Shadowrocket Official Manual / Protocol Reference
Shadowrocket
Protocol & Settings Reference
Choose a Type based on your server details, distinguish authentication, transport, and routing, then compare speed and battery use.
This guide focuses on configuration and performance on iPhone and iPad. The protocol concepts also apply to other Shadowrocket platforms. Available fields vary by platform; check the current Add Server screen and your existing service configuration.
01 / Before you choose
Protocol, transport, and routing: what each controls
Choosing a protocol in Shadowrocket does not let you switch modes freely on the same server. The server must already be running the selected service and provide authentication and transport settings that match the client. If your server details specify Shadowsocks, changing Type to Trojan will not make it faster. Even with the same address and port, the initial data exchanged by each side will differ. First identify what the server actually offers; then compare the available options.
Start by separating the connection into three layers
The first layer is the proxy or tunnel protocol: it defines how requests are represented, how identities are authenticated, and how data is encapsulated. The second is the transport and security mechanism carrying it, such as TCP, UDP, TLS, or WebSocket. The third is the client’s routing rules, which decide which requests use the connection. Keeping these layers separate helps explain why “authentication succeeded, but websites still won’t load” and “websites work, but a UDP app fails” are different problems.
For example, a VLESS user ID is not the same as transport encryption, and a TLS server name is not the same as a WebSocket Host. They may use the same domain, but serve different purposes. Likewise, Config, Proxy, and Direct in Global Routing are routing modes, not server protocols. Switching to Proxy can help identify a routing issue, but it will not fix an incorrect password, certificate name, or server port.
| Layer | Key question | Typical fields or settings |
|---|---|---|
| Protocol & authentication | Are both sides using the same data format? | Type, Password, UUID, encryption method |
| Transport & security | How does data reach the server? | TLS、SNI、WebSocket Path、UDP |
| Rules & routing | Which path should a request use? | Global Routing、DOMAIN-SUFFIX、FINAL |
Speed and battery life depend on the conditions
Throughput depends on wireless signal, round-trip time, packet loss, server load, the destination, and the device’s processing capacity. Protocols can change handshake counts, retransmission behavior, and encapsulation overhead, but they cannot remove the physical link’s bandwidth limit. Battery use is not determined by encryption alone: sustained transmission over a weak signal, frequent keepalives, and repeated handshakes can matter more than algorithm differences. For a meaningful test, keep the device, network, destination content, and test duration constant. Don’t generalize from a single speed test.
This guide does not rank protocols as “fastest” or “best for battery life.” If the server offers only one type, configure it correctly and look for bottlenecks. If it offers several, compare them using the same workload. Short web requests are sensitive to connection setup delays; sustained downloads depend on steady throughput; calls and interactive apps also depend on jitter and the UDP path. The best choice may differ by use case.
Getting the app and server details are separate steps
Shadowrocket is available for iPhone, iPad, Mac, Windows, Android, and Linux. The iPhone and iPad app costs $2.99 as a one-time App Store purchase. The Mac (Apple silicon and Intel), Windows, Android, and Linux versions are free. See the download guide for download options and first-launch instructions. Desktop and Android installers are available from this site at /dl/. Check the App Store listing for iPhone and iPad system requirements.
Before troubleshooting, save the original configuration and note where each setting came from. Change one variable at a time. If you switch Type, DNS, transport, and routing mode together, even a successful connection won’t show which change fixed the problem.
02 / Encrypted proxy
Shadowsocks: the method and password must match
Shadowsocks was designed to relay proxy requests with relatively little overhead while encrypting traffic between the client and server. It usually does not require a domain certificate for each service, so deployment and settings are fairly straightforward. But fewer fields do not mean fewer checks: the server address, port, encryption method, and password must all match. A mismatch in any of them can leave the connection with no useful response.
