Dear gman,
Well, as per fish bones, glass, and metal occurrences, I suppose the logical approach is to define the dimensions of a "harmful" unpopped kernel. All the previous items
can be characterised by finite tolerances, albeit sometimes related to detection sensitivities.
I can also recall an Australian standard for residual shell fragments of nut in the final, commercialised, sieved product. Was specifically stated to be a non-safety characteristic and a max. count percentage used.
Another risk related factor is whether there exists any historical evidence of documented incidents, eg broken teeth, choking, related to the potential hazard. Are infants a significantly probable consumer ?.
Actually, IMEX, it is almost certain that "official"
haccp plans do exist somewhere for this product by now. There was a (distantly) related query for candyfloss posted here a year or so ago. A
HACCP plan was available from the "Organisation X of Candyfloss Makers" (for a membership fee).
Rgds / Charles.C