Sample Size

How many parts must you measure before the average means anything? n = (z·σ / E)² — and because the margin is squared, halving it costs four times the samples. That relationship is why measuring "a few more" rarely helps.

Assumes a roughly normal process and a known σ. With σ estimated from a small sample the t-distribution is stricter, so treat these as a lower bound. This sizes an estimate of the mean, not an acceptance sampling plan — use AQL for that.