I've got some pretty good data on the AEMC CA813's accuracy.Methodology
I measured from 10% to 100% stimulus patterns using a spectroradiometer (Orb Optronics SP-100). The display was a plasma. Next, I measured the same patterns using the CA813 by placing the reading head directly against the screen. By comparing the two I could determine whether the relationship between the two instruments remained constant or varied.
If it were constant, then a single comparison could be used to create an offset that would allow accurate measurements for all other levels using the CA813. This would also be strong evidence that the CA813's response was linear across a wide range of measurement.
Here are the results.
As you can see, the CA813's response is very constant from approximately 1 lux to 200 lux.
One unusual aspect of this test is that I compared illuminance readings to luminance readings. However, once the offset between the two has been established, I see no problem with this.
From this data I am confident that the AEMC CA813 can be used within the measured range with confidence that its readings are accurate. Whether you could go below 1 lux and maintain that accuracy I cannot say. My reference instrument won't read any lower.Edit:
Although I cannot read lower using the reference spectro, I can use the Chroma 5 colorimeter to read considerably lower. I have also tested that against the spectro at higher luminance values and determined that its luminance readings are spot on. When I read the black level on the plasma using the Chroma 5 I get a consistent 0.06 cd/m2 or 0.018 fL. Using the CA813, I measure 0.11 lux. Divide that by the conversion value of 1.8 established above and you get 0.06, precisely the same luminance value obtained by measuring directly with the colorimeter.
From this I conclude that the AEMC CA813 is accurate down to at least 0.1 lux.