Systemic-Emergent Design

How to author affordances, not solutions: a small set of consistent, deeply-simulated, interacting rules that produce more meaningful play than any volume of scripted content — and, for a procedural/AI-authored game, the only thing that makes generated content cohere rather than collapse into "10,000 bowls of oatmeal."

When to use this

• Designing the rule substrate / "chemistry engine" before authoring or generating content
• Deciding whether to add a system, a verb, or a scripted set-piece
• Auditing for emergence: are systems actually interacting, or siloed behind scripts?
• Making procedural generation produce perceptual variety, not just numerical variety
• Diagnosing: "the world feels inert", "all variety is scripted", "every generated quest is the same", single-solution gates

Scope

This skill owns the emergence-from-interacting-systems discipline. Adjacent concerns live in sibling skills, cross-linked from the guide:

• The "interesting decision" test and general theory → game-design-fundamentals
• Perceptual uniqueness, the oatmeal problem, handcrafted-anchor + constrained-fill → procedural-generation, procgen-review
• Combat verbs, telegraphing, the feel of interactions → combat-design
• Currencies/items as systemic chains (currency-as-material) → rpg-systems
• The spatial affordances systems act on → open-world-design
• Quests that emerge vs. branch → narrative-and-quest-design

How the pieces fit

• GUIDE.md — the cited why, in five sub-domains: the immersive-sim lineage (universal rules; intention & perceivable consequence), emergence from few deep systems vs. scripted events, the "good GM" analogy and its honest limit, multiplicative vs. additive design (Nintendo's chemistry engine), and how emergence + procedural generation multiply each other. Each rule carries an exemplar, source, test-for criterion, named failure mode, and the procedural/headless implication.
• CHECKLIST.md — Do/Don't + machine-checkable Test-for criteria, grouped by sub-domain, written to be enforced as automated gates in a generation loop.

The one idea to anchor on

You never design play directly — only the rules that give rise to it (the "second-order design problem"). Author orthogonal affordances and let them combine; keep the rules near-exceptionless and mapped to real-world expectation so players learn them by exposure; guarantee every verb has intention and perceivable consequence; and freeze the rule substrate before you generate, so the simulation — not the AI's whim — validates emergent solutions.

Why this matters doubly for a generator: producing more content is cheap; producing more interactions is hard, so a generator told to "maximize variety" produces oatmeal, and one that "helpfully" adds bespoke interactions reintroduces exceptions that break learnability. Score the agent on interaction density, constrain it to tag-assignment and situation-composition over a frozen rule table, and forbid it from ever inventing a new rule or exception.

Start with GUIDE.md, then apply CHECKLIST.md.