The clearest description of what it costs to run a network in China is not in any carrier's annual report. It is in the brief routing career of an animation studio.

On 25 June 2025, the Asia-Pacific registry's record system shows, autonomous system 139187 was registered under the name TWOTENNET to "WUHAN TWOTEN CULTURAL COMMUNICATION COMPANY LIMITED", with a technical contact named Han Taixi reachable at a mailbox on the studio's own domain, an address on the eighth floor of a creative-industry building on Guanshan Avenue in Wuhan's Optics Valley, and maintenance delegated to China's national internet registry. The same allocation round gave the company a portable block of 512 addresses, 165.101.70.0 through 165.101.71.255, recorded in APNIC's database under the same network name. That same fortnight, RIPE's routing archive begins to see half of that block announced to the world from the new number: first faintly, at about two dozen of the collectors' vantage points, then — by mid-July — at some 350 of them, which is to say essentially everywhere the global routing table is watched from.

For ten weeks, a mid-sized cartoon studio in central China was, in the technical sense that matters, an internet operator: it originated its own routes, under its own number, from its own address space. Then the record turns over. In the collectors' August bins the announcement's visibility collapses from 335 vantage points to 57, then 21, then nothing. From mid-August to late October the same /24 appears in the table originated by Google's cloud network instead. In mid-October it flips again, briefly originated by Amazon's, while the studio's own number flickers back for a dozen days and disappears for good. On 26 October 2025 a second autonomous system was registered — WINGMARK-AS-AP, held by "Wingmark Matrix (Wuhan) Computer Systems Co Ltd", listing the same rooms, 810-811 in the same building, and the same Han Taixi as contact — and from the following day to the time of writing it has originated the studio's 165.101.71.0/24 from behind Amazon's network. The other half of the block travelled the opposite direction: since 15 October 2025 the routing archive shows 165.101.70.0/24 originated by China Unicom's backbone, where it still sits today. As of 3 July 2026, autonomous system 139187 announces nothing at all.

That arc — ten visible weeks, then a partition of one small network between a state carrier and an American hyperscaler, with the company's own number left as an empty registration — is not a failure story, and reading it as one would miss what it documents. It is a price signal. Everything the studio tried, and everything it settled for, has a published tariff somewhere, and assembling those tariffs turns an obscure routing curiosity into a measurement of the precise economic space China's telecom triopoly leaves to the private layer beneath it.

Three carriers first, everyone else priced after

Start with the structure that makes the question interesting. China's fixed internet access market is, by the industry ministry's own 2025 statistical bulletin, a market of 691 million broadband subscriptions — and the bulletin counts them as belonging to exactly three companies: China Telecom, China Mobile and China Unicom. Telecom service revenue across the sector reached 1.75 trillion yuan in 2025; fixed broadband access alone brought in 289.6 billion. Divide the one by the other and the average Chinese broadband line yields the carriers about 35 yuan a month — a computed figure, but computed entirely from the regulator's published numbers. There is a statutory fourth backbone, the broadcasting network, and a long tail of licensed value-added operators — the ministry had issued 23,500 electronic licences for value-added telecom services by early 2025, as reported when the first foreign-invested trial approvals were announced — but the pipes, the addresses most Chinese eyeballs sit behind, and the interconnection points all belong to the three.

What the under-storey pays those three is not left to negotiation folklore; twice in the past decade the state has written the number down. Until mid-2020, the settlement price one backbone paid another for interconnection was capped at 80,000 yuan per gigabit per month, a standard preserved in the industry ministry's 2020 adjustment notice, republished by the education-network operator it also affected. That notice rewrote the hierarchy explicitly: from 1 July 2020, China Mobile and the two incumbents would peer settlement-free, as Telecom and Unicom always had between themselves; the broadcasting newcomer and one state-owned niche carrier would get their rates cut by at least 30 percent from that 80,000-yuan standard; the academic networks would connect free. Caixin's reporting put the value of the change to China Mobile alone at roughly two billion yuan a year. Note what the decree did not do: it said nothing about anyone below the named networks. The under-storey — licensed access resellers, data-centre operators, enterprises with their own address space — kept buying at retail, from tariff sheets the incumbents publish themselves.

The under-storey itself is legally elaborate. The ministry's licence taxonomy splits value-added telecom into a first category — data centres, content distribution, domestic virtual private networks, internet access provision — and a second, covering online information services and their kin. The first category is where anything resembling a private carrier must live, and it is deliberately hard ground: capital thresholds, provincial versus national scopes, foreign-ownership limits only now being loosened through a trial that began with thirteen approvals in early 2025. Thousands of firms hold these permits; almost none of them owns transmission. They rent everything — ports, circuits, address space, the very interconnection their customers pay them to arrange — from the three companies whose retail arms compete with them for the same enterprise accounts. The classic customer problems this layer solves are the ones the giants serve badly precisely because serving them well would cannibalise dedicated-line revenue: an office needing all three carriers' routes at once because its clients sit behind different backbones and inter-carrier paths degrade at peak; a campus wanting one bill for a mess of buildings; a firm needing addresses that survive a change of carrier. Every one of those problems traces back to the same structural fact — the three interconnect generously with themselves and grudgingly with each other, and everyone below pays for the difference.

