Official Manual · Technical Reference

Shadowrocket Protocol Reference

Choose a protocol based on the server you already have, then check authentication, transport, and routing. A protocol type does not replace server configuration or determine where all traffic goes.

For help adding a server or making your first connection, start with the quick-start guide. If you can find the entry in Home but aren’t sure which type or fields to use in Add Server, or want to compare how different connection methods perform, start with the relevant section on this page.

Protocol type
The format the client and server use to exchange data
Transport
The connection that carries protocol messages
Authentication and keys
How the client and server verify each other
Routing and DNS
Who resolves the destination, and which route the request takes

Identify the protocol from your existing connection details

Identify the service before comparing performance

Selecting a type in Add Server tells Shadowrocket which message format to use when first communicating with the server, and how to authenticate and carry the destination request. Look for the protocol name in the details supplied by your service provider or network administrator. If you’ve imported a subscription, open the entry and check its existing type. Region names, route labels, and device notes are usually just labels, not technical settings. The fact that several services use the same common port doesn’t mean they’re interchangeable. If the server offers only one type, use that configuration. You can compare protocol performance only when you actually have more than one connection method available.

Record the details separately by address, authentication, transport, and traffic scope. The address and port identify the server endpoint. Authentication data might be a password, a UUID, or a key; these are not interchangeable. Transport determines whether data is carried over a standard connection, TLS, WebSocket, or another method. Traffic scope determines which destinations should use that connection. These details belong to different layers, and a single set of instructions may include multiple domain names and keys. Don’t fill in fields just in the order they appear, or use a previously working entry to guess missing information. Keep credentials in your private source materials. In troubleshooting notes, record only whether a field was filled in and who supplied it.

Common fields and what they mean
Field groupMeaning and what to checkWhat you can’t infer
Address and portIdentify the server endpoint that accepts the connectionThe destination website or protocol type
Password / UUIDClient credentials required by the serviceWhether the server certificate is trusted
SNI / HostUsed separately to identify the virtual host during the TLS handshake and at the HTTP layerThat the two values must match or both equal the server address
PathA request path required by a specific transport; enter it exactly as providedThat every protocol requires this field
UDP / AllowedIPsThe first usually concerns datagram support; the second concerns the address range assigned to the tunnel peerThat UDP traffic from every app will automatically use this connection

Check both the fields and the stage where the error occurs

“Can’t connect” can describe different failures: the server hostname doesn’t resolve, the port is unreachable, the certificate name doesn’t match, or the password is rejected. First confirm the device can access the internet without the connection. Then check that the endpoint address is valid, and compare the settings with the protocol’s handshake and authentication requirements. If the entry came from Subscribe, check whether the server is still present after an update and whether any manually changed fields were overwritten. Keep a known-good configuration for comparison. That makes it easier to identify what changed than editing several fields in one entry. If the subscription can’t be parsed or the list is empty, see Troubleshoot subscription parsing and empty lists.

Don’t reduce “connection speed” to a single number. Time to first response depends on DNS resolution, round-trip time, the handshake, and the app request. Sustained transfers depend more on bandwidth, packet loss, congestion control, and server load. Idle battery use is affected by whether the connection stays open, network changes, and background activity. A connection may start quickly but slow down under sustained traffic, or work well on a stable network but need to reconnect when switching from Wi-Fi to cellular. First identify the problem: slow page loads, unreliable large downloads, lag in real-time apps, or disconnections after the screen locks. Then choose the relevant metric.

Start by defining what each protocol does. HTTP and SOCKS5 are primarily application proxy interfaces; WireGuard works with IP packets inside a tunnel. They handle destination addresses and return traffic differently. Other protocols may all carry web requests but differ in datagram support, name verification, or outer transport. If you only need to reach one internal address, whether the route reaches it matters more than peak throughput. If you need access to multiple internal subnets, also check the DNS results and whether each subnet is reachable.

Shadowrocket is closed-source commercial software. Get the iPhone / iPad version from the App Store product page for a one-time US$2.99 purchase. Check the App Store page for system requirements. The Mac (Apple silicon and Intel), Windows, Android, and Linux versions are free, with installers available from this site at /dl/. See the download page for platform links, system notices, and installation requirements. The examples below focus on checking settings and testing on iOS.

