Rendering state

tyrant.engine.renderer turns a GameState into something an LLM can read. Both functions take the state and the viewer's uid, and scrub the state for that viewer before rendering, so the full state can be passed directly:

from tyrant import render_dict, render_text

prompt = render_text(state, uid)
payload = render_dict(state, uid)

render_dict

render_dict returns plain dict, list, str, int, bool and None values, so it can be passed straight to json.dumps. It has these sections:

Key	Contents
viewer	The viewer's uid, role, party, whether they're alive, and their known teammates
players	Every player in seat order, with party, role and whether they're alive
board	Policies enacted and needed to win, the fascist track, the election tracker, the Hitler zone, veto, and the pile sizes
government	The president, the chancellor and the previous government
turn	The phase, who the game is waiting on, the active power, the viewer's hand, the winner and the discussion
timeline	Every public event
knowledge	The viewer's private knowledge

render_text

render_text builds Markdown from the same dict, with sections for the viewer's role and known teammates, the players, the board and election tracker, the government, the current phase, who must act and the discussion topic and round, a numbered timeline of public events and discussion, and the viewer's private knowledge. Hidden values always render as unknown.

# Secret Hitler: Player 0's view

## You
- Player 0
- Role: liberal
- Party: liberal
- Status: alive
- Known teammates: none

## Players
In seat order:
- Player 0 (you): party liberal, role liberal, alive
- Player 2: party unknown, role unknown, alive
...

To keep prompts small, render_text shows speeches from only the last two discussions; older ones collapse into one line each. render_dict keeps every speech.