Those tariff sheets are the honest map of the space the triopoly leaves, and they deserve close reading, because the same three companies that charge each other zero for interconnection publish prices for everyone else that run three orders of magnitude above their consumer offers for the same nominal bandwidth. That spread — not any licence fee, not any formal prohibition — is the wall around the niche this piece set out to price. And in June 2025, one animation studio walked up to the wall with its own ladder.

The studio behind the network name

The company that owns TWOTENNET is not obscure; only its appearance in a routing database is. Wuhan Twoten Cultural Communication Co., Ltd. — 武汉两点十分文化传播有限公司, trading as 2:10 Animation, the name a joke about the after-lunch minute when animators start drawing — was registered on 8 September 2011 with 18.14 million yuan of capital under founder Wang Shiyong, per corporate-registry mirrors. The studio itself is older: the domain 2-10.cn was registered on 9 March 2007, a fact preserved in China's national domain registry and verifiable through its public lookup service, and trade-press accounts of the company's funding history describe a handful of art students starting that year on pocket capital. The web-content filing trail is equally continuous: 鄂ICP备17001223号-2, first filed in June 2017 and most recently reviewed in December 2024, per the filing-database mirror, held by the same legal entity. The company's Chinese site footer displays a value-added telecom licence, 鄂B2-20170130 — a Hubei-issued second-category permit of the kind that covers commercial online information services, not internet access provision. An attempt to verify that licence in the industry ministry's public query system from outside China returned only the application shell, an outcome recorded here rather than smoothed over; the filing number and class are as the company's own footer states them.

What the studio sells is animation at the premium end of the Chinese market: cinematic trailers and character films for games — its own client roll names League of Legends, Honor of Kings, Arknights, Onmyoji and Dungeon & Fighter — plus original series, brand films, and contract work on theatrical features. The revenue logic behind that mix, as the founder described it at the 2021 raise, splits roughly a third of output into wholly owned original properties and two-thirds into commissioned work for platforms and consumer brands: the commissions pay the payroll at service margins, the owned properties are the call options, and the studio's persistent public claims — four hundred projects, twenty billion cumulative views, a hundred brand clients — are the marketing surface of that structure rather than auditable facts. The state has endorsed the model in its own currency: a national cultural-industry demonstration base designation in 2024, a national copyright exemplar in 2020, high-and-new-technology enterprise status, and a place in Hubei's "golden seed" listing-reserve programme, honours that carry tax and subsidy consequences in China's system and that provincial and industry profiles corroborate. Wuhan matters here too. The studio sits in Optics Valley, the city's subsidised technology quarter, where the founder has been explicit about staying — "Wuhan is our base", he repeated in November 2025 — which means second-class regional pricing on the carrier tariff sheets examined below, provincial patronage, and salaries a coastal studio cannot match. Who pays is therefore a short list with deep pockets: game publishers led by Tencent, NetEase, Riot and Hypergryph, the video platforms that commission series, and consumer brands. That client mix bought the studio an investor list unusual for central China. FreeS Fund led an A round in 2016; the company's English site names Alibaba Group among its backers from a 2017 round; and in September 2021 Pop Mart led a B+ round in the several-hundred-million-yuan range, joined by the listed publisher Chinese Online and the Arknights developer Hypergryph, with contemporaneous coverage framing it as Pop Mart's entry into animation. The 2021 reporting had the company at five hundred staff, releasing one to two works a week, and eyeing a 2022 listing. No listing followed; more on that below.

This is the entity the directory records, on the strength of its network registrations, under a regional internet-provider category — and the classification is worth a paragraph, because it is not exactly wrong. The directory's evidence boundary is narrow: two domain references, one public website, a pair of supporting records. Seen only through registry glass, a company holding an autonomous system number, a portable address block and a network name genuinely is indistinguishable from a small carrier; that is the shadow any enterprise casts the moment it touches the numbering layer. The analytical task is not to correct the category but to explain why an animation studio came to cast a carrier's shadow — and what its subsequent retreat says about everyone else standing in the same light. The explanation begins with what the studio's business actually moves: terabytes of frames, review sessions with publishers in Shanghai, Shenzhen and Los Angeles, render jobs, and — per a September 2025 account of its AI unit's work — model-assisted production that cut a one-minute piece from roughly three months of turnaround to two weeks. A firm like that has the traffic profile of a small studio-cloud, distributed collaborators (its site claims a community of a thousand), and clients on both sides of the Chinese border filter. Multi-carrier redundancy and stable global reachability are not vanity for it; they are delivery deadlines.