The encryption method is not a client-side performance switch
Common methods include options named AES-GCM or ChaCha20-Poly1305. The method name describes the key and authenticated encryption used by the protocol; it is not a reason to choose based on “higher numbers are faster” or “newer names are better.” Use the method configured on the server. Hardware acceleration affects processing costs only when the client and server already support the same method.
Some Shadowsocks variants also differ in key format and protocol generation, so AES or ChaCha20 in the name does not mean they are interchangeable. If an imported method is not recognized, check the fields supported by your current client and the server’s requirements. Don’t replace an unfamiliar method with a similar-sounding one. Where a configuration requires an encoded key, that key is not the same thing as an ordinary password.
Copy the Password exactly as provided. Leading or trailing spaces, punctuation changed by an input method, or a missing character will change the result. Reformatting a long string to make it look neater can easily introduce errors. If the subscription imported correctly, there is no need to edit the password manually. A share link may encode the whole set of details or only the authentication portion. If parsing fails, refer to the original service instructions rather than repeatedly trying different decodings.
Check TCP and UDP separately
Shadowsocks can carry TCP and, when supported by both sides, UDP. The network conditions for the two paths may differ. A working website proves only that some traffic succeeds. If calls, games, or certain DNS requests fail, check whether the server enables UDP forwarding, whether the port allows that traffic, and whether an intermediate network is dropping datagrams. The UDP switch is not a universal fix.
If the service uses an additional transport extension, its name and settings must also match. A basic Shadowsocks configuration does not automatically include WebSocket or TLS. Adding an extension’s path to an ordinary address field will not work. First identify the base protocol, then check whether an extension is in use and whether the client provides the corresponding fields. Don’t add Host, Path, or certificate settings unless the service instructions include them.
| Setting | What it means | Common mistake |
|---|---|---|
| Server / Port | Actual server address and listening port | Using the admin page address as the server address |
| Method | Encryption method matching the server | Changing the method based on personal preference |
| Password | Information needed for authentication and key derivation | Missing a character or adding spaces when copying |
| UDP options | Allow the relevant traffic over the datagram path | Enabling it in the client without checking the server |
Understanding overhead on iOS
With a simple transport and a stable connection, fewer encapsulation layers can reduce processing complexity. Actual battery use still depends on traffic volume and wireless conditions. Compare both bursts of short requests and sustained downloads on the same network to see the difference between connection setup and long transfers. If battery use is noticeably higher only on cellular, check signal strength, retransmissions, and screen usage before blaming Shadowsocks.
Include Shadowsocks in a comparison only if your server explicitly supports a compatible method and you have verified TCP and UDP as needed by your apps. If the connection closes immediately, check the method and password first. If it waits and times out, check the address, port, and reachability. Logs can narrow down the cause, but silent packet drops may make an authentication error look like a timeout. Check the server logs too.
03 / Separate authentication from transport
VMess and VLESS: settings beyond the UUID
VMess combines user authentication with proxy data encapsulation. Its settings commonly include a UUID, security options, and a transport method. VLESS uses a more minimal protocol design and relies more on an outer layer for transport security. Their names and some fields are similar, but they are not interchangeable protocols that can be made to work by changing Type alone. The server configuration must specify the protocol, transport, and applicable security requirements.
A UUID covers only part of identity verification
A UUID is a user identifier, usually written as a hyphenated string. It is neither the server address nor a certificate name. Entering the UUID correctly does not complete the connection setup: also check the port, transport, TLS, and other server requirements. When entering settings from written instructions, it is easy to copy the most prominent UUID and miss the transport path that delivers data to the server process.
VMess authentication depends on time, so a significantly incorrect system clock can cause failures. Set the device to update the time automatically and check the server clock. Older configurations may contain legacy compatibility fields; do not overwrite current service settings with a fixed value from an old guide. Security options must also match the server’s requirements. If import tools use different labels for the same field, compare what the settings actually mean rather than their display order.
