I think this would be the most sensible next step in complexity for you. There’s nothing preventing you from jumping straight to a 3-digit setup, but a PAO would be a good intermediate step.
Virtual Flashcards were super-helpful and it was pretty much exclusively what I used. Number on one side, IMAGE on the other. Since the end goal is to see the number and associate an IMAGE that you can then animate in a scene, I found it more helpful to provide that visual as the “answer” instead of just seeing the word. It forced me to practice directly converting number to image quicker. Being honest with your success/failure in working with the flashcards is very important since you should be prioritizing heavier review of “difficult” number/images. Also, it’s important to practice with randomized number order. If you only drill sequentially 00-99, you may develop a dependence or habit of having to “count yourself in” to the target number. Random review keeps you on your toes and helps develop DIRECT association.
For PAO specifically, here’s a link to some thoughts on how to practice independently associating each element with it’s number: Tip for practicing PAO
So the unbreakable rule I set for myself for my 3-digit system was that the FIRST three consonant sounds of the word(s) MUST indicate the number. I was fine with trailing extra sounds, like in the “SPiDer man” example (which I’d encode to 091) Since I knew that it was a 3-digit number, I just mentally discarded the “…er man” portion. I have NO plans to create a 10,000 image 4-digit system so I am unconcerned about those extra sounds causing issues down the road.
For my 2-digit system I was more strict in not having ANY trailing consonant sounds. So 12 would be something like DaNNy, not DaNiel. DaNNy is 12, DaNieL would be 125. This strictness did make it tougher to build the list, but I’m glad I stuck to it because there is no ambiguity with what number each element represents. I really tried to future-proof my 2-digit list so that I wouldn’t need to change previously fluent associations if I wanted to expand to 3-digits. (Here is my complete 2-digit system list if you’re interested: Post your 0-99 List here for other people to use... feel free to add - #14 by TheHumanTim)
In your example you mentioned the idea that “sPiDer Man” could represent 913. In my opinion, this is NOT a good idea because you’ve broken the consistency of encoding the first three sounds. You will drive yourself crazy if you introduce exceptions to your structural rules and then have to remember which images or numbers those exceptions applied to, especially if you do this within a 3-digit, 1000 image list.
Absolutely a fixed list for anything requiring speed. Improvisation is ok when there is no concern about time. More on that below.
If you want to increase speed, then a pre-defined fixed list is generally preferred. With a single element list, this does cause a lot of repetitiveness as you’ve noticed. This is why fast times with large digit sequences are almost unheard of with only a single-digit, single element system. Too much repetition and no data compression. A 2-digit, single element system is “better” but you’ll still get tons of repetition and inefficient scene construction with lots of extra “stuff” in your scene that doesn’t encode any information.
The fix for this is expanding to something like a 2-digit PAO. With 100 unique, fixed people, 100 fixed actions, and 100 fixed objects, there are 1 MILLION combinations for scene creation. Compare that with 1000 combinations for a 1-digit PAO, 100 combinations for a 2-digit, single element system, and only 10 in a single/single.
Just comparing a 2-digit / 1 element system with a 2-digit PAO, it is much less likely that you’ll find repetitive 6-digit sequences in whatever you’re trying to memorize (with PAO) vs. multiple instances of the same 2-digit single element… Even if there is only one digit off, the PAO scenes still become unique enough that they’re pretty easy to differentiate.
Example:
127497 - DaNNy CaRRYing a PiG
127197 - DaNNy CuTTing a PiG
Those two scenes are very different, and especially if they take place at different loci within a memory palace sequence, you’ll find they’re very easy to retain because its almost like you’ve come across DaNNy twice in your “story” and in each instance he did different things.
I memorized a ton of digits of Pi using 2-digit PAAO system, so sequences of 8 digits, and I came across many that started with the same person, had the same object in the object place sometimes with the same adjective descriptor, or the same action was performed many times by a bunch of different people in those mental scenes, but it never really confused me because each one stood on its own within the context of the location.
Now, if you don’t care at all about speed, then having a pre-defined fixed word/image set is less important than having a wider major system vocabulary. You can improvise in the moment and find a nice word that matches and fits well in context of the topic you’re memorizing. The trade off is that it will take you MUCH longer to encode your numbers this way. Having a preset list allows you to save a ton of mental effort generating the elements of your scenes. The numbers do that for you.
A PAO system will generate the entire SCENE for you. “Person does Action to/with Object.” Once your list is fluent, the only brainpower you need to use is translating those prompts into a memorable visual. This is the big advantage of PAO. You can outsource the scene construction to the system and just basically read a description of the scene within the number sequence.
I know this was long. Hopefully it helped clarify some of those questions!