Reading the failure, not the symptom

Several of these surfaced far from their cause. Worth remembering:

• Longhorn pods crash-looping on one node — the node was memory-starved by hypervisor ballooning.
• external-dns "up to date" while doing nothing — the zone was filtered out.
• HTTPRoute with an empty status — the Gateway controller never started, so nothing had accepted the route.
• "Could not connect" in ~15 ms is a TCP reset: the packet arrived and was refused. A firewall drop times out instead.