A VPN provider ranking scored to two decimals, with the only measurements in a comment

Twelve services scored to two decimals, with no criteria given

The headline is a ranking table with a score column running from 9.20 at the top down to 8.40 at the twelfth entry, in steps small enough that the bottom four services sit within half a point of each other. Alongside each score there is a reference starting price in yuan per month, and a one-line note on line type or plan shape.

Nothing in the file says how a score is produced. There is no weighting between price, line type, node count and support, no test data behind the numbers, and no statement of when the table was last measured. A two-decimal score implies a precision that a reader will reasonably assume comes from somewhere, and here it does not.

The only measured figures anywhere in the visible text are an availability rate of 100%, an average latency of about 25.5 ms and 0% packet loss over a 30 day reference window. Those three numbers belong to a provider that has been taken offline, and they sit inside a commented-out block, together with the note that they come from the provider's own page. So the single piece of quantitative evidence in the repository describes a service that is no longer recommended.

Every provider link in the file carries a referral code

Each provider entry ends with a registration link, and each of those links carries a tracking or discount parameter rather than pointing at a plain homepage. Some use a `code` query parameter, one uses an `aff` parameter, and one entry is described in the text as having the promo code built into the registration link already. There are also percentage discounts mentioned for annual and first-time purchases, and a named discount code for one operator's annual plan.

None of that is hidden, and the file does state that promotions are subject to the operator's own site, which is the correct caveat. But the structural point stands regardless of disclosure: the list is not neutral between the operators. The author earns from the signups, which is a reasonable thing for a person to do and a poor thing for a ranking to be.

There is no score criteria that could control for it either, because there is no criteria. A reader who wants an unconflicted comparison has to go to each operator directly and compare plans there, using this file only as a list of names to go and check.

The delisted provider is still in the file, commented out

One provider is marked as temporarily offline, and its entire entry is retained inside an HTML comment rather than deleted. The comment states that the original images and material are kept there so the entry can be restored by uncommenting it when the service comes back.

Everything survives: the description bullets, the paragraph about the service's channels and support, a full plan table, the fit-for-audience line, and its registration link. So a reader viewing the raw file on GitHub sees a longer list than a reader viewing the rendered page, and the two disagree about what is currently recommended.

That entry also contains the most useful piece of arithmetic in the whole repository. Its plans are sold with a 6x multiplier on nodes, so the page's 1200GB does not mean 1200GB of transferable data: the file converts it to roughly 200GB, 1800GB to about 300GB, 3500GB to about 583GB, and a 1000GB pack to about 166GB. That single conversion is the difference between comparing two services and comparing two products.

Elsewhere the file notes that some operators have no multiplier at all, and that others open low-multiplier lines only in certain tiers.

Two package shapes: a monthly allowance, or a one-time pack

The market splits into two shapes, and both appear for most operators.

The monthly subscription gives a traffic allowance per month with a device cap and often a speed cap. The entry at the top of the list sells 128GB a month for 15 yuan with no device limit, then steps up through 192GB, 256GB and 512GB, each tier also without a device limit. The second ranked entry shows the other end of that shape: five tiers from 19.99 yuan with 100GB and two devices at 60 Mbps up to 99.99 yuan with 1200GB, ten devices, no speed cap and IPLC nodes throughout. Price, traffic, device count and speed move together, which is why a single starting price per row tells you very little.

The one-time pack is the other shape. The entry ranked sixth works entirely this way, from a 1GB trial up to 10TB, sold as a single payment. Other operators mix both, and one makes the mix explicit by putting a half-year minimum on its cheapest plan while offering monthly tiers above it.

Device caps are the quiet differentiator. Several entries sell unlimited devices at the entry price, several cap at two or three, and one operator reportedly supports no customer service at all on its cheapest tier. Those are the details that decide whether a plan is usable, and none of them appear in a single score column.

An unlimited-time plan does not promise an operator

The file is careful about one term, and the care is worth keeping.

An unlimited-time plan means the traffic is not reset on a monthly boundary and is consumed until it runs out. The entry that sells only this way starts at 130GB for a single payment of 19.9 yuan and scales to 1660GB, and it is described on its own page as having no node multiplier and no device limit. The top ranked service sells a comparable one-time 128GB pack for 36 yuan alongside its monthly tiers, so the two shapes sit next to each other in the same list at very different effective rates.

What the term does not mean is stated in as many words: it does not mean the operator has committed to keep operating. That is a distinction a buyer pays for, and the file is doing the reader a service by making it.

It also warns against assuming that a new pack stacks on top of an unspent balance. That is exactly the sort of detail that is easy to miss in a product page and expensive to discover after paying, and it is the strongest argument in the file for checking the operator's own terms rather than trusting a comparison table.

Several listed services share one upstream and cannot back each other up

The second ranked entry is described as a long-running branch site of a network called SuYing, with comparable entries existing under other names in the same family. The file draws the obvious conclusion: services from the same family should not be treated as each other's disaster recovery.

That is a real operational point rather than a footnote. If two subscriptions resolve to the same upstream, having both does not give you a second route when that upstream has an incident, which is the entire reason to keep a backup subscription. The same caution applies to the plan ladder inside one operator: an entry that sells domestic relay lines and premium IPLC lines in separate tiers means the cheap tier and the expensive tier may share the fate of the same operator on the same day.

Two other naming details from the file are of the same kind. One operator was previously published under a different name, and the file says it now uses the new name throughout, which matters if you are comparing against older lists. Another entry is explicitly ranked lower because its package details need confirming with the operator, which is an admission inside the ranking that the ranking is not finished.

Three release tag schemes in three weeks

The repository is three files: this README, a client configuration guide, and a contributing note. There is no license file, and the license field on the repository is empty, so the text here has no stated terms for reuse or redistribution. That matters more than usual for a document other people are invited to translate, mirror and republish.

The release history is untidy in a way that makes version pinning awkward. Three releases in the last week of September and the first days of October use three different tag formats: one is tagged `v10`, one is tagged `version1.0.9`, and one is tagged `version1.0.8`. The release named 1.0.9 carries the description that it was to test more airports, which suggests the list grew as a trial rather than through a stated editorial process.

The declared homepage is a separate site that carries the same ranking, so the GitHub file and the website are two copies of one list with no stated rule about which is authoritative when they disagree. The README's own advice to readers is to bookmark the repository or the site because the operators change domains, which is the most useful sentence in the file and also a description of why a static list ages badly.

Editorial conclusion

Read this repository as a map of how paid proxy subscriptions are packaged, not as a shortlist. The twelve entries, their plan tables and the client guide are genuinely informative about the shape of the market: monthly versus one-time traffic packs, device caps, speed tiers, line types, and the fact that several operators share one upstream network. Do not treat the score column as a measurement, and do not buy through the links in the file, since each one carries a referral or promo code that pays the author. Before paying anyone, do what the README itself tells you to do: test latency and throughput on your own network, treat every promotion as valid only on the operator's own page, and expect domains to move. Two facts about the repository itself are worth carrying: it has no license file, so its text is not formally reusable, and it publishes no stable version scheme, having used three different tag formats in three weeks. Nothing here is a bypass of anything you have paid for; it is a list of paid subscriptions, and the decision is whether a subscription service is what you want at all.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. Releases
5. Zirakin/airport-recommendation on GitHub