Official Guide · Protocol Reference

Shadowrocket Protocol Guide

Match the protocol to the server first, then compare speed and battery use. Use this page to check existing settings, understand fields and troubleshoot connection issues. For your first import, server selection and connection, follow the steps in the Quick Start Guide, then return here for protocol or settings questions.

Go to the download page for your device: the iPhone / iPad app is a one-time $2.99 purchase on the App Store; check the App Store listing for system requirements. The Mac (Apple silicon and Intel), Windows, Android and Linux versions are free, with installers available from this site at /dl/. The one-time app purchase does not include a service plan. This site does not provide, sell or recommend proxy servers or subscriptions.

1. Choosing a Protocol: Start with the Server Settings

Identify the connection type from your existing settings

Only protocols and parameter combinations explicitly supported by the server are valid options. In Shadowrocket's Home, check entries imported from an existing Subscribe. If you have server parameters rather than a subscription, select the protocol in Add Server according to the original settings. You cannot identify a protocol from a server domain, port or entry name alone: one domain may have multiple endpoints, and the same port can serve entirely different purposes. A successful subscription refresh only means the subscription URL is reachable; it does not confirm that every listed server can authenticate or forward traffic.

When entering settings manually, group the information into three layers. First, how to reach the server: its address and port. Second, how to authenticate: for example, a password, UUID or key. Third, how the two ends exchange data: transport, TLS, WebSocket path or plugin options. A correct address does not guarantee valid authentication, and successful authentication does not mean the intended destination is reachable. Enter the original settings first, then use the error to identify the layer at fault. Avoid changing several kinds of fields at once.

Protocol lookup for existing server settings
Existing settings or use caseProtocolCheck before connecting
Shared password and encryption methodShadowsocksEncryption method, password and plugin
UUID and specified transportVMess / VLESSProtocol, UUID, transport and security layer
Password and TLS serviceTrojanServer name, certificate and password
Web proxy or general-purpose proxy endpointHTTP / SOCKS5Authentication method, encryption boundary and destination scope
Device peers and subnet accessWireGuardPeer public key, Endpoint and AllowedIPs
Explicitly provided QUIC endpointHysteria2UDP connectivity, certificate and authentication method

Narrow down options based on traffic needs

Start by identifying what you need to access. For everyday web browsing, look at time to first response, how reliably pages load, and whether small requests remain stable. For large transfers, also check sustained throughput. For internal network access, reaching the required subnets and using internal DNS may matter more than benchmark results. For voice, real-time sync and other services, first confirm that the protocol, server and client all support the required UDP forwarding, then test the actual service. A webpage loading does not validate all traffic types.

If the service provider supplies only one configuration, choosing a protocol means selecting the correct type in Add Server—not changing the connection to another protocol. Comparison is worthwhile only when you actually have multiple working endpoints. Two entries using the same protocol can perform very differently because of server load, network path or destination. Conversely, perceived differences between protocols on the same device may mostly reflect those conditions. Unless the provider confirms that the two configurations have comparable conditions, limit your conclusion to which one works better on your current network.

Distinguish connectivity, functionality and suitability

Connectivity means the device can exchange data with the endpoint. Functionality means the intended traffic goes through that endpoint and gets a valid response. Suitability means it continues to meet expectations in everyday use, while the device is asleep and when networks change. For example, a successful Connectivity Test only confirms that one test succeeded; it does not prove that background sync or network switching works. Start with a known destination, note what failed and under what conditions, then expand testing to the traffic you actually need. This approach reduces unnecessary setting changes and makes issues easier to reproduce.

When recording the original settings and test results, note the protocol, whether authentication succeeded, the network used and whether the network changed. Do not copy the full configuration into a public report. A subscription URL may contain access credentials, as can a QR code. Never use real passwords or keys in screenshots of example settings. When asking the service provider for help, describe the failure stage and share field names with credentials removed. See the glossary for interface terms, or the FAQ for purchase, installation and import questions.

