The answer to choosing a VPN without getting burned isn't in the spec sheet — it's in four things you can check before paying: whether the route count is inflated, whether peak-hour bandwidth is oversold, whether shutdown risk can be spotted in advance, and where the refund and support terms are written. What these four pitfalls have in common is that each leaves verifiable traces before you pay; most people just don't look when they order.
Here's the order to check things in: first tell direct connections, relay routes and IEPL dedicated lines apart, then go through how to verify each kind of pitfall, and finally condense it into a checklist you can follow as-is. Every number in this article can be counted and verified on public pages.
First, Tell the Route Types Apart: Direct, Relay and IEPL Dedicated Lines
The route type determines how a line performs at peak hours, and how much substance there is behind that advertised route count. Three common types:
| Type | How traffic is routed | Peak-hour performance | Best for | How it's usually labeled |
|---|---|---|---|---|
| Direct | Connects straight from your local network to an overseas server, running the whole way over public international gateways | Noticeably affected by gateway congestion; speeds fluctuate widely | Light browsing, backup routes | “Direct”, “Standard” |
| Relay | Connects first to a relay node in mainland China, then travels out along an optimized path | More stable than direct, but quality depends on whether the relay bandwidth is shared | Everyday cross-border access, streaming | “Relay”, “BGP” |
| IEPL dedicated line | Runs on an International Ethernet Private Line rather than public international transit | Path and bandwidth are relatively fixed; little degradation at peak hours | Long-lived connections, video calls, AI tool sessions | “IEPL”, “Dedicated” |
Each type is a trade-off, and what matters is matching it to the use case: direct is enough for light browsing, relay suits everyday cross-border access and streaming, and only scenarios that keep sessions open for hours justify an IEPL dedicated line. The real red flag is a single total that lumps all three together: the same total means something completely different when it's all direct versus when some of it is dedicated.
The takeaway: A total route count tells you nothing on its own; only a breakdown by type is comparable. When you see a number like “thousands of servers,” ask first: how many are dedicated, relay and direct, and can the list be opened route by route?
Inflated Server Counts: The Bigger the Number, the More You Need to Break It Down
Inflated route counts usually come in a few flavors: counting several entry points on the same server as separate routes; using “nodes” and “routes” interchangeably; leaving dead or long-unavailable routes in the list to pad the total; or simply claiming “thousands of servers worldwide” with no list to open at all.
There's only one way to check: find the itemized list and count it yourself. A credible route list should carry at least three pieces of information per entry: country/region, route type, and what the route is best suited to. A count that doesn't match the marketing, or a list that never gets updated, are both signals.
This service's route list page lays out 190+ routes across 120+ countries by region and type, so you can check the count and the type mix yourself. Whether a page like this exists at all is one of the screening criteria.
“Node count” and “route count” are two different measures: a region may have several entry routes, or just one. Blurring the two usually serves to make the total look bigger. When you look at a list, first confirm which one it's counting.
Overselling: Only Peak Hours Show the Real Bandwidth
Overselling means selling the same slice of bandwidth to far more users than it can carry: everything tests fine during the day, then between 20:00 and 23:00 — peak hours — speeds drop and connections start to fail. You can't see it on a marketing page; you can only test for it.
Fix two things when you test: the target — use the sites or services you actually visit every day, not a random speed-test image — and the time window, testing in the same slot for several days in a row to see whether the slowdown reproduces consistently. If switching to another route in the same region at the same time shows a clear difference, you can be fairly sure the route's own bandwidth is being shared out, not that the target site is at fault.
-
Fix one real target
Pick the site or service you use most, and stick with it for the whole test to keep the variables down.
-
Test at peak hours for several days
Run one round between 20:00 and 23:00 and log the results, then run another during the day as a control.
-
Switch routes as a control
At the same time of day, switch between different route types in the same region and see whether the difference reproduces consistently — if it does, the problem is the route, not the target site.
-
Hold a long-lived connection
Keep a video call or a large file transfer running for a stretch and watch for mid-session drops or frequent reconnects.
-
Keep your records inside the refund window
Screenshot the times and results so you don't have to reproduce them when you request a refund.
That's why the refund window is worth treating as a trial period: this service offers a 14-day no-questions-asked refund, long enough to cover two straight weeks of peak hours. Do your testing inside the window, and if it doesn't work out, refund inside the window — no gambling required.
The takeaway: Overselling never shows up on a marketing page — only peak-hour testing will reveal it. Before you order, confirm the refund period and the refund path, then treat it as your trial window.
