U. That would let me use "pulling" as a concatenation indicator: another 'null word' that isn't necessary in one clue, but is vital for the other.)