We have identified several lines of evidence that can supply quantitative bounds on systematic measurement errors, and on some kinds of modelling errors. Taken together, these checks furnish some confidence that, for front-illuminated devices, the absolute quantum efficiency errors are no larger than 5% in the 0.4 - 4 keV band.
This result suggests that, although it cannot be ruled out at present, it is unlikely that broadband errors as large as 5% could remain in the absolute quantum efficiency models of the reference detectors, at least in the BESSY passband.
Of course our internal consistency check cannot rule out all systematic errors. For example, this check would not reveal a geometry error that affected every BESSY measurement identically (though the normalization results described above suggest such an error is highly unlikely.) It is noteworthy, however, that the synchrotron measurements in question were made on different runs separated in time by about a year, and with different sets of electronics.
Nevertheless, since the SSD detector is thought to be fairly well understood at higher energies, and since we have no other external check on our high-energy absolute calibration, we have compared the ACIS flight detector model predictions to ACIS-to-SSD relative QE measurements made at XRCF. The XRCF phase-I results used for this purpose are detailed in table 4.36 in section 4.6.6. If we use the ACIS detector model quantum efficiencies, together with these relative QE data, to infer the detection efficiency of SSD5 at 8.040 keV, we obtain ten estimates of the SSD5 efficiency. The mean value of these estimates is 0.966, and the standard deviation of the estimates is 0.033. For the FI detectors alone (for which the QE model is based on the branching ratio technique), the mean is 0.952, and the standard deviation is 0.018. The a priori expected quantum efficiency of the SSD5 at this energy is 0.997 (according to R. Edgar). This comparison is thus consistent with the supposition that systematic errors in the absolute efficiency of our models for the FI detectors are no more than 5%.
We emphasize that a more precise characterization of ACIS systematic errors, especially at energies above 4 keV, awaits absolute calibration of the XRCF BNDs to an accuracy of order 5% or better. The calibration of SSD5 would be the most useful for this purpose.