Most companies flatter themselves in their registered names. This one filed a confession. The RIPE Database record created on 2 December 2022 describes autonomous system 200959 as the "Sunflyer Global Experiment Network", and the organisation object behind it is not a company at all: it is a single name, Yaoxuan Chen, with an address on Xinghu Street in the Suzhou Industrial Park, a maintainer handle, and a contact role grandly titled NOC whose mailbox sits on a personal domain. There is no capital, no licence, no staff. There is, however, a real network. As of this week, RIPEstat's routing view shows ten IPv6 prefixes announced continuously to the global routing system, corroborated by Hurricane Electric's routing observatory, which also records two live transit relationships — one with Vultr's network in Singapore, one with the hosting firm xTom in Germany — and memberships of two Frankfurt-area internet exchanges.
Between China's licensed triopoly and the merely connected sits a stratum that rarely gets counted: privately run networks, thinly documented, holding real routable resources under foreign registration for research, practice or ambition. They are invisible in Chinese statistics because no Chinese institution issued them anything, and they are easy to miss in global statistics because each one is tiny. The Sunflyer Global Experiment Network is as clean a specimen of the stratum as the public record offers — clean because its operator, unusually, documents himself in two languages and leaves thirteen years of traces.
An experiment that owns routable resources is a strange economic object. It produces no revenue and books no assets, yet it occupies entries in the same global registries that carry China Telecom and Deutsche Telekom, and its routes propagate through the same tables. The question this piece pursues is the one the name invites: what is such an experiment for, what does it cost to hold and run, what may it legally carry, and what does the existence of a whole stratum of networks like it say about demand the licensed Chinese market declines to serve? The answers turn out to be measurable to a degree unusual for anything labelled grey. Nearly every input has a published price, because the experiment layer runs on tariff sheets — just not Chinese ones.
One person, entirely on paper
Identity is usually the murky part of profiling an unlicensed network. Here it is the clearest part, which is itself informative. The registry organisation object gives a personal name and a Suzhou address. The PeeringDB entry — the peering industry's self-service directory — classifies the network as non-profit, IPv6-only, carrying 20–100 Mbps of mostly inbound traffic, with an open peering policy and 1G ports at two exchanges. The network's public face, a page on the operator's blog at blog.sunflyer.cn, states the purpose in plain terms: an experimental backbone "maintained by myself to understand the global networking infrastructure and connect my own network to the world", IPv6-only because of what the Chinese text calls, with self-mocking parentheses, a funding problem — poor. The blog itself translates as "some noob's dust-gathering blog", and its footer signs the work of "CrazyChen (aka. Sunflyer)".
The paper trail runs back thirteen years. A query against CNNIC's whois service shows sunflyer.cn registered on 5 September 2013 under the real-name-verified registrant 陈耀璇 — the same Yaoxuan Chen as the RIPE organisation object — with the registration paid forward to 2030. The blog's tag cloud preserves a student's history: posts on campus-network dial-up tools, on a Chongqing university's network quirks, on Android teardowns, dating back to the mid-2010s. Somewhere between then and 2022 the student became an engineer in Suzhou with a few hundred dollars a year to spend on infrastructure tuition. A trading-name search finds no corresponding company: the National Enterprise Credit Information Publicity System cannot be queried from outside China without human verification, and that attempt failed here, but open-web searches in Chinese and English return no registered firm under the Sunflyer name or its Chinese rendering. The name exists in exactly two kinds of places: internet registries and the operator's own site. On the evidence available, Sunflyer Global Experiment Network is one natural person's registered hobby, and nothing about the record suggests otherwise. That completeness matters analytically. In China's licensed market, opacity usually hides ownership. In the experiment layer, the ownership is published to the world in a European database — because publication is the price of participation, and because the thing being protected is not an asset but a person's standing to keep playing.
