01Choosing a protocol
Protocol Types and Global Routing
Start with the Protocol Your Server Supports
A protocol is an agreement between a client and a server, so compatibility comes first. If a server accepts only one protocol, enter the details as provided. A single server address may offer multiple endpoints, but each endpoint must be used with its own port and credentials. Changing an entry from Shadowsocks to VLESS will not create a different connection, and two entries using port 443 are not necessarily compatible. Compare protocol designs and real-world performance only when the administrator provides multiple working endpoints.
Before choosing a type in Add Server, sort the details into four groups: server address and port, credentials, transport and security settings, and traffic scope. The address directs the connection to an endpoint; credentials determine whether access is allowed; transport settings determine whether both sides can establish the same channel; and traffic scope determines which requests use it. You can inspect these fields in an existing subscription, but don't infer the protocol from an entry's name. A name that includes a location, purpose, or transport abbreviation doesn't guarantee that the current settings match.
Protocols can't be reduced to a simple speed ranking. Whether a protocol supports datagrams, relies on a certificate trusted by the system, or assigns an address to a tunnel interface are different concerns. WireGuard, for example, carries network-layer packets, while an HTTP proxy typically handles requests or connections. Both may produce the same result for a web visit, yet differ in network coverage, domain handling, and idle traffic. Start by defining what you need to do, then decide which differences matter.
Distinguish Protocol from Routing Mode
| Interface option | How it works | What to check |
|---|---|---|
| Config | Routes traffic according to the rules in the selected configuration. | Whether a destination matches the expected rule. |
| Proxy | Sends requests handled by this mode through the selected proxy. | Whether the proxy path can complete the same request. |
| Direct | Sends requests handled by this mode directly. | Whether the destination is reachable without a remote proxy. |
With Config selected, an app page and the APIs or resources it calls may match different rules. A page loading doesn't mean all its requests use the same route, and one failed resource doesn't mean the proxy is entirely unavailable. Note the domain or address that actually failed and check the matching rule; don't rely only on the domain in the address bar. If you temporarily switch to Proxy for a test, switch back to the original mode afterward so later measurements use the same route.
[Rule]
DOMAIN-SUFFIX,example.com,PROXY
FINAL,DIRECT
This snippet illustrates rule order; it is not a complete configuration. The example domain and its subdomains go to PROXY, while requests not matched by the preceding rule go to DIRECT through FINAL. To check your own setup, first confirm which file is selected in Config, then look for earlier rules that could match or override the destination. Changing a rule's action to DIRECT changes the route, not the server entry itself. Direct in Global Routing also doesn't mean the system connection is turned off. When testing a direct baseline, state exactly which method you used.
Separate App Access, Import, and Troubleshooting
If the app isn't installed, start with Getting the App and FAQs. Get the iPhone / iPad version from the App Store for a one-time payment of $2.99; 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/. A one-time client purchase is not a service plan. This site does not provide, sell, or recommend nodes or subscriptions. This guide covers using details you already have; downloading the app does not activate a connection service.
If you have one existing subscription, first check whether it updates, then whether the entries it provides work. These are separate checks: a successful update only means the client retrieved the details; an update failure doesn't mean existing entries have stopped working. If old and new settings are mixed together, note which entry failed and which subscription it came from, then compare its type and settings. Don't edit the subscription, change rules, and alter system networking all at once. Even if the connection recovers, it will be hard to tell which change helped.
02Encrypted Proxies
Shadowsocks: Encryption and Traffic Coverage
Design and Scope
Shadowsocks is designed as a lightweight encrypted proxy: the client sends the destination and traffic to a remote server, which connects to the destination. Apps don't have to implement this encryption layer themselves. It is not the same remote protocol as ordinary SOCKS5. A client may provide a local access method suited to an app, but the client and Shadowsocks server exchange Shadowsocks protocol data. Don't enter a remote Shadowsocks endpoint as a SOCKS5 type and expect it to work with only a host and port.
