I wasn’t originally going to write a post mortem for this, but the final word count is above 80,000 words, and it’s certainly been a unique project, so fuck it, what do I have a blog for if not this? Spoilers for Worth the Candle follow. Link here, in case you haven’t read The Exclusionary Principle.
History
The exclusions document was originally promised as a Patreon reward a long time ago, while Worth the Candle was still ongoing. Aerb was a huge world, and there were tons of exclusions in the background, and they seemed like one of the things that people would like to see the most. Besides that, I thought it might be fun to do a bit of a splatbook, which feels very appropriate to the spirit of Worth the Candle.
I can’t remember what the goal actually was, but it got hit, and I committed to doing the exclusions doc … and then just kind of didn’t for a long time. Some of that was because all the work had to happen between other projects, and part of it was because of some of the difficulties inherent to the project, namely the parts that made it unfun for me (more on that later).
Eventually, to get the thing to stop being a persistent obligation, I decided that it had to get done, even if it was at the expense of other things, and in order to help motivate myself and keep up progress on it, I streamed most of the writing process on Twitch. AI image generation was coming along at the time, and I wanted an excuse to play with it (more on that later), and so I decided to add in images for some of them, which eventually became all of them.
Then I sat on my hands for a bit, putting off editing for several months, mostly because of personal feelings, partly because I had other things that I was working on. Eventually, while the exclusions doc was only half edited, I decided to just start publishing them, one day at a time, again as a way to force myself to do something that I really didn’t feel like doing and to keep myself accountable in a way that I wouldn’t be if I had just said ‘it’s done when it’s done’.
And it’s all out there now, finally over for good or ill, and it’s time for reflection.
(This is actually a lie, since I’m writing this a week or so before the end of posting, but I cannot imagine that my opinions are going to majorly change, given that 99% of the work and reception of the work is in the past.)
Personal Feelings
![](https://alexanderwales.com/wp-content/uploads/2023/07/alexanderwales_the_artist_with_his_head_in_his_hands_paint-stai_1162e541-c23f-40d8-910d-18ab2d52dfda-1-1024x574.png)
I guess I should clarify what I mean by “it”, since there were a lot of things, and obviously I’m not going to write the exclusions doc twice. It actually takes some introspection in order to find some lessons here, rather than just being generally dissatisfied, and at least part of this post mortem is trying to extract some lessons from the pain, but in brief:
Holy hell am I not doing that again.
- Writing within an existing universe isn’t too bad. Writing in one with a pretty large canon and a lot to look out for is kind of bad. Writing in one that has a number of elements that aren’t designed to work with your project is just flat out bad. And trying to write in a specific format, from a specific viewpoint, which isn’t always conducive to the idea, actually just sucks. (But it sucks unevenly, see “Canon” below.)
- I tend to find creativity pretty easy, but the one thing that’s guaranteed to induce ‘creativity fatigue’ is having to draw from the same well over and over again. I think this is actually a brain chemistry thing, maybe. If you ask someone to list all 50 states, the last few are going to be painfully hard, and I think (unscientifically) that at some point the brain just says ‘You need me to name you a state? But I just did that!’ The exclusions doc thankfully switches tracks a lot, but it’s definitely a case of going to the same wells over and over. I was at least somewhat penned in by what was in Worth the Candle, which was written without the exclusions doc in mind, and I couldn’t just cut things down to their more natural size. Ask me to make up five cool magic items and I’ll say “Okay, sure!” Ask me to make five more and I’ll say “Er, can do!” Ask me to make fifty and … I mean, yeah, I can do it, but it’s going to smear my brain into a thin paste.
- I have tried my best to live a life where I can, as much as possible, not care about money. I live modestly and don’t splurge often. Unfortunately, I can’t completely ignore the issue of finances, especially for my job, which is writing. At best, this project somehow contributed to an increase in money from Patreon, which would be very difficult to calculate assuming that it was true. At worst, it represents a lot of work with no return aside from making something that I enjoyed making and getting to share it with people who enjoyed it too. That’s definitely worth something … but I could also have made something that I enjoyed making and that people would like and get actual income from it and it would be more accessible because people wouldn’t have had to read 1.6 million words of a prior work first. These 80,000 words could have been a short novel. Actually, given the extra time this project took, these 80,000 words could have been something like 120,000 words of a regular novel. I think the only way that I can possibly justify spending all the time is that I said I would, and as much as I can be, I am a man of my word. But I also can’t help feeling like this was a huge time and labor sink. (I have generally avoided this relationship with my artistic endeavors, and this has been the first time I’ve really felt it. I haven’t liked it, and wish I could deny it, but the feeling is there.)