Even the hosting of the blog participates in the pattern. The domain is Chinese, real-name registered under CNNIC's rules, yet the site resolves through a content-distribution hostname to Cloudflare's edge network, and no ICP filing number — the registration code mainland-hosted sites must display — appears in its footer. A .cn identity, served from outside the mainland hosting regime: the same shape as the network itself, Chinese in name and person, foreign in infrastructure. None of this is concealment. It is the standard posture of the stratum, which keeps identity onshore, where identity is compulsory, and operations offshore, where operations are affordable.
There is one commercial-looking artefact in the record, and it deserves attention rather than dismissal. Among the address blocks the network registers is a /46 created in March 2023 whose database object is named "SGENv6-CUSTOMER" and described as a customer IP range. An experiment with customers is either a network rehearsing the shape of a business or a hobbyist borrowing the vocabulary of one. Which of the two it is cannot be settled from the registry alone; the distinction carries real legal weight in China, as discussed below, and it is flagged here as the single most consequential open question in the record.
The offshore cottage industry that supplies it
Nothing in this network touches a Chinese supplier. Its autonomous system number exists under the sponsorship of Zappie Host LLC, a Delaware-registered company operating from an Auckland address, which acts as the RIPE-accredited local registry of record. One tranche of its address space sits inside a Zappie allocation whose country field reads Isle of Man. Another tranche descends from a /32 pool block jointly maintained with the successors of Route48, a community project that handed out free IPv6 prefixes to hobbyists, now administered care of Cloudie Networks, a US LLC that also appears — as autonomous system 924 — among the transit relationships declared in Sunflyer's own registry object. The /46 customer range descends from an allocation held by Securebit AG, a Swiss registry-services firm. Connectivity comes from xTom, an Estonian-registered hosting group with German operations, and from Vultr in Singapore. The exchange ports sit at LOCIX, a community-priced Frankfurt and Dusseldorf exchange whose PeeringDB notes advertise free 10G, 40G and even 100G peering ports and which counts 226 connected networks in Frankfurt alone.
The registry object's declared routing intentions widen the supplier map further: alongside the live transits, its import lines name Hurricane Electric — the American backbone whose free tunnel services have been the traditional first rung of hobbyist routing — and iFog, a Swiss host that caters explicitly to personal networks. Declared-but-dormant relationships of that kind are the experiment layer's equivalent of credit lines: cost-free to state, activated when a free or nearly free session becomes available, dropped without ceremony. A licensed carrier's registry object encodes contracts. This one encodes possibilities.
This is not an exotic arrangement. It is a cottage industry with its own economics, and Sunflyer is a representative customer. An inverse query against the RIPE Database run for this piece on 3 July 2026 returns 87 autonomous system numbers currently sponsored by Zappie Host alone, and the naming pattern of the objects — personal surnames rendered as network names — makes plain what the book of business is: individuals, worldwide, holding personal autonomous systems through a sponsoring registry that files their paperwork. Zappie is one of a dozen such firms; Securebit publishes a shop where an address block is provisioned "within 60 seconds"; the operator's own page thanks xTom for the Frankfurt presence. The supply chain of China's experiment layer, in other words, is a set of small companies in Delaware, Zug, Tallinn and Auckland whose collective product is the retail decomposition of RIPE membership: the right to exist in the global routing system, sliced into personal portions and sold at tariffs a student can pay.
The dependency structure this creates is worth stating precisely, because it is the experiment layer's real balance sheet. Sunflyer's continued existence requires: a sponsoring registry willing to keep filing EUR 50 of annual paperwork; two or three hosting firms willing to run BGP sessions to a personal ASN for the price of their smallest virtual machine; a community exchange that keeps ports free; and a free-prefix pool that stays administered. Every one of those is a favour-adjacent commercial relationship with no contract depth. Switching costs are near zero in money — renumbering a /44 of hobby space is an afternoon — but the supply is concentrated in a handful of firms whose own economics are thin. When Route48's free service wound down in 2023, the hobbyist space that depended on it went dark with it; the pool block behind part of Sunflyer's space was re-created under new administration in December 2024, a date visible in the object itself. The experiment layer's suppliers die and resurrect at the frequency of side projects, and its members treat that churn as weather.