This proxy encrypts the segment between the client and proxy server; it doesn't replace HTTPS on the destination site. If the service uses HTTPS, the app still verifies the certificate for its connection to the destination. Assess proxy forwarding and destination trust separately. For an existing, reliable endpoint, the best reason to choose it is usually that its settings are clear and it supports your actual needs—not an assumption that one protocol is fastest on every network.
Check Authentication and Extensions Against the Source
The Shadowsocks server host and port must come from the same configuration. The encryption method and key material must match the server. Common authenticated encryption methods provide both confidentiality and data integrity, but methods aren't interchangeable. If the source distinguishes a traditional password format from a variant that requires a specific key format, don't keep moving the same string between methods. Don't add characters, spaces, or change the encoding. After copying, check the beginning and end, separators, and that the source is complete.
If the provider specifies a plugin or additional transport, check its required format and settings for that endpoint. A plugin can change how the client reaches the proxy server, so the host, path, or security settings may differ from the base address. Don't add an extension just to make an endpoint match another entry when the source doesn't specify one. Likewise, don't remove a specified extension to test the connection. If a regular entry works but one with an extension fails, check that extension's endpoint and compatibility before changing the rules for the whole configuration.
If ordinary web pages work but voice, games, or certain app requests fail, first check whether the request uses UDP. You can't infer Shadowsocks UDP support from a single successful TCP request. The result depends on whether the server accepts and forwards UDP and whether the network in between allows it. TCP and UDP remain separate transport paths even when they use the same port number on the same host. Check the actual service, then ask the administrator whether the relevant endpoint supports it; don't assume every failure is a password issue.
Test Performance on iPhone and iPad by Symptom
Start with a short request that you know will use the proxy, then test a longer transfer. If the short request fails, check DNS resolution, endpoint reachability, the encryption method, and key material first. If short requests are reliable but large downloads fluctuate, check the remote egress, the device's network connection, destination rate limits, and packet loss. Some servers don't return a clear error when authentication fails, so the client may show only a timeout or disconnect. Don't assume the displayed message proves the password is right or wrong.
When the device has enough processing capacity, weak signal, repeated retries, and server load often affect everyday performance more than differences between encryption methods. When checking battery use, don't count the extra wakeups caused by running Connectivity Test continuously as normal idle cost. Test ordinary activity on a fixed endpoint first, then compare idle connections on the same network. If the difference appears only with the screen on and heavy traffic, record the foreground app and data volume too.
If you have two Shadowsocks endpoints, first consider which reconnects more reliably on your usual networks, then compare the same download and connection stability. If only one destination performs poorly, check the route from the proxy egress to that destination and any service policies. If all destinations are affected only after switching networks, record the connection type before and after, recovery time, and whether reconnecting was needed. This is more useful than changing several settings and keeping only a one-off test result.
03Authentication and Transport
VMess and VLESS: UUID, Transport, and Security Layers
Why the Two Types Aren't Interchangeable
VMess was designed to combine a remote proxy with different underlying transports; the protocol defines authentication and message protection. VLESS simplifies the proxy protocol layer and relies mainly on a separately configured security layer for transport confidentiality. Both may use a UUID to identify a user, and both may use transports with the same name, but that doesn't make their authentication, message formats, or protection methods the same. Use the type specified by the server. The same UUID doesn't make one configuration interchangeable between the two types.
Think of a connection as passing through several checks: the device reaches the network endpoint, establishes the outer transport as specified, identifies the user through the relevant proxy protocol, and then accesses the destination remotely. If the outer layer uses TLS, it also checks the certificate and name. Reaching the same address and port doesn't mean any of the later checks have passed. Finding the failing stage is often more useful than memorizing field names—and can explain why one entry on a server works while another times out.
