I even find “goes through” difficult in this context. I understand that’s not what you’re implying, but one cannot interpret the numbers for SF12 to indicate that SF11 and better would not have “gone through” for those. (Unless you know ADR was used, or at least know that the number of ADR-enabled devices is evenly distributed over the SFs in that list.)
It may be interesting to see how many of those did use ADR, and on how many of that city’s gateways a packet for each SF was received on average, regardless ADR. (And even then there’s more to know, like how many of the SF12 devices are ABP or pre-October 2019 OTAA devices that do not get ADR MAC commands due to TTN erroneously using SF12 for RX2 when one confirmation of a MAC command failed.)
Agreed. 
No, there’s no blocking in place. But after your testing you’re not allowed to fix your device to use SF11 or SF12. And see also Is there a correlation between payload length and potential packet loss?