What the experiment costs to run
The arithmetic of this network can be assembled almost entirely from published price lists, which is itself the finding: the grey layer is grey in Chinese regulatory terms, but its cost base is fully documented in euros, francs and dollars. Start at the registry, because the registry's fee design is what makes everything downstream possible. Under the RIPE NCC Charging Scheme 2026, becoming a member registry costs EUR 1,000 to sign up and EUR 1,800 per year to maintain — sums that price out individuals, exactly as CNNIC's do. But the same document prices sponsorship: a member pays EUR 50 per year for each autonomous system number it sponsors for an end user — the wholesale cost, to Zappie Host, of keeping Sunflyer's number on the books, out of a sponsorship book of 87 such numbers that implies roughly EUR 4,350 of annual registry charges passed through to hobbyists worldwide. Spread the sponsor's own EUR 1,800 account fee across that book and the fully loaded wholesale cost of keeping one personal network on the registry's books comes to about EUR 70 a year — the number around which every retail offer in this market clusters. Retail markups vary by sponsor and Zappie does not publish its current fee, a gap flagged here; Securebit, the Swiss firm whose space hosts the customer range, publishes a tariff under which a /44 of provider-aggregatable IPv6 — the size class Sunflyer announces — costs CHF 19 to set up and CHF 15 a year to hold, registered in the RIPE Database within a minute of payment.
Connectivity is the next line. Vultr's published plan list prices the smallest IPv6-only virtual machine at USD 2.50 a month, and its BGP documentation lets a customer announce their own prefixes — minimum a /48 — over a session to the host network with no separate charge stated. That is the entire Singapore point of presence: thirty dollars a year. The Frankfurt presence rides a machine of the same class at xTom, whose exact price for this customer is not published and is treated here as an inference of like magnitude, plus a LOCIX exchange port that the exchange's own record advertises as free. Tunnels between the nodes cost nothing but configuration; the operator's page specifies WireGuard by preference, GRE where needed.
Assemble it and label each part. Documented, from tariff and registry sources: EUR 50 of registry paperwork embedded in whatever the sponsor charges; CHF 34 first-year, CHF 15 thereafter, for a /44 at published Swiss rates; USD 30 a year for the Singapore node at Vultr's list price; zero for exchange ports at LOCIX's advertised terms; zero marginal for tunnels. Inferred, and flagged as such: a Frankfurt machine in the USD 30–80 a year class, a retail sponsorship markup somewhere between zero and double the registry's EUR 50, and the possibility that part of the address space — the tranche descending from the free-pool block — costs nothing at all, which its provenance suggests but no invoice proves. The documented floor is roughly USD 140 in year one; a realistic envelope for the whole operation, including modest markups and a second and third virtual machine, is USD 150–300 a year. The operator's own characterisation — IPv6-only because of poverty — is consistent with the sharpest relative price in the stack: on Securebit's IPv4 tariff, the smallest routable IPv4 block, a /24, rents for CHF 300 a month against the /44's CHF 15 a year. At annual rates that is a factor of 240. The experiment layer is an IPv6 layer not by conviction but because the other protocol costs 240 times as much to sit in the same routing table.
Note what is absent from that ledger: any line where money comes in. The network sells nothing and, on all available evidence, is paid by no one; the "who pays" question has a one-word answer — the operator — and the revenue logic is the household's, not the firm's. That absence is not a data gap. In a jurisdiction where the difference between hobby and business is the difference between a nuisance fine and criminal exposure, a zero-revenue structure is the load-bearing design choice, and every cost in the stack is sized to be sustainable indefinitely out of a salary's rounding error.
What does the money buy? Nothing that produces income. It buys the ability to originate routes, to appear in the peering directory alongside carriers, to misconfigure a session and watch the consequences propagate, to be — in the community's own vocabulary, which the operator's Chinese text uses with a shrug — a "BGP player". The network's declared traffic band, 20–100 Mbps mostly inbound, is a home connection's worth. The unit economics of the Sunflyer Global Experiment Network are therefore the unit economics of tuition: some hundreds of dollars a year of consumption spending that returns human capital, standing in a technical community, and optionality. It is unpaid research and development, and the question of who eventually captures the return is answered the way it always is with tuition — the employer of whoever the student becomes.