Enter Fields at the Correct Layer
| Source field | Purpose | Check first |
|---|---|---|
| UUID | Identifies a user authorized to connect. | Must match the endpoint; a label or remark is not a substitute. |
| Host / Path | Used by transports or endpoint routing that require these fields. | Enter as provided, preserving the leading slash in paths and letter case. |
| SNI | Specifies the server name during applicable TLS handshakes. | Use the endpoint details; don't infer it from a remark or Path. |
| Flow | Flow-control mode used by certain VLESS combinations. | Set only when the source specifies it and the client supports that combination. |
For example, with WebSocket transport, Host and Path may route traffic to an endpoint service, while SNI selects a service or certificate earlier in the TLS handshake. If all three happen to use the same domain or text, that's a deployment choice, not a general rule for entering them. If the source provides a separate host and path, keep each in its own field. Don't paste a full address into a path field or put a path in SNI to try to fix a certificate error. If a field isn't specified, check which transport the endpoint requires instead of guessing a value.
Some VMess authentication methods are sensitive to clock skew. If many previously working entries fail at once after the device time changes, check the system date and time first, then the server's time. Don't apply historical settings from a source to every entry automatically. For VLESS, check that the outer security settings and required fields match the source. Don't mistake the protocol's lack of built-in transport encryption for permission to use any outer configuration securely.
A successful subscription import only shows that the details could be parsed. If manual entry and import behave differently, compare the type, address, port, transport, security layer, and outer routing fields in person. You don't need to post the full share details publicly. If the entry fails only after import, check whether additional settings from the original configuration were omitted. If both versions fail, verify the endpoint first instead of deleting every entry or repeatedly scanning the same details.
Find Which Layer Affects Speed
Real-world latency for VMess and VLESS isn't determined by proxy messages alone. Additional connections in the outer layer, relays in front of the server, and connection reuse can all affect short requests. If two entries change both the server and outer transport, one page load can't show that the protocol layer itself is faster. When you can't control every variable, record the result as an end-to-end comparison of two separate endpoints instead of trying to assign a contribution to each layer.
On iPhone and iPad, also observe recovery after switching between Wi-Fi and cellular, and whether the next request succeeds after a few minutes of inactivity. Keep On Demand settings consistent; otherwise, the number of connection attempts becomes another variable. If both entries handle the same task in about the same time, the one with easier-to-check details and more reliable recovery after network changes is usually more practical than the one with a slightly lower test latency once.
04TLS Proxy
Trojan: Passwords, SNI, and Certificate Verification
Verify TLS Before Proxy Authentication
Trojan places proxy requests inside a TLS channel, using established mechanisms for verifying server identity and protecting transport. In a typical connection, the client reaches the endpoint, establishes the outer connection, completes the TLS handshake, then sends authentication and proxy requests according to Trojan. Reachability is only the first step, and a valid certificate doesn't prove the password is correct. If proxy authentication succeeds but the destination remains unavailable, check whether the remote server can reach it. For an endpoint that specifies an additional transport, match the source details in full; the protocol name alone isn't enough.
Don't confuse Trojan with HTTPS requests in a browser. They may use the same TLS technology, but the content sent after the handshake and the server's function are different. Opening the endpoint domain in a browser and seeing a page, redirect, or error can confirm that one service responded; it can't verify a Trojan password or proxy forwarding. Conversely, the absence of a browsable page doesn't by itself prove that the endpoint is misconfigured.
Address, SNI, and Certificate Each Answer a Different Question
The address and port say where to connect. SNI specifies the requested server name in applicable TLS handshakes. Certificate verification checks whether the other side can prove it is the expected service. The source may specify an address for the connection and a different domain as the TLS name; that isn't necessarily wrong, but use the pairing provided by the administrator. If replacing a domain endpoint with a bare address causes a name mismatch, restore it and check the source rather than disabling verification to make the connection work.
Certificate failures commonly result from a name mismatch, incorrect device time, an expired certificate, or a missing chain. Check the device's date and time first, then the name in the endpoint details, and finally the certificate and chain provided by the server. For an organization-managed private trust system, confirm the required trust settings through an organization-approved process. Don't install certificates from unknown sources. Keeping verification enabled helps prevent a connection to the wrong server from looking like a successful test, especially after switching networks.