VLESS does not provide general payload encryption by itself. Successful UUID authentication does not mean the connection is encrypted. Check separately whether TLS or another explicitly configured secure transport is in use. Some deployments also require a flow setting or other conditions. Do not choose these fields arbitrarily or copy them from a different server. Enter them only when the server specifies them and the current platform supports them.
WebSocket, Host, Path, and SNI each have a different role
WebSocket is a transport method, not another form of user authentication. When enabled, the connection uses an HTTP upgrade request. Path identifies the server endpoint and usually needs its leading slash. Whether capitalization, a trailing slash, or a query string matters depends on server routing; don’t omit them without checking. Host affects virtual host selection at the HTTP layer, while SNI identifies the server during the TLS handshake.
The address can be a domain or IP used to establish the network connection, while certificate validation typically relies on the name associated with the service identity. They can differ, but must match the deployment design. If TLS fails, don’t leave certificate verification disabled just to make a test pass. Check the device time, SNI, certificate names, and certificate chain, then confirm that the connection is reaching the right port and frontend service.
Connection reuse and the iOS experience
Multiple transport layers can add connection setup steps, but may also help manage connections in some deployments. Whether to enable connection reuse depends on client support, server implementation, and the workload. Reusing one underlying connection can avoid repeated handshakes, but packet loss or congestion on that connection can affect every request sharing it. More reuse is not always better, and a fixed value found online is not necessarily right for every server.
On iOS, frequent short page loads can feel different from continuous playback. The former is more affected by handshake and resolution time; the latter depends more on steady throughput and buffering. To compare VMess and VLESS, keep the server location, transport, and security layer as consistent as possible. Otherwise, you are mostly measuring differences in the overall deployment, not the protocols themselves. Background connection behavior also depends on system scheduling and app traffic; a status icon alone cannot tell you how much battery is being used.
Troubleshoot one setting at a time in this order: UUID and protocol type, transport, Host and Path, then TLS and SNI. To retrieve the complete settings from an existing subscription, see Four ways to import: Subscribe, Add Server, QR code, and clipboard. Keep the original imported fields where possible to avoid transcription errors.
04 / TLS transport
Trojan: verify TLS before checking the password
Trojan uses TLS to protect the connection between client and server, then handles proxy authentication and data transfer within that connection. Rather than asking users to select an encryption method like Shadowsocks, its main security properties come from TLS configuration and certificate verification. The settings may look more compact, but certificates, server names, and time are essential to establishing the connection—not optional details.
Address, port, password, and server name
Server determines where to connect, Port identifies the listening service, Password is used for protocol authentication, and SNI helps the TLS service select the right identity. Each has a distinct role. If your service details provide only an IP address, confirm how the certificate name should be validated. A reachable IP does not mean the TLS name is correct. Verifying directly against an IP is appropriate only if the certificate includes that IP identity.
Preserve capitalization and symbols in the password exactly. If TLS succeeds but the connection then closes, authentication may have failed, or the server’s forwarding target may be unavailable; check the logs. Conversely, if TLS reports a certificate error, repeatedly changing the Password is unlikely to help. Identifying the stage where the connection fails avoids toggling settings at random.
Some ports are commonly used for particular services, but that does not mean every service must use the same port. Follow the details for your existing service; don’t change the port just because TLS is involved. If a frontend gateway is in use, it must also forward the connection to the correct backend. Opening an ordinary webpage on the server does not prove that the proxy authentication endpoint is working.
Don’t leave certificate verification disabled
Certificate verification checks whether the service identity on the current connection can be trusted, including the certificate chain, validity period, and name match. If it fails, first enable automatic system time, then check SNI and the names covered by the certificate. An incomplete renewal, missing intermediate certificate, or incorrectly configured site can cause failures on some devices while others appear to work because their cached certificates and trust chains differ.
