Start with the spread, not the average. A run reporting min/avg/max of 24/26/28 ms describes a healthy path. One reporting 24/58/410 ms has the same floor but is queueing badly somewhere, and the average hides it. The statistics line gives you a fourth number, stddev: that is your jitter figure. A few milliseconds is fine; a standard deviation approaching or exceeding the mean means the path is unstable, and anything real-time (VoIP,
remote desktop, interactive SSH typing) will feel bad even while bulk downloads still complete normally.
Packet loss needs context, and the percentage alone is nearly useless. Look at the icmp_seq values or the probe order to see the shape of it. Ten probes with one gap scattered in the middle is 10% on paper, but it is usually a single ICMP packet deprioritised by a router that had better things to do; TCP retransmits it and you never notice. Ten probes where numbers 4, 5, 6 and 7 vanish in a row is a different animal (a link flap, a re-route, or a Wi-Fi roam) and it will visibly stall an interactive session. On Wi-Fi and cellular, occasional isolated loss is routine. Sustained loss on a wired path is a fault worth chasing, and a
traceroute helps you find the hop where it begins.
Read the ttl field too. Hosts start at 64 (Linux, BSD, macOS), 128 (Windows) or 255 (many network appliances), and every router decrements it by one. A reply with ttl=52 came from something that started at 64 and crossed twelve hops: a quick sanity check that you are talking to the machine you think you are, and a hint about which OS family answered.
The first probe is often the slowest. Blame ARP or neighbour discovery, DNS resolution and a cold route cache, not the server. Judge the run from the second probe onward.
Finally, do not read 100% loss as "down". Plenty of hosts, CDNs and edge networks drop ICMP echo by policy. And on the TCP engine the semantics change completely: you are timing a three-way handshake to the port you picked, so every value carries connection-setup cost on top of the true network round trip, and a host whose firewall silently drops that port reports total loss while being perfectly healthy. A closed-port annotation is the good outcome: it means the host answered with a reset.