2. Shadowsocks: Shared Password and Encryption Method

Design and use cases

Shadowsocks is designed to establish an encrypted connection between the client and a remote proxy using a small set of parameters, then forward traffic from the remote end to its destination. Unlike a tunnel for routing network segments, it does not require assigning a tunnel address to the device. A typical configuration starts with the address, port, encryption method and shared password. If the server specifies a plugin or other options, those four basics are only the starting point; they do not replace the additional requirements.

Encryption between the client and proxy protects only that connection. Whether the destination app has end-to-end protection still depends on the app itself. For example, a browser's TLS connection to an HTTPS site provides a separate layer of protection. If the original service uses a plaintext protocol, the connection from the remote proxy to the destination is not encrypted just because you selected Shadowsocks. To assess whether a setup is suitable for sensitive data, check both the service protocol and the trust boundary of the proxy endpoint—not just the name of the encryption method.

Check the password and encryption method together

Open the existing entry and confirm the full encryption method name against the original settings. Similar-looking suffixes may use different key formats or packet handling, so the two ends must match exactly. In traditional configurations, the password may be a shared string; some newer methods require key material of a specific length and encoding. If a credential looks encoded, do not decode it yourself, add spaces or remove trailing characters. After pasting, check for line breaks or full-width punctuation, and make sure you have not copied the password from another entry.

If the original settings specify a plugin, distinguish its options from the standard encryption fields. The server address identifies the endpoint; a domain or path used by the plugin transport belongs to another layer and should not be entered as the server address or password. Compare an unchanged entry from the same subscription with your manually entered one first. If the original works but the manual entry does not, check for differences in how the fields were interpreted. If both entries fail, ask the provider whether the endpoint, port or settings have changed. Do not infer whether a plugin is needed from a familiar port number.

A working webpage does not confirm UDP support

A webpage test usually cannot confirm that UDP works over the same path. If ordinary pages load but a service such as voice does not, first check whether the server has enabled UDP forwarding and whether the current routing policy sends that traffic through this entry. Repeat the tests on the same network: first try an ordinary request with a clear response, then test a service that actually requires UDP. If both fail, recheck the endpoint, encryption method and password. If only the UDP-dependent service fails, focus on UDP support and routing; there is no need to change authentication settings that are already working.

Measure speed and battery use with real traffic

Fewer configuration layers can make it easier to avoid mixing up fields, but they do not guarantee low latency or longer battery life. On iPhone and iPad, short text requests are more sensitive to round-trip time, while long transfers use both encryption processing and wireless resources. Even the same encryption method can behave differently across devices due to hardware acceleration, concurrent requests and device temperature. If the phone gets warm, pause the comparison. Test again under the same screen, network and battery conditions so you do not mistake heat-related changes for protocol differences.

For intermittent browsing, repeated success when opening the same kinds of pages—and whether the first request works after the screen has been locked—matter more than a single peak-speed result. If short requests work reliably after connecting but large downloads slow down, retest other known-good entries against the same destination and at a similar time. This helps rule out destination-side throttling or network congestion. Compare encryption methods or transport combinations only when the server conditions are also similar.

3. VMess and VLESS: Identity, Transport and Security Layers

The same identity credential does not mean the same protocol

VMess includes identity verification and data protection in its protocol design, and commonly uses a UUID to identify the user. VLESS is designed to do less at the protocol layer, with traditional configurations generally relying on an outer security mechanism to protect transport. Both may have UUID, transport and security fields, but their handshake formats and server-side handling differ. If an existing entry specifies VMess, configure it as VMess. Do not switch the type to VLESS just because you have a UUID and the address and port look the same.

A UUID here is neither a display name nor a certificate. Preserve its original characters and separators; do not regenerate it based on the device name. Some VMess handshakes also require the clocks at both ends to be within an acceptable range. If several previously verified VMess entries start failing authentication at the same time, check that automatic date and time are enabled on the device and ask the provider to confirm the server clock. Changing the displayed time zone alone usually will not fix a clock offset. Preserve any legacy compatibility fields in older settings; do not copy values from other entries.