A password is proxy authentication material, not a certificate name or device account password. If entering it manually, check capitalization, leading or trailing spaces, and extra line breaks from copying. If all devices fail at once and the endpoint's certificate or authentication details recently changed, ask the administrator what changed. If only one device fails, check its saved entry, time, and network. For field roles and common handshake failure points, see Trojan Field Guide.
Separate Initial Handshake Time from Ongoing Speed
When a new connection is required, the TLS handshake adds a round trip and cryptographic work. Once the connection is established, later requests depend more on remote forwarding, network round-trip time, and bandwidth. Record whether a measurement is for the first request after switching entries or a later request in the same session; don't combine them into one latency figure. On lossy networks, common deployments over TCP may also pause for retransmissions, especially when signal strength fluctuates.
For everyday browsing on a fixed network, first confirm that the endpoint name and certificate are stable. Then test the first page load, continued browsing, and recovery after the device wakes from sleep. If the main issue is battery drain in the background on cellular, check whether connections are being re-established frequently rather than estimating overnight battery use from the cost of one handshake. Different endpoints can have different reconnect rates, routed traffic, and foreground workloads; even with the same protocol, battery results may differ.
For a single failure, first record whether the symptom is a connection timeout, certificate error, authentication failure, or successful connection with an unreachable destination. Then choose what to check next. The remote server may not provide a clear reason for failure; in that case, retry with a destination known to work. If most destinations work but one service domain doesn't, check its route in Config and remote egress conditions instead of re-entering a password that has already been verified.
05Traditional Proxies
HTTP and SOCKS5: Requests, Tunnels, and DNS
The Difference Is What the Proxy Understands
HTTP proxies are commonly used to forward web requests and manage network access. They can process ordinary requests according to HTTP semantics. For services that need a tunnel, the client can send a CONNECT request to ask the proxy to connect to a destination. For an HTTPS destination, the client still performs its own TLS handshake with the destination through the established tunnel. That destination-side TLS is separate from whether there is also TLS between the client and the proxy. An HTTPS destination URL alone doesn't tell you that the outer connection has the same protection.
SOCKS5 focuses on letting the client specify a destination for the proxy to connect to, without requiring it to understand the semantics of each higher-level request. It can carry domain names or numeric addresses, and the specification includes UDP relay, but an implementation or service policy may not provide every feature. An endpoint can require a username and password or use another permitted method in a controlled environment. Plain SOCKS5 doesn't provide encryption, and username/password authentication doesn't protect the connection itself.
Enter the Existing Endpoint and Check Where DNS Resolves
First confirm whether the administrator provided HTTP, HTTP proxy protected by TLS, or SOCKS5. The server address, port, and credentials must all come from the same endpoint. If an organization's access details include internal domain names and a specific network, test them in an authorized environment. A familiar port number doesn't identify the proxy type, and being able to access HTTPS websites doesn't prove that the proxy service itself uses TLS. For an authentication failure, check the account and its authorization for that endpoint; don't try passwords from other systems in turn.
With SOCKS5, if the client sends a domain name, DNS resolution may happen remotely. If it resolves the name locally and sends a numeric address, the remote server connects to that address. An HTTP proxy may also resolve the request destination or the name in a CONNECT request remotely; the actual behavior depends on the target the client sends and how the configuration handles it. If a domain resolves differently on the device and the organization's network, different results don't mean one DNS answer is inherently wrong. Check which side is expected to resolve it.
If an internal domain fails but its numeric address is reachable, don't permanently replace the domain with the address. HTTPS may need the correct name to verify the certificate, and servers may select an app based on the domain. First check which side resolves the name, whether the result is in the expected address range, and whether the rules still match the same destination. Connecting directly to a numeric address and visiting the original domain are different tests; record them separately. Follow the organization's policy for destination ports accessed through the proxy.
