First thanks for your quick response.
Metal can comes from several sources including raw materials, closure systems, and the same manufacturing equipment. Especially from those that have moving parts like mills and tablet press.
My dilemma is that due to several factors (e.g.: metal check sensitivity, limited room space, vibration, etc.) a significant amount of cores (> 1% of 800 kg batch) are typically rejected during the compression process.
Most of this material are false rejections. It is a fact that at least a dozen of tablets are “rejected” everytime the diverting mechanism of the metal detector is activated. Not all of them necessarily have metal. I did a quick trial in one occasion where more than 15 kg (~10,000 cores) were collected as combined wastes from right and left sides metal detectors. After the completion of the batch, the rejected tablets were slowly passed through the corresponding metal detector and about 20 tabs in total were rejected. However, this is not our normal practice since our effective SOP only instructs the operator to document the amount of waste and discard it.
From the quality point of view, the excessive amount of false rejected tablets eventually may mask a real metal content issue that could put at risk the patient health. That is why the auditor is requesting that a limit is established, either by weight or number of tablets, to discriminate between what would be an acceptable amount of rejected tablets in a typical batch and what should rise a flag (investigation) to determine if a batch can be released for distribution if the limit is exceeded.
The actual observation was:
There is no limit defined for rejects in the metal detection during tablet compression. The rejects are not checked, but just discarded at the end of the compression of the batch.
My intention is to define a technically sound limit that fit the purpose but at the same time do not create a problem due to excessive number of investigations due to false rejects. I believe that for the last part the reprocessing at the end of the batch to differentiate the true and the false rejects can be a fix.
Where I need more help is with the rational I must use to define the limit.