Check the endpoint, transport and outer security separately

With WebSocket transport, the server address, TLS server name, HTTP Host and Path may each serve a different purpose. The address determines where to connect. The server name is used for the name indication and identity check during the TLS handshake. Host and Path can tell a reverse proxy which backend should receive the request. These values may be the same, but sometimes they must differ. A domain can resolve and its certificate can be valid, yet a mismatched WebSocket path can still leave the endpoint reachable while preventing the data connection.

The transport may also be plain TCP or another option specified by the provider. Changing the transport is not just a performance preference; it often changes what the server expects to receive first. For additional Flow, security type or outer-handshake settings, follow the explicit instructions for that entry. Do not assume the current endpoint accepts TLS on the same port just because another configuration uses TLS. Nor does a domain in the address guarantee that the certificate is valid for that name.

Troubleshoot the earliest symptom first

If the server connection times out, start with the address, port, current network and what changes when you switch networks. If you see a certificate or name warning, check the server name, device time and certificate status first. If the endpoint is reachable but apps cannot access their destinations, check the protocol, UUID, transport path and whether the server still has this configuration enabled. Logs do not always distinguish the layers clearly, so avoid guessing at a plausible-sounding cause. Keep one unchanged entry, change only one field at a time, and compare the results.

If only a few entries in a subscription fail, check the transport settings for those endpoints first. If all entries in the same group fail at once, look for a shared cause: whether the subscription actually refreshed, whether the system clock was correct, and whether changing networks resolves the issue. Do not overwrite an old entry with settings from a similar-looking entry in a new subscription. When multiple endpoints share a domain, their paths, authentication and security settings may be what distinguish them.

Compare first response time and sustained throughput

An outer security handshake and additional transport can add network round trips when a connection is first established. Connection reuse and the number of short requests also affect the experience beyond peak speed. On iOS, frequent disconnects and reconnects cost energy both to process handshakes and to wake the wireless radio, so do not infer battery use from protocol overhead alone. Test just after connecting, on repeat visits and after unlocking from sleep. Record when things slow down. If only the first visit is slow, focus on name resolution and handshakes; a large-file speed test cannot explain a delay on the first request.

You can compare two protocols on your device only when the same service explicitly offers both and their endpoints, network paths and loads are sufficiently similar. Even then, frame the result as how a particular transport and security configuration performed on a particular network—not as proof that one protocol is faster in all conditions. For everyday use, settings that can be verified, restored after a failure and used to complete real requests reliably matter more.

4. Trojan: TLS Identity Checks and Proxy Authentication

What TLS and the proxy password each do

Trojan's basic design carries a proxy session over a TLS connection, then uses protocol-level credentials to verify that the client can access the proxy service. This separates server identity from proxy account authentication into two steps: TLS checks the server name and certificate chain, while the proxy password determines whether the connection can use the proxy. A valid certificate does not mean the password is correct. A correct password cannot make up for a name mismatch or an untrusted certificate. Keep the original verification settings in place while identifying which stage fails.

Some configurations connect directly by domain; others specify the network address and TLS server name separately. The actual address may point to a specific host, while the certificate covers a domain provided by the service. Keep the values separate only when the original settings say to do so, and do not copy an ordinary website domain into the server name field. SNI indicates the requested server name during the handshake; it is not the same as HTTP Host. Even when the text values happen to match, do not treat the fields as interchangeable.

Keep certificate verification enabled to confirm the server identity

If you see a certificate warning, first check that the device date and time are correct, then compare the server name and destination address with the settings. If the connection used to work but the same warning now appears on multiple networks, ask the provider to confirm that the new certificate is deployed on the actual endpoint, covers the expected name and includes the complete certificate chain. Do not bypass verification just to make one connection succeed; doing so cannot confirm that the device reached the correct server.