Disabling certificate verification changes the security boundary and is not a general optimization. If you temporarily compare results for diagnosis, remember that this only helps narrow down the cause; it does not fix identity verification. In an environment with a deliberately managed trust system, follow the administrator’s certificate and trust settings rather than accepting every certificate. If you cannot confirm the service identity, check the server configuration.
| What you observe | Check first | What it does not prove |
|---|---|---|
| TCP connection not established | DNS resolution, address, port, network path | That the password is definitely wrong |
| TLS handshake failed | System time, SNI, certificate chain, listening service | That verification can be disabled permanently |
| Connection closes immediately after handshake | Authentication details, protocol endpoint, server logs | That bandwidth is insufficient |
| Websites work, but UDP fails | UDP forwarding support and the app’s traffic path | That all protocol features are working |
Handshake overhead and long-lived connection behavior
TLS exchanges required information when a new connection starts, so a longer round-trip time increases setup delays. Session resumption and connection reuse may reduce some of that cost, depending on both implementations and whether the connection is actually reused. TLS does not mean sustained transfers must be slow: once a connection is stable, throughput still depends mainly on path capacity, packet loss, and server performance.
On iOS, check how often the app establishes new connections, whether it repeatedly retries after a network change, and how much traffic it handles with the screen off. When switching from Wi-Fi to cellular, an existing TCP connection may need to be re-established; that does not mean the password or certificate has changed. If the connection works on a stable network and pauses briefly only during a switch, focus on connection recovery instead of changing protocols immediately.
You can test Trojan as a TLS transport when your existing service provides complete settings and the device can verify its certificate. Decide whether it suits your needs based on how reliably your apps work, the behavior of long-lived connections, and battery use—not a single handshake time.
05 / General-purpose proxy protocols
HTTP and SOCKS5: a proxy is not the same as encryption
HTTP proxies arose from the need to forward application-layer requests and are common on managed networks, in debugging environments, and as shared access gateways. SOCKS5 offers a more general connection negotiation interface and does not require the forwarded application to use HTTP. Both are common in existing network setups, but the protocol names do not guarantee encryption between the client and proxy server. Keep proxy support and link confidentiality separate.
HTTP CONNECT and TLS to the proxy
An HTTP proxy can handle ordinary HTTP requests or use CONNECT to create a byte tunnel to a destination. When the destination uses HTTPS, its TLS connection usually runs inside that tunnel. This does not automatically encrypt all negotiation between the client and proxy. With an unencrypted HTTP connection to the proxy, the protection available for the proxy address, connection negotiation, and some authentication details differs from a TLS connection to the proxy.
If a configuration lists an HTTPS type, confirm whether it means “connect to the proxy server using TLS,” rather than merely “can access HTTPS websites.” These are not the same. The former requires checking the proxy server’s own certificate and name; the latter describes what destinations it can reach. If the service details only say “supports HTTPS,” ask the administrator of your existing service how it listens for connections.
Username and Password are for proxy authentication, not your account on a destination website. Entering a website login password in the proxy fields will not fix the proxy handshake and exposes information unnecessarily. Some managed networks use additional authentication methods that cannot be represented by a simple username and password. Check the authentication types supported by the client and the administrator’s settings; don’t assume every HTTP proxy can be imported directly.
SOCKS5 CONNECT, UDP, and DNS handling
SOCKS5 is commonly used to ask a server to establish a TCP connection. The protocol also defines UDP ASSOCIATE, but you must check separately whether the server implements it, the network allows it, and the client can use it. A successful TCP test does not guarantee UDP will work. Some services return a UDP relay address that is unreachable on your current network, causing datagram apps to fail while ordinary websites appear normal.
SOCKS5 can carry a destination as a domain name or as an already resolved IP address. This affects which side resolves the domain, but Type alone does not determine all DNS behavior. Shadowrocket DNS settings, whether the app resolves names itself, and routing rules all affect the final path. If a destination has already been resolved locally, the server may receive an IP instead of the original domain.
