The datasets by which we showed that NBS 127 causes an offset in yields, 
voltages, and isotopic results in subsequent phosphate analyses were 
produced using reagent-grade silver phosphate from Alfa Aesar rather than a 
material requiring pre-processing.  Therefore, this offset is cannot be a 
consequence of precipitation technique.

At the Duke facility, we ran std-devs of 0.15 or less on large groups (>20 
samples) of AASP numerous times; it makes a great lab standard, running 
about 11 per mil, except under standardization by BaSO4.  AASP was seriously 
impacted by antecedent NBS127 samples without exception, and always in the 
same way: reduced oxygen yield and voltage, and an isotopic depletion of >2 
per mil; both responses tapering back toward optimum over 15-20 subsequent 
samples.  Barium sulfate standards had significantly less of an impact on the 
isotopic results of subsequent sugar, nitrate, or cellulose standards, but in 
some cases very seriously reduced the oxygen yield from them.

Naturally, we did not merely clean the reactors for this project whenever 
backgrounds began to rise (at or below his threshold), we replaced the entire 
assembly outright on a regular basis.  And of course like all of the instrumental 
parameters, linearity was regularly monitored; it seems that at Duke, linearity 
was not as dramatic a problem as it is at NCSU.

The 'rapid' precipitation process--the second of two described by Matt Kohn in 
the online repository for the Dettman reference (Geology Jan 01, if memory 
serves)--reliably produces ~100% phosphate from starting samples as small as 
1.5 mg, whereas the Crowson technique becomes unpredictable below about 
10 mg and in terms of yield underperforms the 'rapid' technique throughout its 
functional range.  The fossil tooth data included in the paper was ancillary 
data designed to show that the BaSO4 standards were interfering with real 
analyses and explain why unprocessed enamel (fossil or otherwise) isn't a 
good material for TC/EA analysis.  It also demonstrates that the problem 
occurred in real data sets from three different TC/EA facilities, including one 
produced under the full range of laboratory protocols established at NCSU.

Thanks for your interest.
Will Straight