On public Wi-Fi, the device may detect the wireless network but still be unable to reach external services until you complete the network sign-in. If the issue occurs only on that network, complete its normal sign-in first, then retry with the original settings. If switching back to a known-good network fixes the issue immediately, record the network conditions rather than changing the password. If TLS succeeds on every network but proxy traffic still fails, focus on the password and the transport actually enabled by the server.

Check the actual connection, not the port number

A server offering a default TLS endpoint does not mean you can connect by entering a commonly used HTTPS port. Conversely, using another port does not mean you can skip certificate verification. If a subscription entry includes specific transport fields, use the complete set of supplied settings rather than copying just the address and password. To verify the setup, confirm the selected server in Home, then make an ordinary request that clearly requires the proxy path and check that the expected content is returned. A webpage loading on its own does not prove that proxy authentication succeeded.

If a test works once but the service is intermittent, note whether failure occurs on the first visit, after a long idle period or in the middle of a continuous transfer. For the first two, check DNS, connection recovery and the endpoint's idle timeout. For the third, also note network changes and destination load. When comparing failures, record whether the connection dropped, the response was delayed or the content was wrong. Calling all three “a slow protocol” makes it difficult to identify whether the delay occurred during connection setup, forwarding or service processing.

Compare recovery costs on iOS

Reusing an active connection and frequently establishing new ones can have different battery impacts. Network changes, background state and the endpoint closing idle connections can all require the device to establish a new session. If your typical use involves brief page checks while moving around, include a Wi-Fi-to-cellular switch and a check after unlocking back to Home. If you mostly transfer large files on a fixed network, focus on sustained throughput, retry costs after a failure and device temperature. Do not combine these scenarios into a blanket claim that one setup is “more efficient” or “faster.”

5. HTTP and SOCKS5: Proxy Interfaces and Encryption Boundaries

HTTP: distinguish the proxy endpoint from the website connection

HTTP proxies were designed to forward web requests. For plain HTTP requests, the proxy can receive and forward the request. To access an HTTPS site through a proxy, the client typically sends a CONNECT request to create a tunnel to the destination, then performs the TLS handshake with the website inside that tunnel. Whether the connection to the proxy itself uses TLS is a separate question. An HTTPS URL does not mean the device-to-proxy connection is also protected by TLS, nor does it mean proxy authentication fields receive the same protection automatically.

If an existing organization proxy requires a username and password, check the original settings for whether authentication is required, whether the endpoint supports TLS, and which destinations and ports are allowed. A successful proxy response does not mean every site is accessible; an organization may allow CONNECT only on certain ports. If you receive an explicit proxy authentication challenge, check the account and authentication method. If only certain destination ports are rejected, check the proxy's allowed scope. Do not assume access to other destinations just because a few sites work.

SOCKS5: forwards connections without built-in general-purpose encryption

SOCKS5 provides a more general-purpose proxy interface than web-specific requests, without requiring the proxy to understand the application data as HTTP. It can negotiate authentication and forward TCP. The specification also defines UDP forwarding, but actual support depends on the client, server and network path. The protocol itself does not mean that traffic is encrypted. In particular, do not treat ordinary username-and-password authentication as an encrypted tunnel on an untrusted network. If additional protection is needed, use a security layer explicitly included in the existing settings and verify that it works.

Distinguish the proxy address from the destination address: the former lets Shadowrocket reach the proxy; the latter is where the app is trying to connect. Which side resolves DNS depends on whether the app gives the proxy a domain or an IP address, as well as the DNS settings and routing conditions. For example, if an internal domain must be resolved by an organization DNS server, do not assume SOCKS5 always makes the server resolve it. Nor should you replace an internal domain with a fixed IP as a workaround for DNS issues.

