Hi again - very helpful information!
Here is the list of HOLU's where I noticed the -999;
it comes from May 11th and October 12 data snapshots (click for access, the -999 is coloured yellow in the NO2 column) I once did using the https://data.waag.org/api/holu/holukit/calibrated/recent API.
May 11 has these ID's with a -999: 87, 84, 83, 81, 79, 77, 66 and 62. All Basic HOLU's!! Nevertheless a NO2 -999.
October 12 same story and for the most part the same ID's: 87, 84, 83, 79, 77, 74, 66, 61; all Basic HOLU's with a NO2 -999.
October 12 also has -999 for ID's 193 and 191 which areExtended HOLU's. Given your explanation this is odd?!
Lastly: if I understand you correctly, you are explaining that in the code running in the HOLU there is "no filter in case of unreasonable data"; that sounds like a good approach: have the data out of a HOLU as raw as possible and decide later / downstream what is used or not in analysis. Question to be sure ( @denis too): is the HOLU code inserting a -999 or some step later in the data infrastructure?