Temperature coefficient and the humidity effect on the LTC6655 voltage output

Between the 11 April and 14 May 2020 I tried to measure the 1 month stability of the LTC6655. During this period I made 80 measurements and the result is shown in the following chart:

Stat	RH %	Tamb °C	V ref, V
Min	24	19.8	2.5003369
Max	54	26.6	2.5003616
Peak to Peak	30	6.8	0.0000247
ppm	0.33ppm/%RH	1.45ppm/°C	9.88ppm

Based on the statistic data, the temperature coefficient was 1.45 ppm/°C and the dependency from humidity was 0.33ppm/%RH. The TC for LTC6655B was below the 2ppm/°C. There is no explicitly given value for the humidity's effect on the voltage output, but was mention that "... humidity sensitivity can be reduced to less than 35ppm for a change in relative humidity of approximately 60%." if PCB slots cuts are made around the LTC6655. This is equal to 0.58ppm/%RH. Because I used small MS8 adapter, I thought that the measured value corresponds to the datasheet value.

However, looking into the chart it is visible that the output voltage is affected more by humidity than by temperature. So I tried to extract measurement data, where the fluctuation of the temperature is relative small and got the next chart, where the dependency is more clear:

According the datasheet, only the LS8 package is not affected by the humidity. I have one LS8, but I do not have PCB to solder on it. So the only option to see what is the real temperature coefficient is to insulate with Fibran XPS the box where the LTC6655 is placed and to fill it with several silica gel packages. This would allow me to keep the humidity in very narrow range.

For period of 12 hours, I was able to keep the relative humidity inside the box in the range of 0.9%, while the external humidity was in the range of 6.87%. The relative humidity and temperature measurements were performed with pair of BME 280 sensors. The result of the measurements are shown in the chart below. It is clearly shows now that the dependency between the temperature in the box and the voltage output is linear. The measured voltage deviation was 5.32ppm for 2.8 °C difference, which makes 1.9ppm/°C within the datasheet specification. 

I wanted to make the same experiment for longer period of time, so the silica gel bags were baked into the oven for about 3 hours in 100°C. In the next 6 days, after 23 measurements, the humidity in the box was changed with 4.3% RH (0.72% per 24h) which is about 3 times better compared to the previous time. The temperature coefficient was not changed so much from the previous time: 1.89ppm/°C for 4.36°C temperature range.

Here an idea popped into my head : if this linearity can be proved for the +/- 5°C and there is no hysteresis within this interval, it would be possible to characterize the LTC6655 for this range and to use the data for an artificial calibration like the HP 3458a and the Advantest R6581 multi-meters. But for that, I have to make PCB for the LS8 package and to make temperature controlled camera.