Introduction
We've been spending a lot of time lately looking at how fandom communities store and share knowledge about the fictional universes they care about — game data, character abilities, lore timelines, item stats, event history. The more we looked, the more one thing stood out: there's a lot of it, it's genuinely impressive work, and almost none of it is built to talk to anything else.
The knowledge already exists
Take almost any game with a dedicated fanbase — Genshin Impact, Honkai: Star Rail, Blue Archive, whatever — and you'll find wikis, datamining repositories, community-run APIs, character build calculators, and fan-maintained spreadsheets tracking things official sources never bothered to document cleanly. Someone, somewhere, has already extracted the data and structured it well enough to build a tool on top of it.
That's not a small thing. It represents years of volunteer effort, reverse engineering, and careful cross-referencing against patch notes and datamined files.
But none of it agrees on a shape
Here's where it gets messy. Every one of these projects tends to invent its own schema, its own naming conventions, its own assumptions about what a "character" or an "ability" even is. A field called effect in one dataset might mean something structurally different from a field with the same name in another. Version history is tracked inconsistently, if at all. Some datasets update within hours of a patch; others lag for weeks or get abandoned entirely.
None of this is a criticism of the people building these things. Most of them are solving a specific problem for a specific community, not trying to build something interoperable with datasets for entirely different games. There was never a reason for them to agree with each other.
Nobody tracks where a fact came from
The part that stands out most to us is provenance — or the lack of it. A number in a spreadsheet might be an officially confirmed value, a datamined value that hasn't shipped yet, a community-inferred estimate, or something someone typed in three years ago that nobody's rechecked since. Most of the time, there's no way to tell which.
That distinction matters more now than it used to, because these datasets are increasingly being fed into AI tools, chatbots, and assistants that answer questions about these universes. When an AI states a fact with total confidence, it inherits whatever confidence — or lack of it — existed in the source data, and that inheritance is usually invisible.