Charles,
I have a few thoughts. Much of if comes down to the fact that random variation makes it hard to be sure on any results. Also, remember that this spreadsheet is based on a binomial distribution, which assumes that you are checking only a small part of the total "production" (i.e. calibration reports). Some of the later number suggest you might have to check a large proportion of the monthly reports to get adequate accuracy, which means a more accurate model would be needed.
As I see it, the goal of a sampling plan is to take products that can be categorized into three groups: "good" = better than AQL, "bad" = worse than LTPD, and "indifferent" = between AQL and LTPD, and try to sort them into two groups ("pass" and "fail").
Ideally, bad lots will never pass and good lots will never fail. Indifferent lots may pass or may fail. In reality, bad lots will sometimes pass, and good lots will sometimes fail.
In your case a lot is the monthly output of 1 tech. A good tech produces 1% or less defects. A bad tech produces 5% or more defects.
If alpha=beta=5%, then you need to test at least 124 samples from that month to get to this level of confidence. Note that indifferent techs could easily pass or fail. Perhaps any tech that fails 3 times in a row could be required to have extra training.
Going the other way, if you wanted to be 95% sure he really was outstanding, you would need to test at least 299 pieces and find no defects (set alpha = 95 in the spreadsheet). On the other hand, to be 95% sure he really "needs improvement", you would just need 1 of 7 bad.
You could also work backwards. Your typical sample size seems to be about 20. If AC = 0, then alpha ~ 18% (there is an 18% chance that a good tech could fail) and beta ~36% (there is a 36% chance that a bad tech could pass and be told he’s OK).
You might also consider a 2-proportion test to see if two results (for example, tech 1 vs tech 2, or tech 1 vs the overall average) are indeed different.
Finally, a control chart could be a valuable tool. See if there any of the techs go "out of control". A p chart should do the trick.
NOTE: I'm still waiting for confirmation from a second source for the accuracy of the spread sheet. Also, I didn't double-check the numbers above, so you ought to get a second opinion before blindly accepting them.
Tim F