Ruleset Mechanics

Bardo no longer assumes that mechanics only come from a small built-in rules registry. The current system can load workspace-defined mechanics manifests and treat them as first-class rulesets for validation, discovery, and resolution.

Workspace rulesets#

Uploaded or workspace-authored rulesets can define action vocabulary, intent aliases, target ranges, modifiers, contested checks, outcome bands, resources, clocks, and consequence chains. That gives Bardo a way to work with the actual game the table uploaded instead of forcing everything through one hardcoded model.

The discovery entry point for agents is ruleset_mechanics_overview. It exposes the available rulesets, action types, support level, and consequence structure before the client tries to validate or resolve anything.

Support tiers#

Each action is treated as full, partial, or advisory support.

• full means Bardo can validate and resolve the action authoritatively.
• partial means Bardo can scaffold the action, but the table still needs to finish the ruling.
• advisory means Bardo can explain the likely procedure, but it should not present the result as authoritative.

This keeps custom rulesets useful even when they are not fully executable yet.

Consequence composition and branching#

Rulesets can now define consequence chains that apply multiple follow-up steps in order. A single action can spend a resource, tick a clock, and then branch into follow-up chains when a later consequence becomes relevant.

That branching model matters for fiction-first systems. A narrow success can carry a cost, that cost can create leverage, and that leverage can branch into follow-up chains instead of forcing the agent to invent disconnected consequences. In other words, consequence chains can now branch into follow-up chains instead of stopping after one flat list of effects.

Table decision nodes#

Not every procedure should auto-resolve. Some systems explicitly want a human choice after the mechanics settle.

Bardo supports explicit ask_the_table decision nodes inside the consequence plan. Those nodes are surfaced in structured output so an agent can say, in effect, “the rules procedure points here, and now the table needs to choose.”

That is the intended use for high-value human judgment: preserve the procedure, preserve the context, and be honest about where the table still has to decide.