game-balance-designer

Game Balance Design

Balance makes the numbers feel right — and balance failures are public: an economy exploit or a broken curve is discovered by players within hours of ship.

When to reach for this

• Designing or tuning progression curves, XP tables, or difficulty scaling
• Building an economy: currencies, sinks, sources, loot tables, pity systems
• Setting matchmaking MMR, rank decay, or placement parameters
• A live-ops tuning change needs a model and a telemetry plan before ship

Principles

1. Model before you ship. Build a spreadsheet or Monte Carlo sim that predicts the average and the 95th-percentile outcome — the unlucky player at p95 is the one who churns or posts about it.
2. Tune with config, not code. Every balance knob lives in remote-config or data files so a live patch never requires a client update or store review.
3. Design the sink before the source. Every currency has an inflation path; model net flow per player-day and per cohort before adding any new source.
4. Respect the grind curve. Hours-to-next-milestone is the most-felt number in the game — chart it across the whole progression, not just the first ten levels.
5. Find the exploit before players do. Adversarially playtest the economy: max one axis (time, money, trading, a single repeatable action) and see what breaks.
6. Pity systems bound variance. Any low-probability reward (roughly < 2% drop rate) needs a hard or soft pity ceiling, or p95 players experience it as never.

Balance change worksheet

For any tuning change, produce these five artifacts before it ships:

Artifact	Contents
Model / sim	numeric model with assumptions stated; mean and p95 outcomes
Proposed values	exact tunables in the game's config format, with old → new
Grind delta	hours-to-milestone before vs. after, per affected segment
Exploit surface	what combinations this opens or closes; the maxed-axis test result
Telemetry plan	the specific metric (and cohort cut) that confirms or refutes the change post-launch

Pitfalls

• Tuning from aggregate averages — segment by cohort and spender tier, or the whales mask the curve everyone else feels
• A new currency source with no matching sink — inflation shows up weeks later and is far harder to remove than to prevent
• Drop rates stated without variance — "1% drop" without pity means 1 in 20 players hasn't seen it after 300 attempts
• Difficulty tuned on developer skill — devs are p99 players of their own game
• Hardcoded balance values that turn a number tweak into a client release
• Shipping a tuning change with no pre-registered success metric, so the post-hoc telemetry read becomes a Rorschach test

---

Related: game-liveops (validating tuning with live telemetry), game-netcode (matchmaking/session lifecycle) · domain agent: game-architect (core loop the economy hangs off) · output/ADR format: playbook-conventions