- This is really a continuation of the above, but I think a fair estimation of the work involved is about 80 hours of writing, 40 hours of editing, and maybe 10 hours of image stuff. Ballpark, there’s actually over 40 hours of image stuff, but some of it was exploration, bashing my head against walls I should have just avoided, or playing around, or redoing work, and I won’t count that against this project. Image stuff also costs money, but I’m not including that here. These estimates are all probably on the conservative side, and at an upper bound, the whole project might have been as much as 200 hours if I factor in that not every hour is made equal and I’ve likely spent some very unproductive hours on the project.
- On the flip side, I did make some money from Twitch, which was a nice surprise, and I guess should be factored into any calculations on whether this was a sensible thing from a money standpoint. I mean, it’s still not sensible, but it should be mentioned.
- There are certain elements of the exclusions doc that I feel just didn’t work. Some of them don’t work, but work as well as they could have given the constraints, and some don’t work, but mostly for lack of time, planning, or some other element that I had power over. I’m happy with the exclusions document, overall, though I sometimes have to remind myself of that, but I’m not happy with all of it, and I think when I compare it to other projects, the lows are lower. There’s no real lesson to take away here, just ‘make better things’, and some of the badness of what I feel are the bad parts is just downstream of other things.
- I enjoyed 70% of the AI art stuff. It’s nice to have a visual to go with the words. The other 30% was frustrating, and I ended up settling for something that wasn’t what I had in mind more than I would have liked. (See “Art” below.)
So yes, not the greatest experience for me, though I think the final product is good, overall. The weakest parts are probably the parts that interact with canon the most, which is at least partly because they were meant to be stories, not rigid entries in a textbook. I’ll probably go into some examples later on. Overall good though, and entertaining, even if the appeal is unfortunately niche.
I’d do a splatbook again, or a worldbuilding document. I’ll almost certainly try to augment my writing with some AI art. There were some good lessons learned here, I think, some skills picked up in the process, some growth as a writer and creative.
But the overwhelming feeling is ‘never again’.
Scope Creep
![](https://alexanderwales.com/wp-content/uploads/2023/07/alexanderwales_a_king_surveying_his_lands_landscape_oil_paintin_681e32bb-6d06-4c51-bba8-1cfcc7caf055-1024x574.png)
Initially, the exclusions doc was going to be short and sweet, maybe a few paragraphs for each entry. In the end, the average ended up being more than 2,000 words. Initially, a few of the entries were going to have some AI-generated pictures, and in the end, all of them have one. Initially, there were going to be a handful of notes from characters scattered throughout, but in the end, most of them have one.
I genuinely do think that all these things have made the exclusions doc better: more compelling, more entertaining, more interesting. I think they also made the whole project take a lot more time. In very few cases do I feel like the extra stuff was bloat.
One of the best things I did, in spite of all the scope creep, was reigning in the scope creep. There could have been a lot more creep. I sat down and figured out all the directions that I could creep in, what extra stuff there could be, and I culled down as much as I could until it was just the creep that made the most sense, the creep that people would, in retrospect, say that they wouldn’t be able to live without. I had a bunch of exclusions that were ‘okay, if I’m feeling it, fine, I guess’, and toward the end of the project I took a knife to them. It’s good to know the lines along which you’ll eventually be making cuts.
Lesson for the future: control your creep, make sure you’re creeping in the right directions rather than just creeping for the sake of creep. I do think this applies to web serials in general as well, which have a definite tendency toward creep.
The Art
Back in fall of last year, I was working on getting some AI art for all of the exclusions, mostly using Midjourney but also Stable Diffusion and DALL-E. A lot of these were very abstract, in part because that’s what the image generators could do. This summer, as I’ve been publishing the entries one by one, I’ve basically been redoing all the old work using Midjourney v5 and generally getting far, far better results, with only a handful of entries using the older generations. I have a few takeaways:
- Current AI art generation can produce things that I definitely find moving and evocative. There’s a debate to be had about what ‘art’ means, but I think that debate is kind of useless to me. There are certain images I generated that I can take time to sit and appreciate, that make me feel things. This is great, and if I didn’t have a bunch of caveats, I would be making at least one image per chapter for everything I ever wrote.