Username/password authentication is not encryption. On an untrusted network, be clear about which data on the connection can still be observed and which content is protected by the destination app’s own TLS. A website loading successfully does not prove confidentiality. Nor should you assume the website provides all the encryption needed, since not every type of traffic is protected in the same way.
Use cases and iOS testing
If you already have a corporate proxy or managed local network service, its settings are often straightforward to enter using the administrator’s details. Still, confirm that the service accepts connections from your current network and whether it restricts source addresses, concurrent connections, or destination ports. If some websites fail, access policy may be the cause rather than a client protocol error. Start with a test destination approved by the administrator.
On iOS, the proxy endpoint’s response time and authentication round trips affect short requests, while sustained transfers depend more on the underlying connection. A simple-looking transport is not necessarily more efficient if the proxy repeatedly rejects authentication, closes idle connections, or triggers retries. Track connection success and repeated connection attempts as well as latency; don’t compare just one number.
If the network requires a browser sign-in before access, complete that step through the system’s supported flow before testing the proxy. Without basic network access, repeatedly changing Username, Password, or DNS in Add Server is unlikely to tell you anything useful. For more troubleshooting organized by symptoms, see the FAQ.
06 / IP-layer tunnel
WireGuard: keys, addresses, and routing scope
WireGuard is designed as an IP-layer tunnel, unlike application-layer protocols that represent proxy requests individually. It uses a fixed set of modern cryptographic primitives, identifies peers by public key, and carries encrypted packets over UDP. Its configuration is therefore closer to “assign an address to the tunnel interface, then define which destinations go to which peer” than simply entering a server and password. Understanding this distinction is the first step to avoiding routing errors.
PrivateKey, PublicKey, and PresharedKey
PrivateKey is the local private key. Keep it secure and out of screenshots, public logs, and troubleshooting posts. The peer’s PublicKey identifies the remote peer; do not enter your own public key again. The server must also register the public key corresponding to your private key. If you change the key only on the client and do not update the server, the handshake will not work as expected. The keys may look similar, but their roles are not interchangeable.
PresharedKey is an optional extra shared key, used only when both sides explicitly configure it. It is not a login password, and you should not make up a string to fill it in. If the existing configuration has no such field, do not reuse a Password from another protocol. After importing, check that each value landed in the correct local or peer section. Don’t treat every key in the configuration as the same kind of authentication information.
Endpoint is the peer’s reachable address and port; interface Address is used inside the tunnel. They belong to different address spaces. Putting an internal tunnel address in Endpoint usually prevents a connection from the public internet. Using the Endpoint address as the local interface address can also break routing. Keep the server-assigned address and prefix exactly as provided rather than changing them to match your local network.
AllowedIPs, DNS, and dual-stack coverage
AllowedIPs both routes destination addresses to a peer and limits which source addresses that peer can accept. It is not an ordinary list of websites you are allowed to visit. The IPv4 and IPv6 ranges it covers should match the intended use. If only IPv4 is routed but you assume IPv6 follows the same path, some requests may succeed while others take a different route.
When WireGuard is used as a connection in Shadowrocket, distinguish client rule selection from routing inside the tunnel. The client’s policy first decides whether a request uses this connection; only then does the tunnel apply its own routing. A domain matching PROXY does not automatically mean the peer’s AllowedIPs includes the destination IP. Check both layers when troubleshooting.
DNS settings should match what the connection can reach. If the specified resolver is accessible only through the tunnel, make sure the tunnel can reach it once established. Otherwise, you may see a handshake in the logs while domain lookups still fail. Test an allowed IP destination and a domain separately to distinguish routing from DNS issues rather than regenerating keys immediately.
| Field | Purpose | What to check |
|---|---|---|
| PrivateKey | Local identity | Matches the local public key registered on the server |
| Peer PublicKey | Peer identity | Not the local public key |
| Address | Address inside the tunnel | Address and prefix match the assigned settings |
| Endpoint | Outer connection endpoint | UDP address and port are reachable |
| AllowedIPs | Address ranges for the peer | Covers the IPv4 / IPv6 traffic required |
MTU, keepalive, and mobile networks
MTU determines whether an encapsulated packet exceeds the size supported by the path. A common symptom is that small requests work while large pages or uploads stall, though other kinds of packet loss can cause similar behavior. Don’t copy a supposed universal best value. Start with the existing settings, make gradual changes based on path testing, and check both IPv4 and IPv6. Lowering MTU may help with fragmentation, but increases the number of packets needed to send the same data.
