Data

Experimental procedures

The silver samples were mounted in epoxy resin, polished down to 1μm using alumina paste and observed under the optical microscope throughout the polishing process. Four silver reference materials were used for calibration purposes: MBH131XPAg1 is a pure silver standard (99.9%Ag) with trace elements, including gold (120ppm). AGA1, AGA2 and AGA3 have major and minor elements including gold, as well as trace elements (Table 10). AGA2 was used only as an internal calibrant during measurements.

A JEOL JXA-8100 electron probe microanalyser with a wavelength dispersive X-ray spectrometer (WDS) was run with a 20kV accelerating voltage and a probe current of 5×10-8 μA. Samples were examined at a working distance of 11mm. The system was calibrated at ×1000 magnification (about 100μm × 100μm) by curve-fitting to three known standards (AGA1, AGA3 and MBH131XPAg1) for each of the following elements: Ag, Au, Zn, Cu, Pb, Sn, Sb and Bi. Emission lines were chosen to minimise overlapping peaks, and peak/background acquisition times (in seconds) reflected the absolute concentrations present: Ag (La) 30/10, Au (Ma) 60/20, Zn (Ka) 60/20, Cu (Ka) 30/10, Sn (La) 60/20, Pb (Ma) 60/20, Sb (La) 60/20 and Bi (Ma) 60/20. Nine areas were measured on each standard. Linear and quadratic fits were compared for each element during the calibration set up. A linear regression was adopted for all elements as differences between the two fits were found to be negligible. Errors on each element were determined measuring an internal standard (AGA2) before and after all sample measurements. Limits of detection (LOD) were calculated for each element. Measurements below these values were classified as below the detection limit (bdl). Note that the high LOD level for silver reflects the extrapolation from the silver standards, which, by definition, have very high concentrations of silver, i.e. the calibration is only appropriate for silver-dominant systems.

Results

We present compositional and LIA data tables delineated by the find site. Pernicka et al.'s (1983) normalised compositional data and LIA values are included, but we use Stos-Gale's LIA values (Stos-Gale and MacDonald 1991; Stos-Gale 2014) and our EPMA data for the analyses. We have not used the XRF data presented in Stos-Gale and MacDonald (1991) because some elements showed appreciably different concentrations (e.g. Cu) with our EPMA measurements. This may be a consequence of issues surrounding corrosion which can affect surface techniques like XRF or overlapping X-ray peaks (both of which are less of a problem for EPMA on sectioned material).

Each EPMA measurement was repeated at least three times. Table 11 presents the averages of these measurements. Non-normalised values are presented along with the absolute totals. Some of the totals are low because the samples were corroded. Severely corroded samples are denoted with their totals in bold type. Some samples have low totals because the dimensions of the sample were slightly narrower than the field of view of the EPMA. These are denoted by an asterisk. Although, the field of view could have been adjusted, it was decided to maintain the same magnification throughout the analyses to that used for calibration (to minimise issues regarding homogeneity at different magnifications), i.e. normalised values or ratios of the elements are used in the subsequent analyses, calculated from the absolute values and totals presented in Table 11. LIA data from Pernicka et al. (1983), Stos-Gale and MacDonald (1991) and Stos-Gale (2014) are presented in Table 12.

Compositional data and LIA data were analysed together by applying the following strategy: (1) LIA values are used to calculate the Pb crustal age (also known as the model age) of the ore, measured in millions of years (Ma), from which the silver derived (using a two-stage evolution model and the parameters of Desaulty et al. 2011) (see Wood et al. 2017a; Wood et al. 2019); this age is then (2) plotted against the levels of gold found in silver, given that gold is a useful geological indicator for a silver source as it is likely to survive the smelting and refining operations. This approach allows two types of data, isotopic and elemental, which are often geologically linked and can inform on provenance, to be plotted together. It should be noted that different parameters to calculate the Pb crustal (model) age are used by different groups of researchers, and the same groups often update the parameters they use (e.g. Albarède et al. 2012). The aim here, however, is not to calculate crustal ages per se but to define a value for both the ore sources and the associated objects: both should have the same lead isotope ratios and hence the same Pb crustal age, whether or not this corresponds with the actual geological date. In fact, the calculations sometimes yield negative values (i.e. future ages). This could mean that these leads have had a more complicated history than might be expected from the predicted rate of addition of radiogenic lead in the source region, during which they acquired extra amounts of radiogenic lead. Alternatively, it could mean that the model or the parameters are not sufficiently refined to represent the full range of ages. Nevertheless, for the purposes of this article, the parameters used to calculate crustal ages for both the ores and objects were the same.

All LIA plots show ± 0.1% error, which was the maximum error on repeat measurements on an isotopic standard for the thermal ionisation mass spectroscopic (TIMS) technique used by the Oxford group (Stos-Gale 2014). Repeat LIA measurements on samples are within this error, as are the calculated crustal ages, i.e. three measurements for SG479-1, SG479-2 and SG479 (15, 10 and 14Ma, respectively), two measurements for SG478 (-17 and -22Ma) and SG481 (18 and 22Ma).