soak-stability
Soak stability
Choose duration and workload from the product risk, not a universal hour count.
| Mode | Example workload |
|---|---|
| active-loop | repeat representative gameplay/content cycles |
| overnight | long mixed active/idle run |
| idle-resume | suspend, background/tab-out, resume repeatedly |
Measurement
Pin build/device/scene route/settings. Sample memory/resident resources, object/resource counts, frame-time p50/p95/p99 (or the project's percentiles), hitch counts, load/save outcomes and subsystem-specific counters over time.
Judge trend/slope, not only start-versus-end. Warm-up/cache growth may plateau; a persistent positive memory/resource slope or increasing hitch percentile after repeated cycles needs investigation.
Repeat save/load against the selected test slot/profile from save-integrity/save-systems; do not hardcode slot 0. Preserve the previous good generation when injecting failures. Repeated audio/network/listener setup must not accumulate duplicate handlers or loops. Effect pools, voices, timers, overlays and generated caches plateau across repeated round/scene cycles and background/resume.
Automation may drive the soak, but label it as automation and do not equate endurance with a human usability/clear test.