Tune with data and math; decide with play
GDC-L1-BAL-000501
Statement
Use quantification — cost curves, spreadsheets, telemetry, win-rates — to reason about balance systematically and to flag outliers. But make the final call by play: math finds what's suspicious; playing decides what's fun and fair. The harder quantities are to compare, the more you must lean on playtesting.
02
Rationale
Numbers are indispensable for balance at scale: a cost curve lets you price a hundred items consistently, and telemetry surfaces the option with a 70% pick rate you'd never spot by feel [S-schreiber-balance]. But numbers also lie by omission — they can't fully capture how an option feels, how it interacts with everything else, or whether a statistically-fine option is miserable to play against. So the two methods are complementary: quantify to narrow the search and catch outliers, then play to judge. Schreiber's own caveat — the harder direct comparison is, the more playtesting you need — captures the balance: math scales, play judges, and the murkier the math, the more the judgment falls to play. This is the balance-specific form of "measure, don't guess" (PERF-0001) fused with "watch behavior, not just numbers" (PLAYTEST-0005).
03
Applies when
Any balance pass, at any scale — from pricing a single item to tuning a live competitive roster. Data-heavy at large scale; play-heavy where interactions are complex or subjective.
04
Does not apply / Exceptions
Tiny option sets can be balanced almost entirely by play (no spreadsheet needed for three weapons). Conversely, at massive scale (live-service with millions of matches) telemetry carries more weight — though even there, the numbers point and humans still decide (PROG-0004's warning against optimizing the metric instead of the experience applies). Pure faith in either extreme — spreadsheet-only or vibes-only — fails.
05
Implementation
Build a cost/value model so options are priced consistently, and instrument the game to see pick rates, win rates, and drop-off (PLAYTEST-0005). Use both to find imbalances; then playtest the flagged cases and decide by experience, not by the number alone. Where quantities resist comparison, weight playtesting more. Keep before/after data so you can tell a change helped.
06
Disagreement
Data-driven balance (quantify, trust metrics, A/B test) vs. designer-judgment balance (play, feel, expert intuition). Data scales and catches blind spots; judgment captures fun and context that numbers miss. The synthesis — data flags, play decides — is what most balance designers converge on; the debate is the weighting, which shifts with scale and how measurable the game is.
07
Notes
The methods principle of BAL, unifying the "measure, don't guess" thread (PERF-0001, PLAYTEST-0001/0005) with the reality that balance is ultimately a judgment about experience. Confidence 4.
↔
Connected principles
S