- Current AI art generation sucks at being specific. As you add on more modifiers, it starts to lose the plot, ignoring more and more of them, and being more prone to including things you didn’t ask for. If AI art stalls out here (and I really really do not think it’ll be in remotely the same state in another year) there will still be tons of demand from people who want their art to show a very specific thing. Trying to get a very specific thing ends up with a lot of beating your head against the wall.
- Current AI art generation sucks at things that aren’t within a relatively narrow band of generic. Generic is what it’s good at. If you want to make a bunch of generic things, great! But if you want things that aren’t generic, that mix genres or show things that the AI model has only seen a few examples of, then you might be shit out of luck. The worst things are the things it can kind of do, and then you’re sitting there tweaking the prompt and endlessly regenerating in the hopes that it’ll chance into producing exactly what you want. Most of the time, this is just a frustrating waste of time.
- Current AI art generation is cheap and will likely get cheaper. I keep thinking about how goddamn expensive it would have been to have this whole thing illustrated to this level of quality by an actual artist. The budget for this project was basically nil, so even a single picture would have been out of the question. Artists are rightfully worried about losing their jobs to the machines, but a project like this is at the long tail, and actually does get some art, some of which is good.
- I really wanted everything to look as though it was painted by the same painter, and think I didn’t really succeed in that. Trying to get a consistent style is difficult, and made more difficult by the wide range of subjects I was attempting to depict. I didn’t put a huge amount of work into it, but at least it’s ninety percent oil painting, which is something.
There were a lot of specifics of this project that made it a very bad fit for AI art, or for art in general. The biggest of them is that some of the exclusions were simply built for prose rather than a painting. Many of them really have nothing to depict at all, whether that’s because they’re more about ideas, because the effect is invisible, or because the effect is in motion in a way that a picture can’t capture.
I think if the start of this project had been “I want a worldbuilding project with AI art” I would have put a lot of my focus on things that I knew the AI could actually depict with some fluency, and would have geared the worldbuilding toward visually impressive displays. These would need to be somewhat generic so that the AI could actually handle them, but I think the end result would gel more, and the vibes of the pictures would better fit the world that the pictures were trying to show.
There was an idea I had early on that I thought would have been cool, which was to have the pictures only show people when it was an enpersoned exclusion and only show objects when it was an entad exclusion, and otherwise show more abstract scenes or landscapes, wide shots that were clearly not focused on anything specific. But it ended up being hard with all of the other constraints put on the art by the things that AI art can’t do yet.
I have to give a special shoutout to Mome Rath, who was (ironically) impossible to get a good picture for. It was again a case where a very specific request just got the model confused and couldn’t be fulfilled. It’s very likely that I could have gotten a good result through inpainting, img2img, or other tools I wasn’t using, but I was really trying to not burn time on learning new tools, especially if I was going to have to use inferior models.
Overall, I would/will use AI art again, it’ll just be with a good understanding of what it can and cannot do, and a story or world that’s designed from the start to be a good match for the tools.
Fistfighting with Canon
![](https://alexanderwales.com/wp-content/uploads/2023/07/alexanderwales_a_bruised_boxer_hanging_in_during_the_last_round_987ff828-cabd-4cd2-9238-fcf43e110fea-1024x574.png)
Worth the Candle is hella long, and its canon is intentionally ‘big’ in a way that made it difficult to work with. It was designed from the start to have lots of weird magic systems that could butt up against each other, and that was in fact one of the major goals when writing the book. I wanted corner cases and interactions and all kinds of fun stuff like that. There’s a long history to the world, at least in terms of ‘lots of stuff is noted to have happened’ if not in terms of literal years.
All this made it significantly more difficult to write the exclusions document.
Overall, the worst exclusions to write, and generally speaking the exclusions with the worst executions, are those with lots of information in the story. I think there are a couple of reasons for this.
- These exclusions were much more constrained, leaving me unable to run with better ideas if I had them. Canon can be a straightjacket. There’s a saying that constraints breed creativity, but I’ve always found that to be only partially true. Sometimes constraints prevent creativity from going in the direction that it naturally wants to go.