PersistentKeepalive keeps a NAT mapping alive when needed; it does not increase bandwidth. Frequent keepalives increase wireless activity, while no keepalives may slow inbound recovery on some networks. On iOS, test standby, network switching, and real-world traffic. Follow the server’s recommendation instead of shortening the interval just to keep the status looking active. WireGuard can handle endpoint changes, but recovery still depends on how the client connects and on network conditions.
07 / QUIC transport
Hysteria2: UDP paths and bandwidth settings
Hysteria2 uses QUIC as its transport for proxy traffic and includes settings related to bandwidth control. Start by understanding that QUIC runs over UDP: it does not turn UDP into a “lossless connection” or increase the physical capacity of the network. Whether your current network allows reliable UDP traffic is an important factor in establishing a connection and maintaining performance.
QUIC has specific advantages and limits
QUIC combines reliable transport and secure handshakes in a user-space protocol and supports multiple streams. Compared with carrying all traffic over a single TCP byte stream, loss on one stream does not have to block ordered delivery on every other stream at the transport layer. However, all streams still share the available path capacity and congestion state. Less cross-stream blocking does not mean packet loss has no effect: lost packets still need recovery and can still reduce effective throughput.
Some networks handle UDP queuing, rate limits, or idle mappings differently from TCP. The same server may work over Wi-Fi but be unstable over cellular, or the other way around. If the initial handshake gets no response, check UDP port reachability, Endpoint, and the server listener before checking whether websites load. A reachable HTTPS website on the same domain does not prove that the UDP endpoint used by Hysteria2 is reachable.
QUIC supports certain connection migration capabilities, but how an app recovers when the network changes still depends on both implementations and system scheduling. A brief pause after switching from Wi-Fi to cellular on iOS does not by itself show whether migration is supported. Check whether you need to reconnect manually, how long recovery takes, and whether the app’s traffic is interrupted—not just the connection indicator.
Check authentication, TLS, and obfuscation separately
The authentication string verifies client access. TLS protects the connection and verifies the service identity. Server, Port, SNI, and certificate verification settings must match the server. If the service also uses an obfuscation mechanism, its type and password must match too. Even if both the authentication password and obfuscation password are labeled Password, they are not interchangeable. Don’t enable an obfuscation setting unless it is included in the service details.
Do not mix Hysteria2 settings with older protocols that have similar names. Share links, converted subscription entries, and manually entered fields must identify the actual protocol. Successful import only means the format was accepted; it does not prove every field was preserved. After converting between formats, check authentication, port, SNI, and bandwidth units. The same server name does not guarantee the same configuration.
Upload, download, and congestion control in practice
Bandwidth settings may use units such as Mbps. First distinguish bits per second from the byte rate shown by file transfer interfaces. Upload usually means traffic sent by the client; Download usually means traffic received by the client, but check the current field description. Don’t enter your home broadband’s advertised download rate for both directions, and don’t treat the Wi-Fi link rate as usable end-to-end internet bandwidth.
Congestion control modes may use bandwidth hints differently, and the server may impose its own limit. Keep the imported values before tuning, check which options the current implementation supports, and compare one setting at a time. If speed spikes and then fluctuates while webpages become sluggish, the bottleneck queue may be filling up; raising the value further may make things worse. This can be more noticeable when other devices at home are using the network.