Shadowsocks: shared keys and encryption methods

Design and configuration fields

Shadowsocks is designed to establish an encrypted forwarding connection between a client and a proxy server using a small set of connection details. The client sends the destination address and data to forward to the server, which then connects to the destination. It isn’t a network-layer tunnel that assigns an internal address to your device, and you don’t need to add each destination website’s domain to the server entry. Keep the two connections distinct: Shadowsocks protects the connection between Shadowrocket and the proxy server. Whether the application data is protected between the proxy server and the destination depends on the app itself, for example, whether it uses HTTPS.

When checking a setup manually, start with the endpoint address, port, encryption method, and password. An encryption method isn’t a performance setting; it defines how the client and server process data together. Select the same method specified by the server. Don’t substitute a similarly named option. Authenticated encryption methods also require the receiving end to verify data integrity. If the password or method doesn’t match, later exchanges will fail even if the port is reachable. Setup instructions may call authentication data a “key,” especially when a particular generation of configuration has specific encoding or length requirements. Preserve the value as provided for that method; don’t re-encode or pad it yourself.

For an entry created from a subscription, don’t paste the full connection URI into the address field. Enter only the endpoint address; if the port has its own field, remove it from the end of the address. Don’t paste the password with leading quotation marks, comments, or the surrounding instructions. Passwords with special characters may need encoding in an import format, but if the app has already decoded and imported the value correctly, don’t encode it again as plain text. If a connection worked before an update and stopped afterward, compare the type, method, and port before and after the update. Changing the protocol type isn’t the most useful first test.

Distinguish TCP connectivity from UDP support

If the existing details specify an additional transport or extension, the basic address and authentication fields are only part of the configuration. Both ends must support the additional method. Don’t assume that removing the option will restore a standard connection. Test a common website first and confirm that a TCP request receives a complete response. Then test the datagram-based app you actually need. If websites load but a real-time app gets stuck loading, check whether both the client and server support datagram forwarding with these settings, and whether your rules send the two types of traffic through different policies.

On iOS, a simple forwarding connection using this protocol may add little overhead on stable Wi-Fi for short connections. Time to first content is also affected by DNS resolution, the destination server’s location, and app caching. On cellular, repeated requests or frequent changes between wireless networks may trigger reconnections, taking more time than a single encryption operation. Don’t assume that fewer protocol fields mean lower battery use. Record sustained transfers on the same network separately from the time needed to make a successful request after switching networks. Different results for these two tests are not contradictory.

This setup suits a complete set of existing connection details when the main use is standard application proxying. If the server also provides another confirmed connection method, compare them using the same destination at similar times. If only some websites fail, record results for the same destination separately under Direct and when routed through the configured proxy. The destination may resolve differently on the two paths, or the connection from the proxy server to the destination may be the part that fails. Check the specific destination, matching rule, and DNS results instead of repeatedly trying other encryption methods.

When moving entries, distinguish notes from identity. Two records with the same displayed name may use different addresses, passwords, or account permissions. The same address may also use different methods on different ports. For the first test, choose one clearly identified record to avoid confusion from automatic switching. If it works only sometimes, note whether success or failure happened before or after a subscription update, whether the same endpoint was used, and whether the issue was limited to one network. Sharing those details with the person who supplied the connection is more useful than sending the full subscription link.

VMess and VLESS: authentication, transport, and configuration

How the two protocols differ

VMess includes user identification and protection for its own messages, and supports different outer transports. VLESS simplifies the protocol’s own encapsulation and doesn’t provide encryption through that encapsulation. For confidentiality in transit, use the protection layer explicitly specified by the server. Both types may use a UUID in a similar format and the same endpoint hostname, but their authentication formats, message structures, and server-side handling differ. Don’t change a VMess entry to VLESS just to try a newer-sounding type while keeping all the original settings.

