Automate regression; human-test feel and emergence
GDC-L1-QA-000201
Statement
Use the right kind of testing for the question. Automate what is deterministic and regression-prone — systems, math, save/load, builds — so it's checked cheaply and constantly. Use human testing for what machines can't judge: fun, feel, difficulty, and emergent behavior. Neither replaces the other.
02
Rationale
Games have two very different testing needs. Deterministic correctness (does the damage formula compute right? does save/load round-trip? did that refactor break anything?) is perfect for automation — fast, repeatable, and it catches regressions the moment they appear, which protects iteration speed (ARCH-0005). But the questions that most determine whether a game is good — is it fun, does it feel right, is the difficulty fair, what emerges when systems interact — are subjective and situational, and only human play can answer them (which is why playtesting, PLAYTEST-0001, exists) [S-games-user-research]. Trying to automate "is it fun" fails; trying to manually regression-test every build wastes people and misses things. Matching method to question (and to risk, QA-0001) gets both cheap regression safety and real judgment about the experience.
03
Applies when
All testing. Automation scales with deterministic, regression-prone systems; human testing scales with subjective/experiential and emergent aspects. Most games need both.
04
Does not apply / Exceptions
Highly emergent or non-deterministic systems resist automated testing (and automating a fast- changing prototype's tests can be wasted effort as it churns — automate stable things). Tiny games may rely mostly on human testing. And automation has real setup/maintenance cost that must be justified. The split is a guide, not a mandate to automate everything automatable.
05
Implementation
Write automated tests for stable, deterministic, regression-prone systems (formulas, save/load, core systems, build validation — CONTENT-0004/0005). Run them continuously (CONTENT-0005). Reserve human testing — structured playtests (PLAYTEST) and exploratory QA — for feel, fun, difficulty, and emergent/edge behavior. Prioritize both by risk (QA-0001).
06
Disagreement
Test-automation-heavy (fast regression safety, but setup cost and can't judge fun) vs. human-testing-heavy (judges the experience, but slow and misses regressions). The synthesis — automate the deterministic, human-test the experiential — is broadly held; the debate is how much to invest in automation for a medium where the most important qualities are subjective.
07
Notes
The method principle of QA; pairs automation with human playtesting (PLAYTEST-0001) and protects iteration (ARCH-0005) via regression safety. Confidence 4.
↔
Connected principles
S