The price of doing it officially
The comparison that gives this piece its number sits across two published fee schedules. China participates in the regional internet registry system through CNNIC, which operates an IP Address Allocation Alliance for entities that want Chinese-registered numbers. CNNIC's published charging policy sets a one-time account-opening fee of RMB 10,000, an annual fee of RMB 10,000 for the smallest IPv6 holding class, and — the cleanest line for comparison — RMB 10,000 per year for each autonomous system number. The registry-side price of the same artefact through a RIPE sponsoring registry is EUR 50 a year under the 2026 charging scheme. At mid-2026 exchange rates of roughly eight yuan to the euro, the official Chinese path prices an ASN's annual existence at about EUR 1,250 — twenty-five times the unofficial path's registry cost, before any Chinese licensing consideration is reached. That 25:1 ratio between two filed tariffs, not an estimated range, is the anchor for everything else claimed here. On address space the asymmetry is structural rather than merely priced: CNNIC's application rules set the default minimum IPv6 allocation at /32 — 4,096 times the /44 this network announces — and frame eligibility around organisations that plan to provide connectivity to customers. Press coverage in 2020 reported CNNIC halving IPv6 annual fees for new members, a discount corroborated only at secondary level and flagged as such; even taking it at face value, the smallest official Chinese address product costs roughly forty times the Swiss-tariff /44 per year while being thousands of times larger than a hobbyist needs. The licensed system does not sell a small size. That, more than any per-unit price, is the market gap the experiment layer measures.
The regional alternative confirms the pattern. Direct membership of APNIC, the Asia-Pacific registry, runs from AUD 1,295 a year in 2026 for the smallest resource-holding example in its fee table, plus an AUD 500 sign-up fee; an associate tier exists at AUD 500 a year, but it is defined by holding no addresses at all — membership without the thing membership is for. Nowhere in the official Asia-Pacific structure — CNNIC or APNIC — does there exist a EUR-50-a-year product for a person who wants one autonomous system and a sliver of IPv6 to learn with. Europe's registry region created that product not by intent but by permitting sponsorship, and a global cottage industry retails it. The result is quietly comic when read in the registry's own metadata: the routing-data service bgp.tools files AS200959 under country code CN, network name "Yaoxuan Chen" — a Chinese network, by a Chinese person, that exists because European institutional plumbing sells existence cheaply and the Chinese system does not sell it at all.
And beneath the registry fees lies the wall that prices are not allowed to cross. Under the Telecommunications Regulations, operating a telecommunications business in China requires a licence, and Article 13 requires the applicant to be a lawfully established company — an individual is ineligible at the threshold, at any price. Article 58 prohibits, specifically, renting international lines or privately installing transit equipment to run cross-border telecommunications without authorisation, and Article 64 requires all international traffic to pass through state-approved gateway bureaus. The 1996 Provisional Regulations on international networking, still on the books, say it in one sentence: no unit or individual may self-establish or use channels other than the state public networks' international gateways — with Article 14 attaching a police warning and a fine of up to RMB 15,000. A 2017 MIIT campaign notice renewed the point for the modern era, barring self-built or leased cross-border channels, VPNs included, without approval, and the ministry's own clarification that summer confirmed the target: unlicensed operators, not licensed use. The official price of a personal cross-border network in China is therefore not high. It is undefined — the licensed schedule has no row for it.
The perimeter an experiment must not cross
Read against that wall, the design of this network stops looking like a technical accident and starts looking like a compliance drawing. Every route the Sunflyer Global Experiment Network announces originates outside China: Frankfurt, Dusseldorf in earlier configurations, Singapore. The traffic its declared band describes flows between rented machines abroad. The operator's page states — in both languages, and the November 2023 archived capture shows the sentence has been there for years — that the mainland Chinese node accepts no peering from overseas, "due to regulation limits", and at that time did not peer domestically either. Whatever crosses the border between the operator's home and the network's overseas body is, on the public record, indistinguishable from an individual's ordinary encrypted traffic to servers he rents. The experiment holds resources globally but is careful never to become a channel — the specific thing Article 58, the 1996 rules and the 2017 notice prohibit. The perimeter is drawn exactly where the fines begin.