If some web services work but a datagram-based app doesn't, don't treat HTTP CONNECT as a general-purpose UDP tunnel. Even if a SOCKS5 endpoint claims UDP support, the control connection, datagram relay, and both network segments can still impose limits. Judge it by whether the actual service works, not by a TCP speed test. For use limited to internal web pages or a fixed set of apps, compatibility with the existing endpoint is usually more important than additional forwarding capabilities.
Don't Treat Speed and Security as the Same Measure
Traditional proxies may have a relatively short processing path, but throughput still depends on proxy capacity, connection policies, and the egress network. Compare the same destination in an authorized environment, then repeat the test to rule out caching or temporary rate limits. HTTPS sites have their own handshake overhead; don't attribute all of it to the proxy protocol, or turn a simple connection-layer comparison into a claim about end-to-end service speed.
On iPhone and iPad, if an endpoint works on an organization's Wi-Fi but fails elsewhere, first check whether it is meant to be reachable outside that network. If it's intended for internal use only, don't try to extend its availability by changing certificate settings or probing ports repeatedly. For increased idle battery use, first check whether background apps repeatedly request unreachable addresses or retry after policy blocks. For an endpoint known to serve only a few domains, use Config to check that it isn't being applied to unrelated requests.
06Network-Layer Tunnels
WireGuard: Keys, Addresses, and MTU
Carries Network-Layer Packets
WireGuard is designed as a network-layer tunnel. It identifies peers using public keys and sends encapsulated packets over UDP. Unlike a proxy that forwards app-level connections, it involves tunnel addresses, peer-allowed prefixes, and routing. This suits configurations with a defined peer and authorized network ranges, but a server hostname alone doesn't tell the client which addresses are reachable. In Shadowrocket, also check whether requests are assigned to another route by Global Routing or Config before they reach the relevant entry.
WireGuard handshakes don't use domain-based authentication like website certificates; trust is based on public keys exchanged in advance. An endpoint hostname can still be used to locate the peer. If the public key is wrong, the UDP port may be reachable without a valid handshake. And a completed handshake doesn't rule out return-route, allowed-prefix, or DNS issues, so don't equate a handshake log with successful access to a service.
Don't Mix Up Key Direction
| Field | Belongs to | How to enter it |
|---|---|---|
| PrivateKey | Local device identity. | The device's private key; don't share it with other users. |
| PublicKey | Peer identity. | Enter the server's public key in the peer entry. |
| Address | Local address inside the tunnel. | Use the assigned address and prefix, not the endpoint host. |
| Endpoint | Network endpoint for the peer. | Use the host and UDP port provided by the administrator. |
| AllowedIPs | Network prefixes associated with the peer. | Keep to the authorized scope; don't expand it just for troubleshooting. |
The server must recognize the public key assigned to this device. Copying all credentials from an old device to a new one may cause both devices to compete for the same identity and tunnel address. If one device works on its own but the connection drops when both are online, first check whether they share the same settings. If a preshared key is configured, it must also match on both sides. Don't enter the server's public key or the device's private key there if none was specified. See WireGuard Keys and Addresses Explained for a field-by-field reference.
AllowedIPs links a peer to network prefixes that can be sent through it. In WireGuard, it helps select a peer for a destination and restricts the source addresses of packets received from that peer. It isn't permission to access a service, and it can't fix a missing return route on the remote side. For an internal service with the same name, first verify that the internal DNS result belongs to the network that should use the tunnel, then check whether that network overlaps with the local Wi-Fi network.
Check Packet Size and Keepalive if Long Connections Stall
MTU limits the size of packets carried by the tunnel interface. Encapsulation adds overhead, and the maximum size supported can vary by network path. If small requests repeatedly succeed but a large transfer stalls, while other paths work normally, MTU on that endpoint is worth checking. Record the current settings and network first, then make one change within the range recommended by the administrator and repeat the same request. There is no reliable fixed value for every network.