Use the failure scope to distinguish protocol, access and routing issues

If the endpoint is reachable but negotiation fails immediately, check that the protocol type and port match. A common mistake is to set the proxy type based on the protocol used by the destination website. For example, the destination may be an HTTPS site while the organization provides a SOCKS5 proxy; use the proxy type specified for the endpoint, not the one shown in the browser's address bar. If only UDP-dependent services fail on the same endpoint, check whether SOCKS5 UDP forwarding is enabled. A working TCP webpage does not prove that it is.

If an HTTP proxy opens common websites but one service fails through that endpoint, compare two allowed destinations on the same network: one that worked before and the failing service. If both fail, continue checking the proxy connection and account. If only the service destination fails, check its port, DNS resolution and the proxy's allowed scope. If the destination works after switching to Direct, that only confirms that the direct path worked at that time; it does not prove the proxy password is wrong or the organization policy is at fault.

Higher speed does not necessarily mean a better fit for mobile use

On a managed network, a simple proxy endpoint can be easy to verify for ordinary web traffic that already uses TLS. On cellular data or public Wi-Fi, reassess whether the endpoint is reachable, whether the proxy connection itself is protected and how exposed the authentication credentials are. When comparing protocols, use the same destination and record first response time, retry count and background recovery. Even with little protocol overhead, repeated timeouts or authentication failures can prolong requests and keep the wireless radio active longer.

In practice, choose HTTP for an explicitly provided web proxy and SOCKS5 for an explicitly provided general-purpose proxy, then verify each required destination and traffic type. If your work depends on reaching a whole address range or services that rely on internal DNS, first confirm that the existing settings provide access to those networks. Do not treat a web proxy as a general-purpose network tunnel.

6. WireGuard: Peer Keys and Network Routes

Think in terms of network-layer routing, not individual connections

WireGuard is designed to carry IP packets between devices or networks. It identifies peers with keys and exchanges encrypted packets over UDP, rather than requiring each app to submit proxy credentials for every connection. Existing settings usually include the device's tunnel address and private key, the peer's public key and endpoint, and the address ranges allowed through that peer. A server domain or ordinary proxy password alone is not enough to establish a WireGuard peer connection.

Endpoint identifies the peer's actual network address, usually a hostname or IP and UDP port. The device's tunnel address identifies it inside the tunnel; the two are not interchangeable. Keep the private key in your own configuration; the peer's public key identifies the other end. Before copying settings, check whether each field belongs to the local device or the peer. Do not overwrite the current device's private key with the one for a backup device. If the settings include a preshared key, both ends must match; it cannot be replaced with the peer's public key.

AllowedIPs determines which traffic enters the tunnel

AllowedIPs uses address prefixes to describe the ranges permitted through a peer. For outgoing traffic, it helps determine which peer should receive the traffic. For incoming decrypted traffic, it also checks whether the source address falls within that peer's allowed range. When accessing a private subnet, keep the range specified by the provider. Expanding it to cover all addresses will not fix a missing return route or an internal service problem, and may redirect traffic that should use the local network.

When troubleshooting, separate the tunnel handshake, destination address and DNS resolution. If no usable handshake is established, check the Endpoint, UDP path, peer identity and whether the server has the device's public key configured. If the peer exchanges data but an internal destination is unreachable, confirm that the target IP is within the expected range and that the peer has a route forward and back. If only internal domains fail, check DNS settings and the resolved address. A successful handshake alone does not prove the path to the internal network is complete.

Balance roaming and keepalives for your use case

When a device changes networks, its outer source address may change. WireGuard can update the observed endpoint based on authenticated packets from a peer, but the new network must still allow the required UDP traffic, and apps need to resume communication after the switch. PersistentKeepalive periodically sends keepalive packets and can help a peer that needs to receive inbound traffic maintain certain NAT mappings. It is not a general-purpose speed setting. If you are unsure whether it is needed, keep the existing value and check whether recovery fails only after the connection has been idle.

A shorter keepalive interval sends packets more often and may wake the wireless radio even when idle. Do not shorten the interval just because one network switch was slow. Test fixed Wi-Fi, cellular and the transition between them separately. If fixed-network use works but the switch fails, check whether the new network allows UDP traffic to that endpoint. If the connection also fails after sitting idle without a network change, check the server keepalive policy, network mappings and device background behavior.