The tolerance economics deserve to be stated as soberly as the fee schedules, because tolerance is a price too. For an individual whose cross-border conduct amounts to using channels other than approved gateways, the written sanction is the 1996 regulation's warning and up-to-RMB-15,000 fine — a deterrent calibrated to a hobbyist's budget, roughly ten times this network's annual cost base. For anyone who crosses into operating a business — selling connectivity, reselling channels — the Telecommunications Regulations' penalty articles escalate to confiscation and fines of three to five times illegal income, or RMB 100,000 to 1,000,000 where income is small, plus the separate machinery aimed at unlicensed VPN commerce that produced criminal convictions through the late 2010s. The legal gradient between experiment and business is therefore approximately two orders of magnitude of fine, plus criminal exposure. This is the context in which the "SGENv6-CUSTOMER" object should be read, and the most economical reading is that it is vocabulary, not commerce: registry templates encourage networks to structure space into infrastructure and customer ranges, sub-allocating a tunnel endpoint to a friend is customary in this community, and no marketplace trace, price list or resale advertisement for Sunflyer space surfaced anywhere in this research. But the object is load-bearing for the network's legal character, and it is the first thing a regulator — or a competitor — would point at. Opacity here is not evasive; it is the absence of any commercial surface at all. Treating that absence as evidence: a network with revenue leaves traces, and this one leaves none.
Precedent sharpens the gradient. The enforcement record of the late 2010s — the cases that made the 2017 notice famous — ran against people selling cross-border access: subscription services, resold lines, tooling with customers. Fines and, in the worst cases, prison terms attached to revenue, and sentencing narratives dwelt on turnover figures. No comparable public case has surfaced against anyone for holding foreign-registered numbers, announcing routes from foreign machines, or writing about routing — activities that describe this network's entire observable footprint. The line the state has actually policed, as opposed to the lines it has merely drawn, is the money line. An experiment with no revenue sits on the safe side of the only boundary with a body count, and the operator's public posture suggests he knows the map exactly: the one place his pages turn formal is the sentence refusing cross-border peering.
Why does the state, which can see the same registry objects anyone can, leave the stratum alone? Three mechanically plausible reasons, none requiring benevolence. The layer is small: personal systems carrying home-connection traffic volumes between rented machines abroad, none of it terminating Chinese users. It is self-policing in the way this network illustrates — the participants' own pages recite the rules, because the participants have careers to protect. And it is informative: a published stratum of technically ambitious young engineers identifying themselves by real name and address is, among other things, a talent census. The regime's revealed preference since at least 2017 has been to prosecute sellers, not students. The stratum's continued visibility is the equilibrium price of that preference.
Signals from the market that is not supposed to exist
The unofficial evidence around this network is richer than its balance sheet, and it points consistently in one direction: the experiment layer is the research arm of a much larger grey economy in China-adjacent connectivity, without being a commercial participant in it. The operator's most-cited work is not the network but a dossier — a long, periodically updated post on the three Chinese carriers' international interconnection, first published in August 2020 and revised as recently as February 2025, mapping which backbone tiers cross the border where, and what each is worth. When it circulated on V2EX, the hosting community's response was collectors' gratitude — bookmarked, saved, reposted. The blog surfaces in NodeSeek's hosting-forum search trails wherever return-path routing into China is argued over. The dossier's own change log records the author trimming what he calls sensitive content, partly at carriers' request — a small, telling signal that the licensed incumbents read the experiment layer too, and occasionally reach into it, by request rather than sanction.
