Surprisingly, Minimum Resolvable Temperature Difference (MRTD) measurements can be made with an accuracy better than the “total system uncertainty” spec of the blackbody. For small incremental temperature measurements made near zero DT and spaced closely in time, measurement accuracy is determined by the temperature noise (short term stability) and linearity of the blackbody, rather than its absolute accuracy.
Recall that MRTD measurements are made by recording the minimum resolvable positive contrast, the minimum resolvable negative contrast, and then averaging the two. This removes any inaccuracies in the reference (target) temperature. Additionally, the MRTD measurement now depends not on the accuracy of the blackbody, but on its linearity (that is, any offset error in the blackbody temperature will eliminated the positive and negative temperatures are averaged. The only remaining error is linearity: how well the slope of the temperature calibration curve matches true temperature).
Temperature measurement linearity for an SBIR 2000 Series Blackbody is better than .0015°C/°C at any point within the blackbody’s range. Linearity is even better near zero DT: on the order of .0010°C/°C. So for small MRTDs (half a degree or less), error due to blackbody nonlinearity is less than the resolution of the temperature readout. The only significant error source is the short term stability of the blackbody system. The data sheet for the 2000 Series Blackbodies specifies short term stability as ±.003°C. This is a conservative spec — the blackbodies easily exceed this level of performance, particularly near zero DT in a stable lab environment. You should be able to make your MRTD measurements with blackbody uncertainty of better than ±3mK.