An empire and lands beyond#
The setting begins in a fictional empire inspired by Russia around the turn of the twentieth century. It is not historical Russia. Steam engines, telegraph wires and the first electric lights share the streets with licensed magic. Slavic folklore gives us house spirits, dangerous forests and a world of the dead. The names, states and events are our own.
A ministry may regulate sorcery. A telegraph office may delay a rumour. A clerk may demand the correct stamp before admitting that somebody has died. The bureaucracy is a joke, but it is not scenery: institutions know things, own things, want things and run up against their limits.
In the long run, the world should reach the scale of Daggerfall or beyond: many lands, not one city stretched forever. One design puts an English-speaking state across the border, with whole regions where English is the local language. Its people meet the empire through travel, trade and a Foreign Quarter filled with consulates, trading houses and hired engineers. Russian-speaking characters may begin as subjects of the empire; English-speaking characters may begin in foreign lands. That is a difference of origin inside the fiction, not a server choice. It is the scale we are aiming for, not a claim that the map already exists.
The same fact in different words#
The rule is simple: there is one shared world. Russian and English players do not disappear into separate language shards.
Beneath the prose, the game keeps facts, beliefs, relationships, rumours and contracts as structured information. The narrator can tell one player about an event in Russian and another in English without creating two versions of what happened. A rumour crossing the market remains the same claim even when the words change. If somebody breaks a promise, both language groups live with the same consequence.
Conversation is messier. One idea we want to explore is an in-world translation artefact that a character buys, then recharges or refuels. Every translated message would drain part of its reserve. The real cost of an LLM request would become a visible rule of the world: translation is a finite resource, not an invisible free service. Players should still be able to read the original speech. Many useful actions need no machine translation at all—pointing, giving, trading and offering a contract can carry structured meaning across the language boundary. A player should also choose the interface and narration language independently of the character’s origin.
The device should look as though somebody had ruled convenience an unnecessary luxury: a strange brass box with two speaking trumpets, a language dial and a crank. One person talks into it, turns the handle and waits through a worrying series of clicks. Then a mechanical voice announces in the other language: “Oh, yes. That seems about right.” Nobody is quite sure whether that was the translation or the machine’s professional opinion. Even the noise has a job: model latency becomes machinery doing something terribly important.
Difference should create cooperation#
Different origins earn their place only if they help people meet. A native character may have the trust of a commune, access to local offices and knowledge of folk customs. A foreign character may have consular protection, trading connections and doors opened by foreign engineers or the upper class. A problem that crosses city districts or national borders becomes easier for a mixed group.
Language learning could matter later: a few words understood without translation, lessons from another player or private speech used deliberately. It must not turn ordinary play into a comprehension test. By default, the game should help people communicate. The differences should create reasons to cooperate, not excuses to build two communities that never meet.
Most of this is still design work, not a list of finished features. We have to prove that translation can preserve meaning and tone at an acceptable speed and cost, that moderation works in both languages, and that Russian-speaking players do not become unpaid support for everybody else. The test is practical: when a language mechanic appears, does it start a conversation between the groups? If it separates them, it has failed.