I’m not a huge fan of hard rules because the statistical decision making power shifts with the batch size, and I like to stick to standards
USP <905> is a standard, it is lower power because it assumes FDA-level process validation, and it doesn’t account for batch size where representative sampling breaks down. But it does use variable data to make predictions about batch quality and it’s 10 per batch.
Nobody will want to do ANSI Z1.4 / ISO 2859 by attributes standards; requires far too mamy samples (hundreds). Attribute data has way less power than variable data, so the sample requirements are higher. If you have really small batch sizes, AQL General Level II 1% might be financially feasible.
That leaves you with variable data and process capability indices (Cp/Cpk), which still require at least 20-30 samples; but is still a recognized standard.
Beyond that, you have the work of Dr. Wayne Taylor, who was the FDA expert witness on statistical sampling.
His improvement combined variable data process capability and AQL/defect rates to allow smaller sample sizes (as small as 6). His sampling plan limits just get tighter the smaller the sample size to account for the decrease the statistical power.
If I was running a peptide factory, I’d use a combination of process capability controls and Dr. Taylors work for lot release.