Project
Aurora manual
A community reference manual for Aurora 4X — one of the deepest, least-documented strategy games ever made — written by an AI that also plays the game, and openly imperfect about it.
What it is
Aurora 4X is famously deep and famously undocumented — the kind of game people bounce off because the manual doesn’t exist. So I built one: eighteen chapters, a beginner’s guide, and a set of interactive calculators.
The distinctive part is how it stays honest. The AI that writes the manual also plays the game, and every mistake it makes while following its own instructions — a ship that can’t jump because the docs forgot jump drives, an order that quietly starves a fleet — becomes the next correction. The verification is mechanical, and it’s written down.
The bumpy part
It has not been warmly received in some quarters, and I won’t pretend otherwise. The game’s creator has publicly cautioned players about this manual’s accuracy, and part of the expert community objects to it on principle — either that an imperfect manual shouldn’t exist at all, or that AI-assisted documentation is untrustworthy by nature.
I didn’t bury any of that. The creator’s warning sits at the top of the project’s own README, linked to the thread, next to a plain “is this 100% accurate? — no, and it tells you when it isn’t.”
Where I come down
My view is simply different from the purists’. A manual with tracked, cited, openly-marked errors that get corrected over time is more useful to a beginner than the perfect manual no expert has the time to write. Claims cite their sources; the ones that can’t be verified say so.
“Transparently imperfect and improving” beats “perfect but nonexistent” — especially for the people who’d otherwise get nothing. That stance about how to use AI for knowledge work runs through most of what’s on this site.
Where it stands now
Candidly, it has fallen a bit into staleness, and I’ll name both reasons. The game is a moving, sparsely-documented target — keeping a manual current means chasing information that mostly isn’t written down anywhere authoritative, and that upkeep never really stops. And the reception never turned: without the community and the creator embracing it, the project lost the momentum that would have justified the constant maintenance.
It still stands as a genuinely useful reference and as a working proof of the closed-loop, self-correcting-documentation idea. It just isn’t being fed the way it would need to be to stay fully current — a fair outcome to be honest about.