Small requests work, but large transfers stall

Tunnel overhead uses some of each packet's available space, so on certain paths small requests may work while larger transfers intermittently stall. Path MTU is one setting worth checking, but it is not the only possible cause. Also rule out destination throttling, congestion, asymmetric return paths and network changes. Compare a small request with a continuous transfer on the same network, then note which network reproduces the problem. Adjust MTU only if the configuration provider gives a recommendation and the symptoms appear related to transfer size; then repeat the test.

If you need to reach existing internal services, verify practical tasks first: whether internal domains resolve, whether their addresses are reachable, and whether other apps still use their expected routes after returning to your everyday network. For this use case, clear routing scope and reliable recovery from idle are more useful than a single public internet speed test.

7. Hysteria2: QUIC, UDP Paths and Bandwidth Settings

Requires a working endpoint and UDP path

Hysteria2 is designed for proxy transport over QUIC. QUIC runs over UDP and combines encryption, congestion control and management of multiple streams. Separate streams can reduce the effect of packet loss in one stream on the delivery order of others, but they still share the network path and congestion-control resources. Retransmissions, queueing and link quality continue to affect performance. QUIC identifies the connection mechanism; it does not guarantee higher speed than other protocols.

Choose Hysteria2 only if the existing server explicitly provides Hysteria2 settings. Check the protocol generation too; similar names do not make protocols interchangeable. First verify the server address, port, authentication method and TLS server name against the original settings. A webpage loading or a TCP request working smoothly does not confirm that the specified UDP port on the same host is reachable. If the endpoint and port are correct but one Wi-Fi network consistently fails, cross-test on another trusted network. That usually provides more useful information than trying random ports.

Handle authentication, certificates and bandwidth settings separately

Use the authentication details supplied in the existing server settings; credentials used to refresh a subscription are not a substitute for proxy authentication. Check the TLS server name for each entry. Even with a numeric address, the provider may require a specific domain for verification. If you see a certificate warning, keep verification enabled and check the system time, configured name and server certificate. If switching networks fixes the problem immediately, note the UDP path difference. If the endpoint is reachable but authentication fails on multiple networks, recheck the credentials in the original entry.

If the settings include upload or download bandwidth values, check their units and direction, and use values supplied by the provider or otherwise verified. These are not a display of purchased bandwidth or hardware network speed. Depending on the implementation and settings, the values may affect sending behavior and congestion-control choices. Values that are too high can increase wireless congestion, retransmissions and queueing; values that are too low can limit throughput. When network capacity varies, keep the existing defaults and use real traffic to check for sustained congestion rather than tuning the settings to a single speed-test peak.

Record symptoms by the stage where failure occurs

If a connection test times out completely, compare the same endpoint on trusted Wi-Fi and cellular, then confirm that the server is still listening on that UDP port. If the symptom is a certificate warning, check the name and certificate rather than increasing bandwidth. If small requests work but long transfers fluctuate, note whether the network changed, signal weakened or the destination was throttling traffic. If other known-good endpoints slow down against the same destination, first check the device's network and the destination service.

Distinguish failures after idle from interruptions during transfer. For the former, use a fixed network to compare short and longer idle intervals, then check the first request after the device resumes. For the latter, monitor a continuous transfer for repeated stalls and recovery. A single Connectivity Test cannot distinguish these cases. Before changing a setting, record its original value, network and symptoms. After restoring it, verify that the improvement is repeatable rather than relying on one successful test.

Choose for iOS based on side-by-side testing

For long downloads or sustained traffic on a variable connection, Hysteria2 can be included in a comparison. Use the same device and destination, with similar screen-on time, background sync and network type. If retransmissions and probing continue for a long time on a particular network, a higher peak speed may still mean more heat or battery use. Conversely, a configuration that completes tasks reliably with fewer retries may take less time overall. Compare task completion, recovery from idle and battery trends; encryption processing and wireless activity are not the only sources of power use.