PersistentKeepalive can send traffic at regular intervals, mainly to maintain address mappings on networks where a device needs to receive traffic from a remote peer. If you don't need this, a shorter interval isn't a way to increase throughput; more keepalives can wake a cellular connection more often. If failures occur only after a period of inactivity, first check whether the initial recovery succeeds, then review network mappings and device transitions. Don't change the keepalive interval and MTU at the same time.
If you need to use the connection across cellular and Wi-Fi, first confirm that both networks can complete the UDP handshake, then test the usable scope with the intended traffic. Reliable access to internal services over time says more about a suitable configuration than a short peak-speed test. Record whether there is “a handshake but no service access” or “no valid handshake.” These point to different checks: internal addresses and routes, or the endpoint, network, and keys.
07QUIC Transport
Hysteria2: UDP Paths and Bandwidth Settings
Designed for Challenging Links, Not Every Network
Hysteria2 uses QUIC over UDP to communicate with the server, with a design aimed at maintaining useful throughput on links with packet loss or high round-trip times. QUIC can handle multiple streams in one connection; while one stream waits for data recovery, others don't necessarily have to wait for the same data. But those streams still share device and network resources. Persistent congestion, remote egress bottlenecks, and unstable wireless connections don't disappear when you change protocols. Measure it on the network and for the task you actually use.
UDP is usually a prerequisite for establishing a connection to this endpoint. Being able to access an HTTPS page on a network only shows that the page's route and transport work; it doesn't prove the specified server's UDP port is reachable. If the same configuration works on Wi-Fi but is unreliable on cellular, or vice versa, first record which network fails and compare the destination port, network conditions, and the service's stated capabilities. Don't simply change the endpoint to TCP; use another compatible configuration provided by the administrator.
Check the Essential Settings First
Use the existing details to confirm the type, server address, UDP port, authentication material, and TLS name. Set SNI and certificate verification according to the endpoint's deployment; correct DNS resolution doesn't replace identity verification. If the administrator provides other transport options, check the specified values and client support. Don't mix settings from another entry. Optional obfuscation settings must match the server. If they're not specified, don't add them just because of a timeout, and don't confuse them with authentication or certificate settings.
When entering settings manually, it's easy to split a port or path incorrectly from an existing address. First check whether the configuration specifies a single endpoint or includes additional options, then enter each value in its corresponding field for the type selected in Add Server. Don't paste everything into the host field. If an entry from a subscription works, compare its non-sensitive fields. For a certificate error, check the time and name. If there's no handshake response for a long time, prioritize UDP reachability, the endpoint address, and the settings agreed on by both sides.
If the source or client includes upload and download bandwidth settings, confirm the units and which direction each value describes before referring to actual network capacity and the administrator's guidance. Mobile network speeds fluctuate; a short test peak doesn't represent the stable capacity of a full connection. An arbitrarily high rate may cause longer queues and packet loss, making throughput appear higher while interactive requests slow down. A rate set too low may also limit sustained transfers. Unless the source requires it, bandwidth figures don't need to be the first settings you change when checking connectivity.
Test Long Transfers and Interactive Tasks Separately
Start with a lightweight request to check whether the connection can be established, then run a longer transfer to observe sustained throughput. If you use video calls, remote control, or frequent small requests, also watch for stalls, recovery time, and latency changes. A fast download doesn't prove that small requests will be faster, especially while other uploads are running. Compare settings on the same network, with the same content and routing mode, repeat the test several times, and note whether the network changed.
If the same endpoint offers multiple entries, first compare which one supports the task more reliably, then look into throughput differences. Record interruptions that happen only after a network switch: when the old connection stops working, how long until the next request recovers, and whether the system signal changes at the same time. Client implementation, remote deployment, and network conditions all affect recovery. Don't assume QUIC's design guarantees that apps won't notice an interruption.