The UUID identifies a user to the server. Enter it in full and protect it like any credential. It isn’t a certificate name or a connection note you can generate on the spot. Some VMess details include settings related to message protection; check them against the supplied values. If you have older configuration details, don’t copy compatibility settings from another user’s entry. VMess authentication also involves time, so a significantly incorrect device clock may affect the handshake. Check this after confirming the endpoint is reachable and the fields match; not every timeout is caused by the clock. Correct the time and test the same entry again.

Match outer-layer fields to each connection stage

First, connect to the endpoint address and port. Second, establish the configured protection layer, such as TLS. SNI tells the server which name the connection is for, and the certificate must pass identity checks according to the configuration. Third, set any additional transport fields, such as Host and Path for WebSocket. The first is the host information in the HTTP request; the second directs the connection to the path configured on the server. These names may match or be completely different. Even if you replace the endpoint with its resolved IP address, keep the correct name fields from the supplied details.

If the server uses WebSocket, entering only the port and user ID won’t restore a complete connection. Check the path exactly as provided, including its leading slash and letter case, and whether a separate host field is specified. If you’re using a different transport, remove any path left over from another entry. VLESS details may also specify flow control or security parameters. These are part of the server’s configuration, not general-purpose optimizations to enable one by one. If the app doesn’t offer a required setting, first confirm that it supports the specified combination; don’t substitute a similarly named option.

When errors occur, check each connection stage in order instead of guessing from the last configuration field. Being able to open websites on the underlying network doesn’t mean the proxy endpoint is reachable. Reaching the endpoint doesn’t mean certificate verification succeeded. Once you’ve confirmed the outer connection is established, check the request path, account permissions, and authentication details. Change one setting at a time and note whether the failure happens on all networks or only one access point. For an imported entry, compare manual changes with the settings after an update so a restored subscription value isn’t mistaken for a random failure.

Compare configurations, not labels

On iOS, don’t compare these connection types by protocol name alone. The outer handshake, connection reuse, server routing, and whether the path crosses between data centers can all affect the results. For first-page-load tests, distinguish a new connection from a later request over an established one. For sustained transfers, use the same destination and similar transfer times. To test background behavior, use the same On Demand conditions and note what happens after switching networks. Repeatedly opening and closing connections creates extra traffic; don’t attribute all of that overhead to the protocol itself.

If two working entries use the same service, ask the configuration provider whether they actually use the same endpoint and forwarding conditions. Results can show the effect of a configuration change only when the main conditions are similar. If the entries use different outer protection or transports, the end-to-end comparison can still help you choose the more reliable option for everyday use. It doesn’t prove that changing the UUID format or core protocol will have the same effect. For terminology, see the glossary.

Trojan: TLS protection and password authentication

Certificates and passwords serve different purposes

Trojan uses TLS to protect the connection, then handles proxy authentication and destination forwarding inside that protected connection. It’s not the same as entering a website address, and you can’t identify it by the port alone. First confirm that the client connects to the right service, then provide the password expected by the server. Establishing the secure outer connection and authenticating at the proxy layer are separate checks. The first verifies the server’s identity and protects data in transit; the second checks whether the server accepts the user’s request. Both must succeed for requests to be forwarded.

The supplied details usually include at least the endpoint address, port, password, and name used for the outer connection. If the endpoint is a numeric address but the certificate is issued to a domain, configure the corresponding domain name as specified; don’t copy the numeric address into every field. SNI helps the server select the configuration for the requested name. Certificate verification checks whether the certificate is trusted, the name matches, and the device time falls within the certificate’s validity period. A correct password doesn’t make a certificate valid, and a trusted certificate doesn’t prove that you’ve reached the right account endpoint.

If the error appears as the connection starts, test the same entry on another network you know works. If it fails on only one network, first check that the network has finished connecting, that the address is correct, and that the app isn’t automatically selecting a different entry. If certificate errors occur on multiple networks, recheck the domain and port in the supplied details and ask whether the server’s certificate or endpoint was recently changed. Don’t disable certificate verification to force a connection through. If a network administrator manages a private trust system, use only trust settings they have explicitly provided and whose source you can verify.