For brief browsing on a fixed network, an existing stable endpoint does not become a poor choice just because another protocol uses QUIC. If your use includes switching networks, compare success rates and recovery time on both networks before single-session speed. Only compare sustained throughput after both configurations meet your reliability requirements.

8. iOS Testing: Tell Protocol, Rules, DNS and Battery Use Apart

Start with a repeatable test sequence

In Home, confirm the selected server, connect, then run a Connectivity Test. See First Connection and Connectivity Test for where to find it. Test a fixed webpage first, then a service you use regularly. Test separately if the service depends on an internal address or UDP. A test result reflects only the conditions of that probe; it does not show that all traffic follows the same policy. An enabled connection switch only means the connection feature is on—it does not guarantee that every new request gets a response.

In Global Routing, Config applies the rules in the current configuration. Proxy is useful for a short comparison of traffic sent through the proxy. Direct checks the connection without the proxy. After switching modes, send a new request; do not count a cached browser page as a valid result. Compare the modes only with public test destinations that are allowed to connect directly. Do not send traffic that should use an internal route out to an external network. Restore your original mode when testing is done.

Narrow down the issue with three modes
ModeWhat it helps testWhat it cannot prove
ConfigCheck whether rules and policies behave as expectedA reachable destination does not prove that a proxy rule matched
ProxyCheck whether the selected endpoint handles a test requestA successful comparison does not prove the original rules are correct
DirectCompare a direct connection to a public test destinationA successful comparison says nothing about server-side authentication

Check rules separately from the selected protocol

The protocol determines how Shadowrocket communicates with the upstream server; rules determine which policy applies to a matching request. The snippet below illustrates matching order. It is not a complete configuration and does not need to replace your current rules. example.com is a sample domain. If a destination matches GEOIP,CN,DIRECT and no earlier rule matches, it uses DIRECT. The FINAL rule handles all remaining traffic. Put more specific domain rules before the fallback so you can tell whether they have a chance to match.

[Rule]
DOMAIN-SUFFIX,example.com,PROXY
GEOIP,CN,DIRECT
FINAL,PROXY

If a destination works in Proxy but fails in Config, first check which Config is enabled and which domain the destination actually uses. Then look for an earlier matching rule or a different domain serving the page's resources. If a rule requires resolving the destination to an IP address, also check DNS results and resolution conditions. A domain resolving does not mean its destination port is reachable. Conversely, opening an IP directly may fail because the website relies on a server name, so it is not an equivalent test for accessing the domain.

If a public test destination works in Direct but consistently fails in Proxy, focus on the endpoint and protocol settings. If it fails in both modes, check the device network and destination service first, then DNS. If only one set of rules behaves unexpectedly, note the matched policy and selected entry instead of reimporting the entire subscription. For a step-by-step check, follow Troubleshooting: Connected but No Internet Access to review the server, mode, DNS and rules.

Test speed, recovery and idle battery use separately

For a speed comparison, use the same destination for short requests and continuous transfers. Alternate between the two configurations at similar times instead of ranking them by a single peak. Test recovery separately: record the first successful request after unlocking, or after switching between Wi-Fi and cellular. For battery use, compare system battery usage over similar periods with screen-on time, background sync, test tasks, signal and temperature kept as consistent as possible. System readings reflect the device's actual workload; they cannot isolate the battery use of one protocol.

When including On Demand in a comparison, make sure both tests use the same trigger conditions. If one configuration connects only on a specified network while the other stays connected all day, their idle usage cannot be attributed directly to protocol differences. If idle battery use rises, check for frequent reconnects, repeated refreshes, persistent keepalives or weak signal. Change one factor at a time and retest with the same tasks. Keep the configuration that reliably handles your usual traffic, makes its routing behavior clear and has a verified recovery procedure.

Download Shadowrocket