Game Balance Lab — 9 Genre Toolkit

Make balance decisions you can explain

Game Balance Lab turns a gameplay complaint into a baseline model, linked parameter changes and a testable experiment plan. Use it for game economies, combat/progression pacing and upgrade tradeoffs when a compatible agent needs practical numerical guidance.

One self-contained English toolkit

The bundle consolidates Whaocaii's numeric orchestrator, shared checks and nine genre workflows. Dependencies are inside the ZIP, so the entrypoint can route to one relevant guide without asking you to install sibling skills.

Genre coverage: adventure/platformer/puzzle progression; deckbuilding roguelikes; loot-driven ARPGs; long-term RPG/MMO economies; shoot-em-ups; management/factory/idle games; survivor-like games; tower defense; turn-based card RPGs.

The guides help you define the right model and checks. They do not implement nine game simulators or promise automatically optimal balance.

What is included

• Root SKILL.md with English routing, input assumptions and delivery criteria.
• Shared unit, probability, resource-conservation, boundary and sensitivity checks.
• Nine compact genre guides with the relevant inputs, analysis and observable checks.
• CSV templates for parameters, resource ledgers and experiment plans.
• A runnable JSON production scenario.
• Standard-library Python reference calculations for capacity-limited sales, incremental net-flow payback, independent drop tails, a single hard-pity cap and specific-card draws without replacement.
• Three complete synthetic worked cases, expected JSON output and a local HTML report.
• Eighteen arithmetic/input-boundary tests, also rerun after ZIP extraction.
• Original MIT license and source/change provenance.

A concrete example

In Pocket Bakery, arrivals and one oven each support 20 items/minute, but service supports only 15. Sale price is 5 coins and ingredient cost is 3, so the baseline net flow is 30 coins/minute.

A 40-coin service upgrade raises actual sales to 20/minute and net flow to 40 coins/minute: incremental payback is four minutes. An 80-coin oven alone adds no sales while service remains the bottleneck. With only 60 starting coins, that oven is also not a valid starting purchase.

At the same demand, buying both upgrades for 120 coins gives only the same 10 coins/minute gain, so simple payback is twelve minutes. These linked checks catch an upgrade that sounds useful but changes no actual output.

The other cases show why 2% independent drops can leave 13.26% of players without a drop after 100 attempts, and how drawing two key cards differs between a 20-card and a 16-card deck. All fixture values are fictional; no telemetry or playtest is implied.

How to use it

Ask your compatible agent to use $game-balance-lab. Supply the current rules/configuration, player/session horizon, units, permitted changes and available observations. Request a design, review or configuration proposal. The skill preserves your requested format and separates assumptions, calculations, simulations and observed play.

Run the included calculations with Python 3.10+:

python3 scripts/balance_math.py --scenario templates/production-scenario.json
python3 scripts/run_examples.py
python3 -m unittest discover -s tests -p 'test_*.py'

No Python packages, network requests or model API key are needed for these reference scripts. Agent availability, its normal account access and local tool permissions remain your responsibility.

Free edition and open-source origin

This consolidated English edition is available free of charge, including reusable templates, verified reference examples and calculators. The public base workflows remain free under MIT at https://github.com/Whaocaii/ai-game-skills . This listing does not hide that source or claim exclusive rights. The English additions are recorded in the package and have not been pushed to the public repository.

Evidence and boundaries

The eighteen tests check the narrow reference arithmetic, invalid input handling and small-deck enumeration. They do not validate gameplay enjoyment, retention, revenue, real-device combat timing or every genre workflow. The production example is a steady-state algebraic model, not a discrete queue simulator. Soft pity, searches, reshuffles, inventory/spoilage events and other rules require an expanded model. This package includes no game engine, telemetry collector, model service or live-economy deployment. Cross-agent installation compatibility was not tested.