Distinguish TLS handshake, proxy authentication, and forwarding failures

A connection can still fail at authentication after TLS is established. For example, a password copied from another server entry may look similar but not belong to this service. Extra spaces copied from surrounding instructions can also change the password. Confirm that the password belongs to this endpoint, reimport the configuration if needed, and retry the same destination. If the connection and authentication succeed but only one website fails, check its DNS, destination port, and server access scope. Don’t classify every failure as a password or certificate problem. Reaching the handshake stage and getting valid data back from the destination are two separate things.

Some deployments also use a specific outer transport. If the details specify an additional host or path, configure that layer as instructed. If not, don’t carry settings over from another entry. For datagram-based apps, test them separately from websites. A successful website request proves only that the connection can forward that type of request; it doesn’t prove that all packet types are supported. Even if the server forwards a particular datagram type, actual performance can vary with the outer transport and packet loss. A single website speed test says little about real-time app quality.

Observe handshake frequency on iOS

The first request usually takes time to establish the connection and negotiate TLS. Whether an existing connection or session can be reused depends on the implementation, server, and state of the previous connection. A short speed test may mostly measure setup time, while a longer download may better reflect path capacity. Switching between Wi-Fi and cellular may break the existing connection and require a new one. Note when the switch occurs, then check whether the first request after recovery and the next few requests complete successfully. If only the switch causes a slowdown and transfers are otherwise stable, report recovery time separately from steady-state speed.

If the existing connection details require TLS and password authentication and explicitly name Trojan, configure that type as provided. If your goal is faster recovery after the device is idle, keep automatic connection conditions the same and record what happens when you reopen the same app after locking the screen. Don’t leave the screen on and run a continuous test. Frequent wireless wakeups, background polling, and repeated reconnections all affect battery use. System battery statistics can provide a rough comparison alongside activity time, but a short test can’t prove that one protocol will always use less power over time.

HTTP and SOCKS5: application proxies and DNS resolution

Similar interfaces, different behavior

HTTP proxies originate as a way to forward HTTP requests at the application layer. For standard HTTP, they can process requests in proxy format. When a client accesses HTTPS, a common flow is to send a CONNECT request so the proxy creates a tunnel to the destination, then establish a separate TLS connection with the destination website through that tunnel. The website’s certificate verifies the destination, but doesn’t automatically protect the connection to the proxy endpoint. If the supplied details specify TLS for the proxy connection, check the proxy’s identity using the corresponding settings. Don’t confuse “can access HTTPS” with “the connection to the proxy is encrypted.”

SOCKS5 separates the proxy protocol from the application’s payload. A common flow is to negotiate an authentication method, then submit a destination address and port for the proxy to connect to. The destination can be a domain, IPv4 address, or IPv6 address; which one is used depends on what the client sends to the proxy. SOCKS5 also defines a way to associate datagrams, but whether it works for your use case depends on the client, server, and network. A proxy that allows standard TCP forwarding may not handle datagrams. Even if the fields look similar to HTTP settings, the server doesn’t treat the two request formats as interchangeable.

Troubleshoot authentication and DNS resolution

When using an existing organization or local network proxy, first check whether it requires a username and password, and whether access is limited to certain devices, subnets, or signed-in users. Distinguish authentication errors from rejections by the destination. If HTTP reports that proxy authentication is required, check the proxy account and authentication method. If a website or business app asks you to sign in, check the destination service’s session instead. If the configuration has no username, don’t substitute your device’s sign-in name. Standard SOCKS5 username/password authentication doesn’t encrypt the connection. Don’t send sensitive business data until you’ve verified the endpoint’s access requirements and protection.

When troubleshooting internal domains, distinguish the proxy server’s hostname from the destination hostname you want to access. Even if the remote proxy resolves the destination, the client usually still needs to find the proxy endpoint first. If the destination is sent as a domain, the proxy may be able to resolve it in its own DNS environment. If the client resolves the domain to an IP address first and sends that address to the proxy, the proxy receives an address and won’t choose a new DNS result for you. Don’t assume where resolution happens based on the protocol name alone. Check the app’s DNS and routing settings, the format of the destination sent to the proxy, and any available connection logs.

