Why would IBM set a hard limit on that, though? Assuming defects are normally distributed, it would make more sense to say: mean of X per N defective parts, wit

Why would IBM set a hard limit on that, though? Assuming defects are normally distributed, it would make more sense to say: mean of X per N defective parts, with a variance of Y with a P-value of Z. Though who am I kidding, no one probably evaluates production like that. Though I imagine chip makers are pretty serious about that kind of stuff.