FAQ
Short answers. If your question is not here, the repository is open.
Do I need an API key?
No. Without a key every session runs on the scripted tier — and that is not a stub, it is a full engine. Rules, meters, effects, win and lose conditions are all computed by the engine; the model never touches them.
What you lose without a key: the connective narration of a standup, character voices, and depth in the retro. The standup becomes a templated digest, the retro a timeline from the event log.
In stage S2 the language model connects: characters will speak in the voice of their role and culture, and meeting scenes will come alive.
How many players?
Each scenario declares its own range. In all five scenarios that is one to four.
Playing solo works: bots fill empty seats and take their turn down a chain of rules, so the table never stalls. With a full table, asymmetric information and competing seat interests are live — which means negotiation. Alone, the game becomes a delegation sim.
Can I write my own scenario?
Yes. A scenario is a YAML file. All five packaged scenarios in backend/content/scenarios/ use the same format yours will.
Schema: docs/scenarios/schema/scenario.schema.json. Scenario model: docs/scenarios/scenario-model.md. Condition grammar: docs/architecture/condition-grammar.md.
A file passes load, validation and composition. The build fails on a card with a dominant option, on an undeclared meter name in an expression, and on an untelegraphed delayed effect.
What stage is the project at?
Stage S1 is built — the game core on the scripted tier. A full session is playable without a language model: five scenarios, the engine, meters, phases, decision queue, standing orders, policies, standups and the retro.
Stage S2 is in progress — the language-model layer: characters with voice, meeting scenes, and a full retro.
Stage S3 follows — onboarding, invite links, leaderboards, and public launch.
See the changelog for details.
What makes this different from other management simulators?
Two decisions set it apart. First: no round has a right answer. The build fails if a card has a strictly better option on the scored meters. This is not a quiz — it is a space of equally valid trade-offs.
Second: consequences are delayed. An effect of a decision can land several rounds later — telegraphed in advance. The cheap moment to prevent a problem passes before the problem arrives.