If the same name resolves to different addresses on direct and proxy connections, first check whether this is expected for internal and external DNS. If your provider says a domain is accessible only on the organization’s network, check whether the selected proxy can reach the destination instead of switching to an arbitrary DNS server. For IPv6, check the whole path too: resolving a name to an address doesn’t prove the proxy has a route to it. A connection initiated from the proxy’s network may succeed or time out because no route is available. That’s a different test from checking whether your device can reach the address directly.

Use cases and iOS behavior

If you only need standard web access and your administrator provided an HTTP proxy, use the configuration they require. If you need to forward different kinds of TCP traffic and your administrator provided a SOCKS5 endpoint, test it with the actual apps you use. The fact that both proxy types can open one website doesn’t show that they support all apps, datagrams, or authentication methods equally. Access to another network also depends on server permissions and destination scope. A standard proxy usually doesn’t provide apps with a full internal network interface, so opening a website doesn’t prove that every internal service is reachable.

On iOS, a nearby internal proxy may respond quickly, but that result applies only to the current path. After switching to cellular, the endpoint may be unreachable if it depends on an office address or network permissions. Toggling the connection can’t change those access limits. Record a baseline on a network that can reach the endpoint, then note which destinations need to be retested after switching networks. Consider repeated connections and app polling when assessing idle battery use; proxy header size doesn’t indicate total device standby power when there’s no ongoing traffic.

WireGuard: key pairs and tunnel address ranges

Understand the configuration through peers and IP packets

WireGuard uses a relatively compact configuration for encrypted network tunnels: each side identifies its peer with a public key and uses configured address ranges to determine how to handle relevant IP packets. It handles network packets inside the tunnel, so each app doesn’t need to speak a proxy protocol. Use cases include connecting to internal subnets with administrator approval or using an assigned tunnel connection. Before selecting this type, confirm that your connection details are actually for WireGuard. A standard proxy hostname and password can’t replace a key pair, tunnel address, or administrator-defined routing range.

Keep the client’s private key on that device or in a controlled configuration. The administrator should register the public key for the client, and the client should use the correct server public key to identify its peer. If the setup also requires a PresharedKey, enter it as specified; it doesn’t replace either public or private keys. Don’t combine configurations from two devices or overwrite one device’s identity-specific details with another’s. Replacing a private key usually changes its corresponding public key. If the server still has the old public key registered, changing the client field alone may not restore authentication. When asking an administrator for help, provide the public key or entry identifier, never the private key.

Endpoint is the address and port that receive outer datagrams. The address assigned to the device for the tunnel is used after entering the tunnel. These are different layers, so don’t swap them just because both are written as numeric addresses in the same set of details. If the endpoint is a domain name, the client must resolve it before it can find the peer. Domains accessible inside the tunnel may also depend on DNS supplied by the administrator. Even with a successful handshake, access can fail if DNS points to a public address, the returned internal address isn’t covered, or the server has no return route.

Check AllowedIPs for the routes you need

In WireGuard, AllowedIPs associates peers with IP prefixes. For outgoing packets, it helps select the peer for a destination address; for incoming packets, it also checks whether a decrypted source address belongs to that peer. It isn’t a domain list and doesn’t configure internal DNS. The outer routing policy also affects when the system or app sends packets to WireGuard. If you need to access only a particular internal range, use the prefix supplied by the administrator. If the details include both IPv4 and IPv6, check each separately; reachability over one doesn’t mean the other is available too.

Troubleshoot in order: check that the endpoint can send and receive UDP, then check the key relationship and assigned addresses, and finally verify destination subnets and return routes. If the handshake succeeds but an internal service won’t open, test a known address the administrator allows. If the address is reachable but the name isn’t, check DNS first. If neither works, check the relevant prefix, server forwarding, and whether the destination allows traffic from this source. If a home network uses an address that overlaps with an internal subnet, the conflict may occur when the device selects a route. Ask the administrator to confirm the intended path rather than masking the overlap by routing more traffic through the tunnel.