On iPhone and iPad, throughput and battery use aren't linearly related. Completing a task quickly may shorten active time, while severe packet loss and repeated connection attempts can use more resources. If the device gets warm, first distinguish heavy continuous transfer from idle with the screen locked, and check for file syncing or app updates at the same time. If another endpoint is already reliable, use Hysteria2 regularly only when the target network allows UDP and testing shows a benefit.
08Testing and Review
A Workflow for Speed, Battery, and Troubleshooting
Narrow Down Options by Use Case
For an organization's existing web proxy, first meet the access and authentication requirements for its HTTP or SOCKS5 endpoint. For access to designated internal networks, check the addresses, prefixes, and DNS in an existing WireGuard configuration. If you already have Shadowsocks, VMess, VLESS, or Trojan endpoints, start with one whose settings match and that has repeatedly worked on your usual network. If you also have Hysteria2 details, treat it as a separate option that requires testing the UDP path. Don't rank speeds without a clear match between the endpoint and its server-side configuration.
Define success separately for browsing, background sync, large downloads, and interactive requests. For browsing, measure how soon the first page works. For background sync, check whether the task completes with the screen off. For large downloads, look at sustained throughput and interruptions. For interactive requests, watch for frequent high-latency spikes. If an entry performs better on only one measure, weigh that against your main use case. For work that needs reliable access, don't replace a proven route just because of one faster speed test.
Set Up Repeatable Test Conditions
After selecting an entry in Home, record the device, network type, Global Routing mode, On Demand status, test destination, and approximate time. First check which rule handles a known request, then test the first request, continued requests, and recovery after sleep. If the test doesn't use the relevant entry, even a precise result can't compare protocols. For a page loaded repeatedly, remember that caching may change the result. For sustained transfers, use the same content and destination each time.
Connectivity Test can check for a response under limited conditions, but it can't replace testing your actual workload. Frequent repeated tests can change connection and wireless states; recording only the lowest latency may miss timeouts and stalls. For repeated comparisons, record the number of successful tests, typical duration, and notable failures side by side. Don't declare a winner over a negligible difference. If two entries share a remote egress, consider whether test order or simultaneous use changed its load.
Compare battery use in two separate groups: sustained foreground transfers and background idle. Use similar screen states, signal conditions, and durations, then check the device's battery usage records. On Demand changes when a connection starts; it doesn't guarantee lower battery use in every situation. Frequent triggers may cause repeated handshakes, while a persistent connection may stay active due to keepalives or traffic. To compare the two approaches, keep the protocol and routing unchanged. See Background Battery Use and On Demand Troubleshooting before testing.
Narrow Down the Failure by Stage
If every destination fails, first check whether the device can access the internet, then verify the server endpoint, port, and outer transport. If a handshake or authentication error appears, check the fields listed in the relevant section. If only some destinations fail, look at DNS resolution, rule matching, and egress policy. For failures limited to large requests or long connections, check packet loss, congestion, and packet-size limits for the protocol in use. If a connection fails only after working for a while, note when the network changes, the device wakes from sleep, or keepalive traffic occurs. Don't keep changing credentials you've already verified.
Change one setting at a time and keep the original value so you can restore it. If you're not sure whether the problem is in Config or the server, compare Proxy with the original configuration against the same destination in an authorized test environment. If Proxy works but Config doesn't, check the rules and DNS again. If both modes fail, use a known-reliable destination to distinguish a remote endpoint problem from a single service issue. For the setup sequence, return to the quick-start guide. For routing behavior, see Proxy Modes.
Write up findings in a reproducible way, such as: “The same entry completes a short request on a fixed network. After switching networks, the first service request fails, but reconnecting restores access.” This is more useful to an internal administrator than “the protocol is unstable.” When you're done, restore the intended Global Routing and On Demand settings, confirm that common tasks use the expected route, and then end the test. If a later subscription update changes the results, establish a new baseline using the updated details.