In the diagram, I write the web server's IP address with a /32 extension, indicating that all thirty two bits of the address must match; i.e, this is a host address, matching a single IP address. While not strictly necessary (an IP address without a trailing prefix is assumed to be a single address), the /32 is written here to illustrate the concept.
The Ethernet containing all of ARS's public web servers is labeled 208.130.29/24, indicating that all IP addresses starting with the twenty four bit prefix 208.130.29 are located on this Ethernet. 18.104.22.168 is, of course, included in this address range.
ARS has been assigned the prefix 208.130.28/22, which includes 208.130.29/24. Why? In binary, 28 is 00011100, while 29 is 00011101. However, because of the 22 bit prefix length, only the first six bits of the third byte are valid. Count: eight bits in the first byte; eight bits in the second byte; six bits in the third byte: 22 bits total; the mask is 255.255.252.0. So, the 28 in 208.130.28/22 should be thought of as 000111xx, since the last two bits aren't part of the prefix. Since 28 and 29 agree in the first six bits, 208.130.29/24 is part of 208.130.28/22.
Continuing along the diagram, we find that 208.128/11 is assigned to MCI. Since 208.130.28/22 and 208.128/11 share their first eleven bits in common, which can be verified by converting 128 and 130 to binary, ARS's address space is part of MCI's. Since MCI is ARS's Internet provider, this should be as much of a surprise as a telephone in Virginia having an area code of 703.