I've hit a bit of a snag on my RTK studies. I guess I didn't fully grasp the core concept of how to make stories. At first I was making simple play-on-words, or just a few hint words, which was fine for the first few hundred, but then the kanji started to get more complicated, and it just wasn't cutting it any more (I was still visualizing something in my head, but nothing super vivid). As a result I started writing more imagery, which gradually got more and more vivid as I got better at it. This was helpful to remember how to write the kanji, however remembering the scene  based on the keyword became a struggle. Especially at 1000 kanji where a lot of the keywords are basically just synonyms of each other.
I asked a question on Matt Vs Japan's Discord (- a patreon perk -) asking what other people do when they struggle to remember their stories based on the keyword. I had my cards setup to show just the keyword on the front, but I also had my story under a hint. My theory was if I couldn't remember the story based on the keyword, I could reveal the hint and still practice trying to visualize the kanji before looking at it. My question was if I should mark the card as hard when I have to reveal the hint (or good if it was like "duh, I knew that", or fail if I was way off). Matt is a guy of few words (or at least he is on the discord) and basically said that hints are worthless and if I can't remember the story than I failed. I only had one or two other commenters and their advice was simply to make sure my stories are as vivid as possible (which was nice, but not really an answer to my dilemma).
So... not much to go on there but Matt is the guru, and as humbling as it was I took what he said to heart.
I did a quick RTFM and reread the book (not all of it - just the intro and section headings). It turns out that I missed something pretty fundamental about constructing stories. I can't help but put part of the blame on the format of the book though. It's broken down into several sections, and each section starts with a few paragraphs explaining the next set of kanji, BUT the headings also contain little bits of piecemeal information on how you should be constructing stories, with gotchas and pitfalls to be sensitive to. I get where Heisig was coming from when he did it this way - he tries to hold your hand through the process while gradually introducing the method. However for me it would have been more helpful to lay it all out in the beginning and let me go to town from day one. To be fair the process is explained fairly well in the introduction, but the gotchas and pitfalls are mostly in the paragraphs above each section.
The concept I missed
Heisig argues that visual memory and imaginative memory are different and that imaginative memory is stronger. He uses the example of a dream, which is visual but really comes from imaginative memory. He claims this is why dreams are powerful and remember-able.
So a "story" should be vivid and imaginative, but there's one missing component:
"...most students of the Japanese writing system do something similar from time to time, devising their own mnemonic aids but never developing an organized approach to their use. At the same time, most of them would be embarrassed at the academic silliness of their own secret devices, feeling somehow that there is no way to refine the ridiculous ways their mind works" 
The above excerpt is the part of the book that stood out to me and made everything click. It's even in the introduction - go figure.
A good story has to be strongly related to the keyword. For me play-on-words and mnemonics are the best approach for that. I abandoned that when I started making more vivid stories scenes (because that's what I thought I as supposed to be doing). My scenes would still be related to the keyword, but mostly in a visual sense, and usually not very well.
So the components are: vivid, imaginative, relate-able, and strongly connected to the keyword with play-on-words and/or a mnemonic device. I had all the components figured out at different times during the processes, but I hadn't applied them consistently, or necessarily together.
Here's an example of a good scene for me, but a poor link to the keyword:
My story: An evil baby with long claws frothing around in the water. Floating in the water.
It has a great visual for me that makes writing the kanji easy. However I've hit this keyword a few times and thought to myself "crap, what was it that floats?" I need a mnemonic or play-on-words to connect this scene of a floating evil baby to the keyword floating.
To wrap up, I'm now at ~1000 kanji "learned" but a large stack of them have poor stories that need to be cleaned up. For a large portion of my cards I need to add a play-on-words or mnemonic device to connect the scene to the keyword, or add a scene to the mnemonic/play-on-words, and in some cases I just need to start from scratch. For the past 2 days I've barely been getting through my reps (yesterday I didn't even get that far) because I've been failing most of my reviews and writing better stories, let alone adding new cards. My plan is to be really strict with myself for the next little while (how ever long it takes), to make sure that I have strong stories for all of my cards. Over the weekend I may even just go through each card one-by-one and update the stories before doing any more reviews. I also need time for these fixed, or new, stories to mature in my brain. There's no point in adding new cards until I've stabilized what I have.
 A scene to me is different than a story. Being able to paint a vivid picture is one thing, but relating it to the keyword is another.
 Heisig, J. W. (1977) Remembering the Kanji vol. I japan publications trading co., ltd.