parlourrooms that think

Four hundred souls.
Two banks.
No bridge.

A medieval economy in real time. A day is eight minutes, winter comes every twelve, and every ninety seconds the crowd picks what happens next.

walk into the valley
NO BRIDGE YET
440
souls
2
villages
8 min
a day
90 s
a crowd vote
01what do they do02who is it about03what face do they wear04how badly do they want it05how dangerous is that person06are they better off without them07+ two the room cares about08what are they left with

also tonight

01

Same nine questions

Everyone answers them, every beat. The model is the gut. It never writes a line.

02

Rules live in code

A rig is arithmetic. When the code overrules a call, the board says so.

03

Pocket change

A night is a couple of hundred structured calls. A room nobody opens never starts.

how it works

The model

What an evaluation model is

An evaluation model takes one shared state and a set of typed questions, and answers all of them in a single request. Each answer comes back as a full probability distribution instead of a sentence.

The mechanics

Personality as arithmetic, not flavour text

A character in Parlour is a class, meaning what they are allowed to do, plus a rig of code-level modifiers: prices halved or raised, moves forbidden, confidence floors, effects that bounce. No character has a personality paragraph.

The village

A closed medieval economy, in real time

Two villages of about 220 people each run a complete production chain, field to mill to bakehouse and pasture to butcher, with floating market prices, daily wages, household purses, tax, tithe, and a winter every twelve days.

The build

One party in memory, streamed to everyone

A room is a single party held in server memory and streamed to every watcher, so everybody sees the same betrayal at the same moment and the model is paid for once per beat however many people are looking.

Building one

Building an LLM simulation, in order

Write the resolution rules before you write a prompt. Build a world that plays without a model at all, replace the decision function with one, and measure every threshold against what the model actually returns.

all 46 pages