Shutdown Risk: Read the Refund Terms and Support Channels
Shutdown risk can't be eliminated entirely, but four checkable points can push the odds down.
Where the refund terms are written
Terms published on the site, with an explicit number of days and a clear process, are not the same as a verbal promise in a support chat: the first can be screenshotted and kept, the second can't be proven afterwards. If you see vague wording such as “the provider reserves the right of final interpretation,” walk away.
What's the lowest entry point
A service that only sells long-term plans is also the one you lose the most on if it disappears. Having a monthly option means you can complete your checks on one month's worth of service instead of paying for a whole year up front. This service's monthly plan starts at ¥9.9.
Whether support actually answers
Send one inquiry before you order: is there a ticket system, can the help center actually return answers, and are route updates logged publicly? At least two of the three should be true before you can say someone is still maintaining the service.
Whether content and routes are still being updated
The year range in the footer, article timestamps, and additions or removals in the route list all hint at whether the operation is alive. A service where every page froze two years ago isn't worth trying at any price.
Screenshot the refund terms before you pay. The more specific the wording, the fewer arguments later; a promise that exists only in a chat log is not a term.
Privacy Claims: Beware Vague Wording and Protocol Details
This is the item most tied to real risk, and the easiest one to be fobbed off on with lines like “your privacy is very important to us.” Look at three things.
How granular the logging policy is
Credible wording separates “connection logs” from “browsing content”: it can say which connection details are kept for troubleshooting while stating plainly that browsing content is not recorded. This service's position is that browsing content is not recorded. Conversely, if a whole policy is adjectives with no specifics, don't expect it to constrain anything.
Sign-up requirements
The less information sign-up requires, the smaller the exposure. This service asks only for a username and password — no email address required.
Protocols, split routing and DNS handling
The protocol itself isn't a privacy guarantee, but it does affect usability. Shadowsocks, VMess, Trojan and VLESS are TCP/TLS-based; Hysteria2 and TUIC run on QUIC (UDP) and perform better on lossy links, at the cost of some networks throttling UDP. What really shapes the experience and your privacy is whether the client does two things: split routing (local destinations direct, overseas destinations through the route, avoiding needless detours) and DNS handling (whether DNS requests go through the tunnel once connected, so DNS leaks don't expose the domains you visit to a local resolver).
Verifying it is straightforward: once connected, open the connection check page and see whether the exit IP's location matches the route you picked, and whether the DNS location matches too. If the two don't line up, something is wrong with the split routing or DNS setup. On transport encryption, this service's stated standard is military-grade encryption; providers differ little at that layer — the details above are what set them apart.
A subscription link is part of your account credentials: don't forward it to anyone or screenshot it in public. You only need clients for the devices you actually use — this service has no device cap, with five devices online at once, and clients for Windows, macOS, Android, iOS and Linux; see Quick Start for import steps.
The Pre-Purchase Checklist: Compare Line by Line Before You Pay
Here's everything above condensed into one list. Go through it line by line before you pay. Only consider ordering once you can tick every box; whatever you can't tick is the risk you're taking on.
- ✅ The route list opens item by item: each entry shows a country/region and route type, and the count matches the marketing.
- ✅ You've run at least one peak-hour test, using the services you actually visit every day.
- ✅ The refund period is written on a page on the site, with a clear process — not something that only appears in a support chat.
- ✅ There's a low-commitment option such as a monthly plan, so you don't have to pay for a long term up front.
- ✅ Support channels are findable: at least two of tickets, a help center and a public update log.
- ✅ Sign-up asks for little, and no email address is required.
- ✅ Clients cover the devices you own, and the subscription import steps are clearly documented.
- ❌ Nothing but a “thousands of servers worldwide” graphic, with no list you can open.
- ❌ Every speed claim is a screenshot from the provider's own site, with no way to reproduce it yourself.
- ❌ Only long-term plans, with refund terms that are vague or say “it depends.”
- ❌ Support exists only as social-media DMs, with no ticket system on the site.
For comparison, here's what this service publishes on its own site — take it and compare it with anyone else:
Compare prices last. A low-commitment tier like the monthly plan from ¥9.9 isn't just cheap — it lets you complete the whole round of checks for a small outlay; and if your usage varies, data packs from 300GB that never expire avoid the waste of a billing period ending with unused data. See the pricing page for the exact tiers.
The takeaway: Of the four pitfalls, “can the routes be checked one by one” and “are the refund terms written down clearly” are the two you can settle before paying — settle those first; overselling and privacy claims can only be verified by testing and by reading the policy word for word, so put those inside the refund window.