Keepalive and mobile networks

PersistentKeepalive periodically sends data to maintain a mapping on intermediary devices when needed. It can help with some connections behind address translation. It doesn’t make the connection faster when you browse, and there’s no fixed interval that suits every network. Shorter intervals increase idle traffic and may keep a mobile device’s wireless interface active more often. Start with the value provided by the administrator. Discuss changing it only after confirming that the mapping expires while idle. If the problem happens only after the screen has been locked for a long time, record the lock duration, network type, and time to recover on the next request. That’s more useful than repeating short tests.

On iOS, a network-layer tunnel can support access from multiple internal apps when the path between peers is working well. Sustained throughput still depends on path capacity, forwarding equipment, and encapsulation. If small requests work but large transfers don’t, check packet-size limits along the path as well as server load, and ask the administrator to assess the tunnel settings. After switching networks, test both internal names and known IP addresses rather than checking only the connection toggle. Both must work to confirm that DNS and routing are available under the current conditions.

Hysteria2: QUIC, UDP, and congestion control

Why the underlying connection requires UDP

Hysteria2 is designed for proxy connections that operate under varying latency and packet-loss conditions. It uses QUIC to create an encrypted connection over UDP, then transfers data using stream multiplexing and congestion control. UDP is the underlying transport between the client and server; the content sent through the proxy can still be a reliably delivered byte stream. The fact that a website uses TCP doesn’t mean the proxy endpoint doesn’t need UDP. If the current network can’t deliver the relevant UDP packets to the server, changing the port to another number won’t automatically switch the client to TCP.

Different QUIC streams each have their own delivery order. Lost data in one stream doesn’t necessarily block delivery on other streams as it would with a single byte order for the entire connection. However, streams still share link capacity and some connection-level limits, so ongoing packet loss or congestion can still affect overall performance. Multiple streams don’t guarantee that every small request will be faster. High-latency links may benefit from connection reuse, but the benefit may not materialize if the endpoint is far away, the server is busy, or the datagram path is unstable. Test repeatedly with the same real-world task instead of relying on a description of how the protocol works in theory.

Verify the endpoint, hostname, and identity separately

The connection details should include the correct endpoint, port, and authentication data. If the setup uses a domain and certificate, also check the required SNI and trust settings. Resolving the endpoint only gives you a candidate address; it doesn’t prove UDP packets reach it. Conversely, opening a website using the same domain may only show that the address’s HTTP service is available. During troubleshooting, confirm that this entry completes its own handshake and can reach a test destination. Don’t lower certificate verification requirements to fix a UDP connectivity problem; they belong to different troubleshooting layers.

Some setups also specify additional connection parameters or a bandwidth setting for the client. Use the supplied values and confirm the direction and units. Don’t swap upload and download, or enter bytes per second in a field measured in bits per second. A bandwidth value can’t increase the network’s capacity. With congestion control that estimates bandwidth, an overly high estimate may cause more queuing, packet loss, and retransmission; an estimate that’s too low can limit utilization. If the details don’t specify a value, first establish a baseline using the app’s or service’s default requirements, then discuss adjustments with the configuration provider.

If real-time apps lag while downloads continue, record what happens to each type of traffic separately. High peak download speed doesn’t prove that interactive request latency is stable. A large transfer can also consume much of the available capacity, slowing smaller two-way requests at the same time. Test interactive traffic by itself first, then repeat under the same conditions while a sustained transfer runs. This helps distinguish a slow baseline path from queuing under load. When comparing protocols, give both entries the same test conditions instead of letting one use the network exclusively.

Network switching and battery use

On iOS, switching from Wi-Fi to cellular can change both the address and network state. Whether a session continues or recovers quickly depends on the client, server, and how the system handles that switch. If the connection fails on only one type of network, record which network works and which doesn’t, then check whether UDP to the same endpoint is reachable there. Don’t rush to change the password or treat successful recovery after switching back to Wi-Fi as proof that the other network can connect. Also make sure the connection you see recovering is the original entry, not a switch to another available service.