- These exclusions were ‘spoiled’ by the book, meaning that the actual exclusion entry often didn’t have that much to offer and felt redundant, offering no tension, answering no questions, and generally just being there for the sake of completeness. I almost didn’t publish the Li’o exclusion because it felt so pointless.
- Many of these exclusions, when initially thought up, were thought up for the story, which means that I was taking something that was envisioned with a narrative purpose or arc in mind and translating it into a relatively dry academic text. The beats don’t hit right, the personal connection to characters isn’t really there, and it’s hard to write around the fact that this is supposed to be an in-universe document. (See “Format” below.)
- Somewhat separately from the actual goodness of the exclusions, some of these have been hyped for like five years, and were just never going to live up to that hype. There’s not a lot that I could have done about that. Hype is a dangerous and fickle beast. It can turn a regular ‘meh’ exclusion into a crushing disappointment.
The two that stick in my mind most are Manifest and Celestar. I think these are both great ideas whose execution would have been great in a story, but which don’t quite work within the exclusions document. Manifest needs a chance to show his personality, his cruelty, the bounds of his power. He needs to be a presence on the page. How’s he going to do that over the course of two thousand words? How are we going to pack meditations on control into that tiny space, most of which needs to be devoted to other things? Similarly, Celestar needs details that don’t make sense to include in the entry, it needs personalities to the elven archmages, disagreements on perfection, the hum of the Perfection Engine as it’s spinning up for something calamitous. It needs Juniper coming face to face with his personal feelings. The format, space constraints, viewpoint, etc. all work against what was originally sketched out for these places and people.
And because these are canonical exclusions, they must be included in the document, and can’t really be changed for the document. So they get crammed into these tiny boxes that they weren’t meant for.
Format
I think the best thing I did with regards to the format was the inclusion of notes from characters and notes from the DM. It really helps to break up the monotony of the scholarly, factual viewpoint.
That said, I like a scholarly viewpoint from time to time, though this has probably drained that battery for a bit. It’s a well that SCP draws from all the time, and there are fun things to do with clinical descriptions of the horrific or funny, irony to be drawn from juxtaposing against the tone, and all kinds of authoritative weight you can lend to crazy fantasy stuff.
Aside from the ‘vital’ stuff at the start, the format I settled on was essentially that there’d be a picture, then an opening description, then a bit of history, some features of the zone, and finally notes from characters. This mostly works, and if I had to write it over, I’m not sure that I would change it … but it’s not necessarily the best way to ‘tell’ every exclusion. The summary gives you the foreshadowing, the history gives you a tiny little story, and the features give you the ‘payoff’. Kind of.
Some entries really felt like they wanted to be something else entirely, though I only did that with a few of them, having the bulk of them be told by characters with some connection to them or handing over more of the explanation to an excited note from the DM as he plots out how an imagined adventure there would go.
There were definitely moments where I felt constrained by the format, like the idea for the exclusion was sound but the shape it was being forced into was wrong. Sometimes I was just itching to write a little short story from someone who was in the thick of it, and couldn’t really do that without just shaking off the whole conceit of the project. Some of the repetition (see “Repetition” below) was definitely caused by sticking to a defined format, but the format was what was promised.
Maybe a way to do it would be “the exclusions document, but also some of them are just flash fiction short story type things”. I think that probably would have ended up being a little worse though, since you do gain something from the consistency as well, a kind of built up goodwill and grounding.
Repetition
![](https://alexanderwales.com/wp-content/uploads/2023/07/alexanderwales_fantasy_painting_in_the_style_of_Andy_Warhol_83854f2b-d912-404a-a780-e47ab4837bf6-1024x574.png)
Sixty-seven might have been the wrong number to do. One of the big things I was worried about going in was that too many of these places would have too much conceptual overlap and ‘spoil’ each other. I think above a certain number, overlap would simply be inevitable, and while I don’t think that number is necessarily lower than sixty-seven, I do think in my specific execution of certain things, the undesired repetition was inevitable.
