Jerry’s Marginalia — The Bus Factor of One
There is a sentence that should scare the hell out of anyone interested in fan translations, game preservation, or ROM hacking:
One extraordinarily skilled person leaves → multiple projects lose the technical expertise required to continue.
Not because that person did anything wrong.
Quite the opposite.
The problem is that we needed them that badly.
When the Wizard Goes Home
Fan translation has always been held together with duct tape, hex editors, caffeine, spite, and a handful of people who somehow understand what a Japanese programmer was doing with assembly code in 1997.
Somebody translates the script.
Somebody edits it.
Somebody tests it.
And somewhere in the basement is The Wizard.
The Wizard knows why changing one character makes the game crash.
The Wizard knows where the pointers are.
The Wizard understands the compression.
The Wizard knows why English text won't fit.
The Wizard knows how the font renderer works.
The Wizard looks at a wall of assembly language and somehow sees words.
Then The Wizard retires.
And everybody looks at each other.
Well, shit.
That isn't hypothetical.
In April 2026, Iwakura Productions announced an uncertain future for several projects after longtime ROM hacker EsperKnight retired indefinitely. Projects including Devil Summoner and Shadowrun were left without the technical expertise necessary to continue as they had before. Iwakura chose to release incomplete work in hopes that somebody else might eventually pick it up.
That's admirable.
But think about what the situation itself tells us.
One person leaves.
Multiple games stop moving.
We call that the bus factor: how many people can disappear from a project before the project is screwed?
In this case?
One.
That's not a criticism of EsperKnight.
It's a criticism of a system that accidentally became dependent upon EsperKnight.
Then Somebody Kicked the Door In
Here's where things get uncomfortable.
Because we need to talk about AI.
Yes.
I can hear the groaning already.
AI translations are soulless.
AI produces garbage.
AI doesn't understand context.
AI is killing translators.
AI is stealing.
AI is going to eat your dog, marry your mother, and replace the batteries in your smoke detector at 3 AM.
We've heard it.
Some of those concerns deserve serious discussion. Machine-generated prose absolutely can be garbage. Translation without cultural understanding can butcher characterization, jokes, tone, and context. An unchecked language model dumping English into a script isn't suddenly a localization team because somebody slapped an .xdelta file onto it.
But while everybody argues about whether AI is good or evil, something considerably more interesting is happening.
Games are getting translated that weren't getting translated before.
Look at Segagaga.
For years, that Dreamcast oddity was one of those games Western Sega fans talked about more than they actually played. It was intensely Japanese, text-heavy, technically troublesome, and never officially localized.
Years ago, I had my own stupid little idea for dealing with it.
Forget patching the actual game.
Translate a playthrough.
Hardcode English subtitles onto the video.
At least then English-speaking players could experience the damn thing.
That wasn't elegant.
But it was achievable.
The idea essentially admitted defeat:
We can't put English into Segagaga, so we'll put English on top of Segagaga.
Fast-forward to 2026.
There is now an actual English translation patch.
And not merely a forgotten v1.0 somebody threw onto the Internet before disappearing.
As I write this, Segagaga English Translation Patch v1.1 exists.
The release fixes text, fixes a text overrun, makes debug mode accessible, supports different Dreamcast image formats, and provides patching options beyond Windows.
There was already a v1.0.2.
Now there's v1.1.
Someone is maintaining the bastard.
That's important.
Because if AI were truly the magical button that both its loudest evangelists and loudest critics occasionally pretend it is, there wouldn't need to be patches after the patch.
You wouldn't have text overruns.
You wouldn't have corrections.
You wouldn't need playtesting.
You wouldn't need somebody deciding that a line sounds like shit and rewriting it.
Instead, we're seeing something much less magical and much more useful:
A person using another tool.
The Bottleneck Wasn't Always Japanese
This gets forgotten constantly.
People hear “fan translation” and assume the difficult part is:
Japanese → English.
Sometimes it is.
But you can have a perfectly translated script sitting on your computer and still have no playable translation.
Because now somebody has to convince the game to display it.
Congratulations.
Welcome to hell.
Japanese games were frequently designed around Japanese text. English brings different character widths, longer sentences, spaces, word wrapping, and all kinds of other inconveniences the original programmers never needed to accommodate.
Want a variable-width font?
Great.
Find the text renderer.
Figure out what it does.
Modify it.
Don't break anything.
Now discover that a menu uses a completely different renderer.
FAAAAAAAK!
And that's before compression, pointers, executable formats, control codes, graphics containing baked-in Japanese text, subtitles, timing, memory limitations, and whatever other fresh nightmare the game has prepared for you.
That's why projects could sit around for ten years despite having people perfectly willing to translate them.
The problem wasn't:
“Nobody wants this.”
It was sometimes:
“The person who knows how to make this technically possible is busy.”
