Change one material variable
Hold the keyboard, plate, caps, mount and environment constant when comparing switches. If multiple parts change, record the result as a complete-build comparison, not a switch-only result.
Use more than one sample
A single switch can be damaged, over-lubed or unusually loose. Record sample size and whether positions were rotated. Manufacturer tolerances make variation a property of the test, not an annoyance to hide.
Separate observation from recommendation
Record smoothness, tactility, loudness, fatigue and wobble as personal observations in the named configuration. A recommendation needs an intended user, board, budget and trade-off—not just a five-point score.