simulation-first-design

Installation
SKILL.md

Simulation-First Design

Purpose: Design and balance games where the loop runs without a human in it — and know what that costs you in evidence.

Sources: No canonical design text covers this; the practice is scattered. Determinism and lockstep replay are standard RTS engineering (see any Age of Empires or StarCraft networking postmortem). Self-play as a balance instrument is established in reinforcement learning (Silver et al. on AlphaGo/AlphaZero; OpenAI Five; DeepMind's AlphaStar league) and the league structure there — deliberately diverse opponent policies to prevent overfitting to one strategy — is the transferable idea. Everything else below is generalised from working implementations rather than cited, and its thresholds are posture B under the Numbers Policy. Treat this skill as a practice being written down, not as received wisdom.


The evidence problem, first

Installs
5
GitHub Stars
2
First Seen
Sep 6, 2026