Forty lab groups each time a pendulum several times to measure g, and each reports an interval. About 95% of those intervals catch the true value, and the few that miss are the 5% you were warned about. With small samples, using z instead of t misses far more often.