I really tried my best not to repeat myself, but early on when I was mapping out ‘okay, why do exclusions happen’ the list really wasn’t that long. Similarly, I had tried to make sure that any magic systems introduced within the exclusions doc would be distinct from each other, but this is very hard to do, especially since I wasn’t just trying to make them distinct from other magic systems in the doc, but also distinct from all the ones that were in Worth the Candle. Beyond that, they all need to ‘feel’ like Worth the Candle magics, like they might have sprung from the mind of a very specific teenage boy from Kansas.
So there is some repetition, and maybe if I had devoted even more time to this project there are certain exclusions that I would have axed and written from scratch — not rewritten, because it would just be a whole new entry with a different concept. I actually do like her entry, but Moljer is one of those that I kind of frown at for being too similar to others, especially given that we have other perfectionists like Pai Shep, Bowdler, Celestar, etc.
There are also a lot of Second Empire cockups, but … that’s just canon? There need to be Second Empire cockups, because that’s an important aspect of the worldbuilding and canonically true. The Second Empire pushed things out of a belief that they could transcend and a worry that if they didn’t they would perish to the things in the dark. They were, from most points of view, evil, and more evil as time went on. But with that said, I think by the fifth or sixth ‘this tragedy was caused when a Second Empire public works project’ entry, it’s a bit boring. Maybe this should have gone up in that Canon section, but this is just a bullet that I ended up biting. It’s a fact of the universe that needed to be within the pages of the exclusion doc.
Separately, but maybe related, I worry there’s a bit too much ‘and then someone pushed things too far’ going on. I did try to shift the focus off that where I could, from ‘this person went too far’ to ‘this is why this person pushed things too far’, but I’m not sure it was enough, and even if the motivations were different, there’s too often an undeniable element of hubris.
A lot of my concerns about repetition clashed with my concerns about telling good stories. Stories have story-shapes, and deviating from those story-shapes in the interests of diversity is probably not that wise, not as a consistent principle. There comes a point where you’re just subverting expectations for the sake of it. Maybe there should have been more exclusions that were simple mistakes, or accidents that happened without mortal intervention. It’s hard to say.
I think overall I’d give myself a pretty good grade on repetition, at least given the constraints I was under. There’s overlap in terms of ‘story’ beats, and a bit of conceptual overlap between exclusions, but there’s nothing that I really felt was derivative because I had a quota to fill. It’s very possible that some readers disagree with this self-assessment, but I haven’t gotten too many complaints so far.
I also don’t think there’s much to take away in terms of lessons for the future, since I don’t see myself doing a project like this again. Maybe for the sake of variety, it would be good to just figure out some matrix of elements to mix and match, but I’m not sure that would actually have helped my process.
Special Thanks
The exclusions doc took something like 80 hours to write, most of which was streamed, archived here. It took another forty hours to edit, along with writing some additional material that was written along the way, especially notes from the characters, many of which were added on the day of. I would estimate I spent ten hours on image generation, but that’s not counting all the time beating my head against the wall, time spent learning, and time spent dithering on pictures for no good reason.
I wasn’t the only one who worked on the exclusions doc though, and want to thank everyone who watched the streams and helped to make the exclusions doc better with their comments, feedback, and discussion. Many people helped with the image generation stuff back when image generation was much more shaky than it is now, and while I ended up using relatively little of it given how astoundingly far image gen has come since I started, I still appreciate the help.
Additionally, I had a private channel set up on Discord where a few people whose opinions I value and trust could give direct editorial feedback on the document as it was completed. I think this might be the most collaboration I’ve had on anything except the podcast I do with DaystarEld, and it took some getting used to, but was very much appreciated, and definitely made the doc better than it otherwise would have been.
Conclusion
![](https://alexanderwales.com/wp-content/uploads/2023/07/alexanderwales_a_popup_book_showing_a_papercraft_magic_kingdom_b07631a9-044e-4770-a9a8-6379a3f88a85-1024x574.png)
I think the final result was good, but I don’t want to do anything like that again, for a relatively narrow definition of ‘like that’. I think a lot of the challenges were caused by the fact that I was working within a huge amount of canon, trying to keep canon compliant while also making entries I felt were compelling, and this was more exhausting than anticipated. The art went well, except for the places where it didn’t, and this could have been solved by worldbuilding with the art in mind, something that was impossible for this particular project (and possibly would have made the end result worse).
