process
The 5-users rule
One test user surfaces roughly a third of usability problems; by the fifth you're mostly re-watching known ones – so run small rounds, fix, repeat.
Nielsen and Landauer’s model – Problems Found = N × (1 − (1 − L)ⁿ), with a typical single-session hit rate L around 31% – shows returns collapsing after about five users (one user surfaces ~31% of problems; five reach ~85%). The popularised version (“5 is the magic number”) misses Nielsen’s actual recommendation: as many small rounds as you can afford, fixing between rounds, because three 5-user rounds catch-and-repair iteratively while one 15-user round just documents a design that never got to improve.
The dropped caveat that matters: the maths assumes comparable users. Genuinely distinct user groups (hosts and guests, say) need 3–4 users per group, not 5 total.