Enable the configuration and note your direct IP

For your first connection, keep three things the same: use the same Wi-Fi or mobile network, the same browser, and the same IP lookup page. Turn off Clash and any other proxy or VPN, then open the lookup page and note your public IP without a proxy. If the page shows separate IPv4 and IPv6 addresses, record each one; compare the same address family after connecting. The local network address shown in your Wi-Fi settings is not the exit IP you need here.

  1. Select a configuration. Using Clash Meta for Android as an example, open “Profiles” from the home screen, select the configuration you just imported, and return to the home screen. Seeing a subscription in the list only means it was imported; it does not mean it is currently active. If several configurations have similar names, check the name and last-updated time first.
  2. Start the client. Start the service from the home screen. The first time the app takes over traffic through a VPN, Android will ask you to approve the connection. Verify which app is requesting access before allowing it. If another app is using Android's VPN connection, stop it first, then try again.
  3. Check which traffic is covered. Make sure the client shows as running and that the browser you're using for the test is included in the VPN's coverage. If per-app filtering is enabled, the browser may be excluded. The VPN icon in Android only confirms that a VPN session is active; it does not prove that a particular website is using the proxy.

Find the policy group used by the rule, then choose a node

Open “Proxies” from the home screen. First, distinguish the three layers: rules decide where a request goes, policy groups choose from available options, and nodes handle the actual outbound connection. Changing the node in a group only affects traffic that matches that group and ultimately uses that node. Group names come from the subscription configuration; names like “Auto,” “Manual,” or a region label don't tell you what kind of group it actually is.

  • select: Manual selection. For an initial check, this type of group is useful for pinning the outbound node so you can tie the result to a specific node. If you select another policy group, follow the chain to see which node ultimately handles the traffic.
  • url-test: Automatically selects a node based on probe results. The probe interval and switching tolerance depend on the configuration; a lower number appearing in the panel does not mean the outbound node switches immediately.
  • fallback: Selects the first available option in the configured order. If an earlier node is still available, a faster result for a later node does not necessarily make it the selected one.

Keep rule mode enabled when checking routing. mihomo evaluates rules from top to bottom and uses the outbound specified by the first matching rule. For example, if the lookup page matches DIRECT, changing the selection in a proxy group won't affect that request. Check the configuration to see which rules may match the target domain, or send a request first and then inspect the client's logs or connection details.

Test latency three times and keep a backup node

Open the target group under “Proxies” and run a latency test for the candidate nodes. The milliseconds shown usually measure how long the core takes to reach a test URL through a node. This is not an ICMP ping to the node address or a download speed test. The test URL, connection setup, TLS handshake, and test server response can all affect the result. Compare values only when using the same network and test target.

  1. Start with three candidate nodes. Keep your phone on the same network throughout; don't switch between Wi-Fi and mobile data while testing.
  2. If the interface allows it, wait about 10 seconds between tests and run three rounds. Record the latency and number of timeouts for each round; don't count timeouts as 0 ms.
  3. Start with a node that succeeds every time and has little variation, then keep another similarly stable node as a backup. Open the page you actually need with your preferred node. For video or large files, check whether they keep loading; probe times can't tell you throughput.

The readings below illustrate how to compare three rounds; they are not real node measurements or universal speed benchmarks. Based on these readings alone, try B first and keep C as a backup. Don't choose A just because it once measured 88 ms.

Candidate node Round 1 Round 2 Round 3 Assessment
A88 msTimed out310 msHigh variation
B142 ms151 ms147 msTest real-world access first
C181 ms195 ms186 msKeep as backup

If all nodes time out at once, first check that your phone can access the internet directly, then try visiting the actual target site through one candidate node. If the request succeeds but probes still fail, check the test URL or probe timeout settings first. If the request also fails, check that the configuration is enabled, whether the subscription needs updating, and whether the connection log reports a DNS or handshake failure.

Verify twice: check the exit IP and the matched rule

First, confirm the public IP used by the test request

Once you've pinned a candidate node, reopen the IP lookup page you used earlier. To avoid reusing an existing connection, you can close and reopen the browser. If the result still doesn't change after switching nodes, use Android's app info page to force-stop the browser, then run the lookup again. Compare the same address family using the same lookup service. A public IP that differs from your direct-connection baseline, together with a record showing the request used the expected node, is evidence that this lookup went through the proxy. Region and ISP labels may vary as databases are updated; rely on the actual address and connection record.

Then check the route used by the site you actually need

With rule mode enabled, open the site you intend to use and note the time and domain. Find the corresponding new request in the logs or connection view, then check the matched rule, policy group, and final outbound. The fields shown may vary by client version. Page loads can also trigger image, video, or API requests to different domains; check those separately. A proxied IP lookup does not mean the target site uses the same group.

  • It shows DIRECT: If the configuration is intended to connect directly, an unchanged IP is expected. If the request should use a proxy, first check whether an earlier rule matched before changing the configuration.
  • It entered the expected group but used the wrong outbound: Check whether the group has selected DIRECT or a nested automatic group. If the node switched automatically, pin a node in a manual group and test again.
  • You can't find the request you just sent: First rule out browser caching and app filtering, then check that the VPN is still running. If your client doesn't show useful connection details, record only the exit IP result you've verified; don't assume which rule matched.

What to check when latency looks fine but a page won't load

Keep the configuration, mode, and network unchanged. Switch only to the backup node you selected earlier, then try the same site again. If the backup works, the issue is likely with the original node or the site's access restrictions for that node's exit IP. If both nodes fail but other sites load, check the failing domain's rules and error details first; don't replace the entire subscription right away.

Pinpoint where the error occurs: for a DNS lookup failure, check DNS; for a connection or handshake failure, use the logs to see whether the problem is with the node or the target site. If the browser returns HTTP 403, a server has sent an HTTP response. Check the domain that returned the error and the page message rather than treating it as a test timeout. Change one setting at a time, then retry the same address.