I worry this reads as too sour a view of the project, but I guess I do feel a little sour about it. It was a labor of love that I wasn’t quite in love with, an obligation from what felt like the distant past. It took energy that could have been spent on other things, and maybe those other things would have earned my love a little bit more.
If you’d like to rank the exclusion zones by whatever metric you choose, bacontime has made a tier list maker here.
I hope that if you read through all the exclusions, you enjoyed them, and where you had gripes, they weren’t enough to ruin things for you.
If you have found a spelling error, please, notify us by selecting that text and pressing Ctrl+Enter.
9 thoughts on “The Exclusionary Principle: Post Mortem”
“I almost didn’t publish the Li’o exclusion because it felt so pointless.”
For what it’s worth, I felt like the Li’o chapter was probably the least “spoiled by the book” entry of all the ones we had already seen. Getting a picture of how things progressed after the heroes left and how the whole affair looked from the outside was really interesting.
“Maybe a way to do it would be “the exclusions document, but also some of them are just flash fiction short story type things”.”
The usual cheat SCP writers use for this is to disguise your sort story as an eyewitness report, transcript, mission log, or the like. I do think that would have worked really well with some of the entries. Maybe we’ll see fanfic that fills that gap a bit.
“There are also a lot of Second Empire cockups, but … that’s just canon?”
I’m not sure if this is my imagination, given I know these were written out of order (although maybe some of this is from the editing process?), but I *felt* like the ratio of “this happened because the 2E are hubristic morons” to “this happened immediately after the 2E, who were preventing it, fell” shifted markedly in the Second Empire’s favour over the course of the document.
Oh no, my formatting broke. Hopefully that’s not too hard to read.
“I think if the start of this project had been “I want a worldbuilding project with AI art” I would have put a lot of my focus on things that I knew the AI could actually depict with some fluency, and would have geared the worldbuilding toward visually impressive displays. These would need to be somewhat generic so that the AI could actually handle them, but I think the end result would gel more, and the vibes of the pictures would better fit the world that the pictures were trying to show.”
I’ve considered a project like this myself and what I decided I’d eventually try was an inverse of your approach. The above paragraph is absolutely correct if you wish to come up with concepts first, and then ask an AI to paint them. But what I’ve realized is that sometimes the AI just creates really weird, cool shit (especially if you deliberately start to experiment with weird prompts) and I decided what I’d try to do would be generate a few dozen of these unique anomalies, make sure they’re reproducible, and try to build the world around them.
I’m always surprised by how much harsher and more disappointed than I expected webserial authors are after a long project is over.. rarely comes across in the work itself.
Following this was a blast, and for all of the obstacles in the way it still definitely felt like a labor of love. It felt very much like another epilogue, in that satisfying and indulgent way. Great atmosphere in its own way too, and really imaginative and exciting at times, just getting lost in these settings and seeing how they all connect.
All that said, thank you for all of the hard work on this and I hope this doesn’t COMPLETELY discourage you from returning to the WTC universe one day, or trying other worldbuilding projects like this.
I still think a lot of the Extras ideas seemed real fun, like the speedrun!
For what it’s worth, I really liked it! One of the fun parts about canon was getting little hints dropped even from very early on about cool EZ, and then later finding out WHY those zones were a problem. But there were tons that were mentioned off-hand that we never saw, and this was a neat chance to see what they might be like.
For the most part, the art style seemed pretty consistent to me. Maybe that’s a difference of just looking at a picture and going “ooh, pretty!”, rather than having to churn through many variations of the same image that an AI has spat out…
Also, I like the idea of ‘Faces’ magic, revolving around dice or coin flips. I think if I were going to implement it, I’d make it so it worked better if you pleaded/persuaded/intimidated or tried to trick the coins/dice before throwing them (partly because that’s something people do when gambling, but mainly because of the pun where the Face is the guy in the party who didn’t dump charisma).
One way to count all states is to imagine a map and fill it from one side to the other. With the US start with the oversees (Hawaii and Alaska) and then from either west or east. I can do that with Europe and count all capitals and countries, but probably not all US states.
This doesn’t hit the same wall you talk about. Not sure if something similar would be usable in other similar repetitive tasks.
With 50 magic items, may be it would make sense to make 5 themes and have ten items around each theme. Or map them to something so that it doesn’t feel like trying to pull another slightly different rabbit from the same hat.