One reason you may not be getting closer to your target gold recovery is that US law allows karat gold to actually be a bit less than the stamped purity. For example, any purity greater than 11.5 K can legally be marked 12K. To illustrate:
If you have 100 grams of 12K gold, the implied amount of gold is 50 grams. But if it is 11.5+ karat, the pure gold content may be as low as 47.916 grams - over a 2 gram difference. 47.9/50 = 0.958, which is roughly 4% less gold than the marked purity.
THanks for your input, yes I was aware of under-karating being one of the factors, think I included in one of the lists. Honestly thats my suspect for the highest amount of the "melt losses" Of course we don't live in a perfect world.
I also don't trust XRF that much, but this is better than than sending a box of gold in with so many questions about why it came out 95% this time, and 88% last time...
Taking responsibility for that first melt myself, that clears the dross, removes the possibility for an easy palming of a couple items and getting some semblence of certainty on the weights and measures of what's being sent in...
The first melt removes almost all the dross which I can preserve for analysis later vs leaving it in the refiner's discards. It also handles any low evaporation metals like zinc. There is a measurable loss of mass after the first melt, when melting the same bar 2nd or 3rd time, there is zero to negligible loss of mass.
The xrf analysis (and eventual pin samples once I acquire equipment) will help to mitigate the calculation uncertainties in averaging of hundreds of items into a single gold percentage of the expected overall yield. The pin samples can be retained for later analysis should the refiners results be outside of reasonable expectations.
Many smaller refiners (probably should call them aggregators since they rarely do refine to pure, just acquire and middleman to larger percentages) use xrf to determine their payouts, usually using the average or mean of 3 different machines etc.
I do trust my refiner, but its a lot safer to take responsibility for these variables that have plagued us for years.