The same analytic role shows up in the blog's later output. A piece from September 2025 dissects the congestion that opened up between the three carriers' domestic networks from mid-2024, province by province — the kind of operational intelligence that used to circulate only inside carrier engineering departments, now assembled from the outside by a hobbyist with measurement points and patience. The experiment network is the instrument; posts like these are the published findings; and the audience — hosting resellers, route arbitrageurs, expatriate users optimising paths home — is precisely the commercial grey market this piece's terms of reference required be left to other profiles. The experimental layer does not sell to that market. It informs it, free of charge, which may be the most economically consequential thing it does.
The demand side shows in what the community gives away. In December 2022, weeks after the network first appeared, a participant in MoeIX — a self-described non-commercial exchange founded in 2020 "for learning and experimentation only", no commercial use, no service guarantees — left a comment on the operator's page offering free transit and a free virtual machine in Seattle, an offer preserved in the archived capture. Free transit, free ports, free pool prefixes: the experiment layer's suppliers compete on generosity, which is what markets look like when the constrained resource is participants rather than capacity. The network's own trajectory signals the other constraint. The 2023 capture lists five points of presence — Frankfurt, Dusseldorf, Amsterdam, Singapore, San Jose — while the current page lists two. Between those dates the routing table shows announcement counts drifting down from a lifetime total of 31 distinct prefixes to today's ten. A licensed operator shrinking 60 percent would be a distress story; a hobbyist doing so is a person with a job. The contraction dates the experiment's peak to 2023 and suggests its budget of attention, more than money, is the binding input now.
What would settle the open questions these signals raise? Whether the customer range serves third parties would be settled by a single route or reverse-DNS trace tying a stranger's service to Sunflyer space, or by any resale listing; none was found. Whether the operator's standing converts to income would be settled by employment or consulting traces, which this research did not pursue past the public record. Whether the community's free-supply equilibrium holds would be settled by watching what Zappie, Securebit and the pool administrators do to retail prices as the registry's own charges evolve — a question with a date attached, discussed below. Each of these is observable from public data; the stratum's transparency means its open questions are, unusually, resolvable by anyone patient enough to watch.
What the stratum is for
Step back from the single network and price the layer as a phenomenon. On the supply side, its existence demonstrates that the global registry system now retails participation at around a hundred euros a year all-in, and that a globally distributed cottage industry — 87 sponsored numbers at one Delaware LLC alone — profitably intermediates between that system and individuals, including individuals in jurisdictions whose own registries start four figures higher and refuse natural persons entirely. On the demand side, it demonstrates that in China there are engineers willing to pay foreign tariffs, in foreign currencies, under their real names, for capabilities with no revenue attached: routing experience, interconnection experience, the ability to hold a fraction of the address plane and shape its announcement. The licensed market offers them no product. CNNIC's smallest unit is a /32 aimed at organisations with customer plans; the carriers sell connectivity, not participation; the universities that once provided this kind of sandbox route everything through CERNET's gateways. The unmet demand the experiment layer reveals is not for bandwidth — it is for standing: the ability to be a network rather than merely to use one.
There is an older model for this equilibrium, and China itself runs it: amateur radio. The state licenses individual operators to hold call signs and transmit on real spectrum, under content rules, precisely because a supervised hobbyist stratum produces engineers, emergency capacity and a census of the technically able, at no fiscal cost. The experiment-network layer is amateur radio without the licence class — the same social bargain, minus the paperwork, because the paperwork was never created. Individuals wanting to operate in the routing system found their call signs in a foreign registry instead, and the state's acquiescence functions as the licence it never wrote. The difference is jurisdictional: a call sign is revocable domestically, while a foreign-registered number is not, which may explain both why the tolerance persists — revoking it would require an enforcement action rather than an administrative lapse — and why it stays informal, deniable on both sides.