Consider Hysteria2 if you already have a working connection and your main issue is long-distance transfers or unstable networks. Use the same content to test sustained transfers, then test interactive use and recovery from idle with the same app. Record transfer duration, any noticeable device temperature changes, and repeated reconnections. Compare those observations with iOS battery activity periods; don’t draw conclusions from a few minutes of battery percentage. If UDP availability varies in your environment, prioritize reliability over peak speed. Compare battery use only among options that can reliably complete your tasks.

Side-by-side testing on iOS: fix routing before comparing performance

Understand the roles of the protocol and Global Routing

Once you’ve checked the server fields, return to Home, select the entry you want to test, and check Global Routing. The protocol type answers “How does this connection reach the selected service?” Global Routing and Config answer “Does this request use that service?” If a rule sends the test destination to DIRECT, a website loading normally doesn’t prove the proxy works. If a request matches REJECT, changing the password or endpoint won’t override that rule. Include the routing mode and test destination in every performance or reliability record.

Questions to check when choosing a test mode
Interface termModeWhat to check
ConfigConfigWith the current rules, which policy matches the destination?
ProxyProxyWhen eligible traffic goes through the selected proxy, does the test work?
DirectDirectHow does the underlying connection perform when the destination bypasses the selected proxy?

In the Config mode, you can inspect a rule snippet like the one below in your own test configuration. Rule order matters: check whether an earlier condition matches before looking at the fallback rule. The first line matches a specific domain, the second a specified private IP prefix, and the last sets a policy for requests that haven’t matched so far. This snippet doesn’t configure a server, and the example domain isn’t a speed-test service. Test using a destination you’re authorized to access, save the original configuration before making changes, and restore your usual mode afterward. For the limits of blocking policies, see How REJECT rules work and where they apply.

[Rule]
DOMAIN,api.example.com,PROXY
IP-CIDR,192.168.0.0/16,DIRECT
FINAL,PROXY

Set up a repeatable comparison

Start by letting the device sit idle on the same network for a similar amount of time. Record a direct baseline for the same task, then select an existing connection for the proxy test. If the destination is accessible only from an internal network, note that direct access isn’t available; there’s no need to use that failure to calculate a speed increase. Use Connectivity Test to check connectivity under the current test conditions, then verify with the actual app. A response from the test destination doesn’t guarantee that the app’s other domains, persistent connections, or datagrams work. For the first test, see First connection and Connectivity Test.

For each candidate, test the first request, subsequent requests, and sustained transfers. Don’t mix a first request with a request for a page already in the cache. For sustained transfers, use the same resource and similar task duration where possible. For mobile testing, switch networks in the same direction and record interruptions and recovery. Alternate between candidates to reduce bias from changing conditions over time. If entries point to different locations, servers, or access scopes, use the results only to choose the configuration that suits the current task. Don’t attribute the difference to the protocol alone or generalize the results to other devices or mobile networks.

When checking battery use, separate continuous foreground use from idle time with the screen locked. Keep test duration, screen use, network connection, and apps involved as similar as possible, and note whether On Demand may restore the connection in the background. System battery statistics can help you check activity periods, but usually can’t isolate wireless activity, system tasks, and proxy processing. Use Data traffic records to confirm whether data was transferred and compare the traffic totals shown in different views. They aren’t battery measurements and don’t directly match a provider’s records. For more, see Understanding traffic readings on the Data page.

Choose a configuration based on the task

If your main task is reaching internal addresses, keep the configuration that covers the required routes and DNS scope. For websites or standard transfers, choose the option that reliably completes the task with the protocol and outer settings specified in your connection details. For real-time interaction, first confirm the required datagram path, then compare latency under load and recovery after switching networks. If the options perform similarly, prefer the one with clearly sourced settings that can be reimported, and whose provider can explain failures. This reduces unnecessary variables in future troubleshooting.

If all options are unreliable, first determine whether the issue is with the underlying network, subscription, routing, or a specific destination before retrying protocol connections. A useful report includes the network type, a redacted entry name, steps to reproduce, routing mode, and time of failure. Keep identity fields, complete address lists, and subscription links private as required by the administrator. If only one mode fails, check the matching rule first. If all modes fail, check the underlying network and destination. For shared troubleshooting steps, see Troubleshooting. To start over with import and first-time testing, return to the quick-start guide.