The group decides the last hop
Once a rule matches, traffic is not handed straight to a node. It goes to a proxy group, and the group picks the line. That is why a node that looks great can still feel slow: often the node is fine and the group policy is wrong.
On the Proxies page, every collapsible heading is a group. The type lives in the proxy-groups block of your config. If the config came from a provider subscription, the interface only lets you switch nodes inside a group, not change its type. Changing types means overriding them with a Merge.
Read the group names first
Providers name their groups very differently. Walk the hierarchy on the Proxies page before you change anything.
How the five types really differ
- 1
select: manual and stable
Whatever you click stays selected, and the choice is stored locally. Keep your main exit group on select so you can switch quickly while troubleshooting.
- 2
url-test and fallback: both automatic, different logic
url-test picks whichever node tests fastest, which keeps latency low but makes it hop around. fallback only moves to the next node when the current one fails a health check, which is stable but can leave you on a slow line for hours. Pick fallback for stability, url-test for speed.
- 3
load-balance and relay: spread and chain
load-balance spreads connections across nodes according to its strategy. consistent-hashing keeps one domain pinned to one node; round-robin cycles through them. relay is a chain: traffic passes through every node in the group in order, so latency adds up. Use it only when you genuinely need a transit hop.
Confirm the group is really in use
Open the Connections page and expand any row, then read the Chains column. It shows the groups and the final node from right to left, for example GLOBAL then your auto group then HK-01. If the group listed is not the one you expected, a rule sent that domain somewhere else. The problem is in your rules, not in the group.
- Keep the main exit on select so you can switch it in one click
- Do not set the interval on automatic groups below 300 seconds; health checks cost traffic too
- Pair load-balance with consistent-hashing so logged-in sessions stop dropping
Common questions
url-test keeps switching nodes. How do I calm it down?
Raise the tolerance value, for example to 150. A new node then has to be at least 150 ms faster than the current one before the group moves. The default is small enough that ordinary jitter triggers a switch.
Will updating the subscription reset the node I picked by hand?
Normally no. Clash Verge stores manual selections locally and reuses them as long as the group name and node name are unchanged. Only when the provider renames a group does it fall back to the first node.