Or worse:
“Nobody knows how.”
Shadowrun Gets Back Up
Then we get the Sega CD version of Shadowrun.
Another Japan-exclusive game.
Another project with history.
Another translation people wanted.
And now, in 2026, an English patch exists with AI involved in its development.
That doesn't automatically make the translation wonderful.
It doesn't automatically make every technical decision correct.
It doesn't mean every future fan translation should fire its translators and ask Claude what Cloud Strife is talking about.
It means something much more consequential:
Someone finished it.
A project that previously depended upon scarce specialized labor found another path forward.
That should make preservationists pay attention.
Not worship.
Not panic.
Pay attention.
What Happens When Said Person Is Gone?
That's the question I keep coming back to.
Suppose your entire workflow requires one extraordinarily talented person.
They're fantastic.
They've helped release game after game.
Everybody knows their name.
Everybody knows that when some horrifying technical problem appears, you send it to them.
That's wonderful.
Until they decide they'd rather spend Saturday afternoon doing literally anything else.
They're allowed to leave.
They're allowed to get married.
Have children.
Change careers.
Become interested in another hobby.
Get tired.
Or simply say:
“I've been doing unpaid ROM hacking for twenty years. I'm done.”
Now what?
If your answer is:
“We wait until another genius appears,”
you don't have a preservation strategy.
You have a summoning ritual.
The Goal Shouldn't Be Replacing The Wizard
And this is where I think both sides of the AI screaming match frequently miss the interesting question.
I don't particularly want AI to replace The Wizard.
I want five more people capable of doing some of what The Wizard could do.
That's different.
Imagine an expert ROM hacker can comfortably solve ten categories of problems.
The average enthusiast can solve three.
Then better documentation, debuggers, decompilers, open-source tooling, and AI-assisted programming allow that enthusiast to solve six.
We haven't replaced the expert.
We've multiplied the useful people.
Now The Wizard doesn't need to solve every pointer problem.
The Wizard doesn't need to explain the same assembly pattern fourteen times.
The Wizard can attack the truly disgusting problems while everybody else handles increasingly sophisticated work themselves.
Better still:
They learn.
That's where I want this technology to go.
Not:
Human knowledge → AI → nobody learns anything.
Because then we've accomplished something spectacularly stupid.
We've replaced:
“We can't continue because EsperKnight isn't available.”
with:
“We can't continue because Claude isn't available.”
Congratulations.
Same cactus.
Different pot.
Open the Damn Toolbox
This is also why open-source fan-translation work matters.
If somebody figures out how to modify Segagaga, I don't merely want Segagaga in English.
I want the next lunatic to be able to ask:
How did they do that?
What did they change?
How did they handle the text?
What broke?
How did they fix it?
Can any of those tools be reused?
Can those techniques help another Dreamcast game?
Can somebody who couldn't have written that system from scratch understand enough of the existing implementation to modify it?
That's how one successful translation becomes more than one successful translation.
It becomes infrastructure.
And infrastructure is how you kill the bus factor of one.
“Nobody Will Ever Translate That”
I've heard some version of that sentence for decades.
Sometimes we were right.
Some games were simply too obscure.
Some were too technically difficult.
Some had enormous scripts.
Some required skills possessed by approximately three people on Earth, two of whom had moved on with their lives.
But I'm becoming increasingly uncomfortable saying it anymore.
Not because AI will magically translate every Japanese game.
It won't.
Not because publishers suddenly care about preservation.
They don't always.
But because the cost—in time, specialized knowledge, and sheer human suffering—of attacking these projects appears to be changing.
And when that cost changes, the list of games considered “not worth attempting” changes with it.
That's exciting.
It's also messy.
There will be terrible translations.
There will be arguments.
There will be people who hide their use of AI.
There will be people who advertise AI involvement like it's a substitute for quality.
There will be excellent translators whose work remains dramatically better than machine output.
There will be hackers who use AI-generated code they don't understand and create spectacular new varieties of broken shit.
Fine.
We've survived bad fan translations before.
We'll survive these too.
What interests me is the person looking at some forgotten Saturn, Dreamcast, PlayStation, Sega CD, or PC-98 game and realizing:
“Wait. I might actually be able to do this.”
Because twenty years ago, the answer might have been:
Find The Wizard.
Ten years ago:
Hope The Wizard eventually has time.
Five years ago:
Maybe somebody else will figure it out.
And increasingly, the answer may become:
Learn. Use the tools. Document what you learn. Ask for help when you hit the wall. Then leave something behind so the next person doesn't start from zero.
That's how this should work.
The important question isn't:
“Can AI replace our best ROM hackers?”
I hope that's not the goal.
The important question is:
“Can the next project survive without them?”
Because eventually every wizard goes home.
And the games should still be there when they do.
— Jerry “The Ankle-Biter” Silverhand