That reframes who competes with whom. The Sunflyer Global Experiment Network competes with nothing in China's licensed market, because nothing there is substitutable for it; its actual substitutes are other tuition instruments — overlay communities that simulate routing without real resources, free tunnel services, employer lab time — all of which it beats on realism at a price within a student's reach. Conversely, the licensed triopoly loses nothing measurable to the stratum today. The cost it bears is subtler and longer-dated: the most detailed public map of its international interconnection economics was written by this layer, for free, and is maintained more currently than anything the carriers publish. A market that refuses to sell participation to its most motivated students should expect them to study it from outside, on Swiss and Delaware paper, and to publish what they learn. The experiment network's real output, in other words, is legibility — of the global system to its operator, and of China's closed interconnect to everyone else. Both outputs are produced at a cost of a few hundred dollars a year, and both would be priced far higher by any institution that had to buy them.
For the directory's purposes, the entity classification should be read with its evidence boundary in view: this is a registered network operator in form — regional scope, exchange memberships, transit contracts — with the substance of a personal research project, documented across one live domain, registry objects in a foreign region, and a handful of community traces. Nothing in the record suggests undisclosed scale, and the scale that is disclosed is the point.
The facts that would move this judgement
The reading offered here — a lawful-by-design personal experiment whose economics measure a licensed-market gap — rests on a small number of load-bearing facts, each of which could move. Evidence of consideration flowing against the customer range would move it furthest: one invoice, one resale thread, one paid tunnel, and the network crosses from the 1996 regulation's RMB 15,000 hobbyist tier into the Telecommunications Regulations' business-penalty tier, and this piece's tolerance analysis inverts. A first publicised enforcement action against any mainland operator of a personal foreign-registered system would reprice the entire stratum's risk overnight, from rounding error to career threat; none is known to date, and that absence is itself part of the price. On the supply side, the registry's own economics carry a date: RIPE members vote on 2027 charging options that include per-resource models, and any change that raises the EUR 50 line materially — or any decision by the few sponsoring firms to shed hobbyist books, as their rising paperwork costs already tempt them to — would raise the grey stack's floor several-fold and thin the layer mechanically. A CNNIC or APNIC product genuinely sized and priced for individuals would do the opposite, collapsing the 25:1 anchor ratio and, with it, the arbitrage that makes the offshore route rational. Finally, the identity reading would move if the person incorporated: a Suzhou company appearing in the enterprise credit system under any Sunflyer-adjacent name would signal the experiment maturing into what its customer-range vocabulary rehearses. Each of these is checkable from public sources on any given week, which is the appropriate cadence for a subject whose entire existence is a set of public records with a person behind them.
Evidence register
The RIPE Database objects for the autonomous system, its organisation, sponsoring registry and address ranges (queried 3 July 2026 via the registry's public interface) carry the identity, dates, the customer-range naming and the Zappie, Securebit and pool provenance. RIPEstat and Hurricane Electric's routing observatory carry the announced-prefix counts, upstream relationships and announcement history. PeeringDB carries the network's self-declared class, traffic band, policy and exchange ports, and the LOCIX Frankfurt record carries the free-port terms and participant count. The operator's site and its November 2023 archived capture carry the stated purpose, the funding remark, the China-node peering refusal, the five-to-two point-of-presence contraction and the MoeIX free-transit comment. CNNIC whois carries the 2013 domain registration and registrant identity. The RIPE NCC Charging Scheme 2026 carries the EUR 1,800, EUR 1,000, EUR 75 and EUR 50 figures. Securebit's tariff pages carry the CHF 19/15 IPv6 and CHF 300-a-month IPv4 prices; Vultr's plan list and BGP documentation carry the USD 2.50 instance and announcement terms. CNNIC's charging policy and application rules carry the RMB 10,000 fees and /32 minimum; Sina's 2020 report carries the secondary-sourced IPv6 discount; APNIC's fee table carries the AUD figures. The Telecommunications Regulations and 1996 Provisional Regulations texts on the Cyberspace Administration's site carry the licensing, gateway and penalty articles; the 2017 ministry notice coverage and CERNET clarification carry the cross-border channel rules. The interconnection dossier, its V2EX reception and the MoeIX site carry the community-signal passages. The measured 87-number sponsorship count derives from the inverse registry query linked in the body and is reproducible by any reader.