On iOS, high packet rates, retransmissions, user-space processing, and cellular radio activity can all affect battery use. Reaching high speeds during a download does not mean the setup is suitable for standby; test heavy foreground transfers and background idle time separately. Hysteria2 is worth comparing when your existing service explicitly supports it, UDP is stable on your current network, and your workload benefits from its transport features. If UDP is restricted, address reachability or use another compatible type offered by your existing service rather than repeatedly changing the authentication string.
08 / Testing and maintenance
Troubleshoot from a single connection to real-world traffic
The purpose of a protocol reference is not to memorize more names, but to find the next condition you can verify when a connection fails. Check in this order: network access, service endpoint, authentication and security, proxy forwarding, DNS resolution, rule matching, and the actual app or service. If an earlier layer has not passed, changes to a later one usually have no basis. For example, changing FINAL will not help if the port is unreachable, and changing DNS may be irrelevant when the certificate name is wrong.
Start with one configuration and one test destination
Keep the original node, make a copy for testing, and note the single field you change. First confirm the device has basic network access, then use Connectivity Test to check whether the endpoint is reachable. The latency result may cover only one type of probe; it does not prove the full proxy path works. Open a permitted destination with stable content and confirm that resolution, connection setup, and data transfer all succeed.
Test TCP, UDP, and DNS separately. If websites work but calls fail, investigate UDP first. If an allowed IP works but its domain does not, check DNS and domain rules. If every request fails during TLS, check the time, SNI, and certificate. When only one destination fails, also check the destination itself, server egress policy, and app account status before concluding that the whole protocol is unusable.
Use Global Routing to isolate rule issues
In Global Routing, Config, Proxy, and Direct correspond to configuration-based routing, proxy routing, and direct connections. Config follows the policy in the current configuration file; Proxy sends requests handled by the client over a proxy path; Direct provides a direct-connection comparison. Briefly comparing these modes can help identify a rule issue, but account for system traffic capture and app behavior. None of the modes bypasses every system-level condition.
The snippet below illustrates rule order. It is not a complete configuration file and contains no server details. A specific domain is matched first, with FINAL as the fallback at the end. PROXY and DIRECT are policy keywords, not protocol names such as VMess or Trojan. Changing rules does not change the authentication method for the selected server.
[Rule]
DOMAIN,api.example.com,PROXY
DOMAIN-SUFFIX,example.com,PROXY
IP-CIDR,192.168.0.0/16,DIRECT
FINAL,PROXY
If Proxy works but Config does not, check which Config is active, which rule matches the destination, and which connection that policy selects. Rules are evaluated from top to bottom, and matching stops at the first hit; a broad rule can hide a more specific one below it. For matching details, see Custom rules and the FINAL fallback. For configuration section roles, see A guide to the .conf structure.
Measure speed, battery use, and failures separately
For speed tests, keep the network, location, device temperature, test file, and duration consistent. Avoid system updates and other high-traffic tasks. For short requests, record time to first response and failure rate; for long transfers, record sustained speed and fluctuations over time. Comparing different servers measures the performance of the entire path. To compare protocols, use compatible options in the same service environment where possible and alternate tests to reduce the effect of time-of-day changes.
Measure battery use separately during idle periods and sustained activity, keeping screen brightness, wireless type, and background apps similar. System battery statistics may attribute tunnel traffic processing to a related process; a high share does not directly mean the protocol is inefficient. Compare actual battery drain, traffic volume, signal, and temperature. Don’t set keepalives too frequently just to improve standby appearance, or sacrifice stability for a short-lived speed peak.
When documenting a failure, include the protocol type, network type, failure stage, and steps to reproduce it. Remove private keys, full subscription URLs, authentication strings, and personal addresses first. If every setting matches the original service details, use server logs to investigate authentication, forwarding, or resource limits rather than endlessly toggling client options. If an existing subscription suddenly changes, see Troubleshooting subscription update failures. For persistent timeouts, follow the server timeout troubleshooting checklist.