The incumbent's own price for autonomy

Here is what exercising that autonomy costs at home, in the sellers' own words. China Telecom's internet dedicated-line tariff standard, 2016 edition, published on its Hubei portal and preserved by the web archive, is the reference sheet for exactly the product an autonomous-system holder needs: "BGP routed access", the flavour of connection over which a customer announces its own addresses. The posted network-usage fee for 10 megabits of BGP access with full transit is 7,600 yuan a month. One hundred megabits: 66,300. A full gigabit at the aggregation layer: 563,200 yuan a month — about 6.76 million yuan a year — with a domestic-only routing scope at 335,100 and reach limited to the carrier's own network at 308,200. The static-route version, without the right to speak routing protocol, runs from 4,200 yuan a month for 10 megabits to 3.23 million a month for 10 gigabits, with the sheet noting that gold, platinum and diamond service tiers add 10, 20 and 30 percent, that Wuhan sits in the second of three regional price classes, and — the detail that defines the under-storey — that these prices apply to "internet access service providers and other government-enterprise customers" alike. The reseller layer buys from this sheet too. Its margin is whatever survives contact with it.

The sheet's small print prices the ingredients of autonomy separately, which makes the comparison with the registry route exact. A static-route customer gets a fixed quota of addresses with the line — one with 10 megabits, sixty-one with a gigabit — and pays 50 yuan per month for every address beyond it. At that posted rate, renting the 512 addresses the studio holds would cost 25,600 yuan a month from the carrier; the national registry allocates the same quantity, portable across any carrier, for thousands of yuan a year. The sheet even publishes an interpolation formula for unlisted speeds, a linear ladder between adjacent tiers, so that no bandwidth escapes the price structure. None of this is hidden knowledge; it is the incumbent stating, in a public document, that what costs pocket change at the registry layer costs a salary at the routing layer — because the registry sells numbers and the carrier sells the only thing that makes numbers mean anything in China, which is reachability through its network.

Now put the consumer sheet beside it. A published 2023 tariff filing from the same carrier's Guangdong operation prices a 500-megabit consumer line at 290 yuan a month — with a dynamically assigned private address, and a clause voiding the service if used "for operational purposes or resale". The enterprise sheet prices 500 megabits of static dedicated access at 196,900 yuan a month and the BGP flavour at 327,200. Same wire speed, same company: a factor of 679 for the right to a public, stable presence, and 1,128 for the right to announce your own addresses. The two sheets are seven years apart, and the state has leaned on the enterprise side in the interval — the 2021 Government Work Report ordered a further 10 percent cut in average small-business line prices, part of a multi-year campaign, and negotiated discounts off list are the norm — so the live multiple is lower than the printed one. But no amount of mandated trimming closes a three-orders-of-magnitude gap, and the no-resale clause on the cheap sheet is precisely the fence that keeps the gap collectible. For scale: even the 80,000 yuan per gigabit per month that the regulator once let the incumbents charge a rival state backbone is a seventh of the 563,200 they post for a private customer's full-transit gigabit. The triopoly's price for peers was set by decree; its price for the under-storey is set by the absence of one.

The registry side of autonomy, by contrast, is almost free, which is what makes the comparison clean. China's national registry operates an address-allocation alliance whose charging policy is published: 10,000 yuan to open an account, 10,000 yuan per autonomous system number, and an annual address fee ladder that starts at 7,000 yuan for a /24-sized holding and climbs to 540,000 for the largest tranches; its application rules cap a first allocation at exactly the 512 addresses the studio received. Community documentation of the schedule as executed in 2022 corroborates the figures and notes annual increases of 4.75 percent from 2025 through 2027. Registry summaries of the alliance's number-resource rules add the operational condition: an applicant network must be multi-homed, establishing routing sessions with one licensed operator within a month of assignment and with at least two within three months. Roughly 27,000 yuan, in other words, buys the paperwork of independence at the schedule's bottom rungs. The paperwork then requires you to buy the 7,600-yuan-a-month line item — twice.

The arithmetic of a silent autonomous system

Lay the ledger out explicitly, separating what the documents say from what must be inferred. The evidence: year-one registry cost of about 27,000 yuan, from the published alliance schedule (opening fee, one system number, bottom-rung address fee; the studio's /23 sits one rung higher on a ladder the policy prints in full). The cheapest compliant way to keep that number speaking at home, per the incumbent's posted sheet, is two 10-megabit BGP-routed lines — the alliance's two-session requirement — at 7,600 yuan a month each with full transit, or 182,400 yuan a year, before the service-tier surcharges the sheet lists and before any negotiated discount it does not. That is the floor for autonomy exercised domestically at a bandwidth no serious studio could actually work with. Scale the same sheet to a single full-transit gigabit and the list price is 6.76 million yuan a year — roughly a third of the operating company's entire registered capital, every year, for one line from one of the two carriers the rules require.

The inference layer sits on top, flagged as such. What the studio actually pays China Unicom for the arrangement now visible in the routing table — its 165.101.70.0/24 originated from inside Unicom's backbone since October 2025, the studio buying reachability as a service rather than announcing it as a peer — is not published anywhere; the observable fact is only that this product exists and was chosen. What the affiliated Wingmark vehicle pays Amazon to originate the other /24 from the cloud is likewise private; Amazon publishes no separate charge for announcing customer-held addresses, and the real cost sits in instance-hours and egress fees that depend on traffic volumes nobody outside the building knows. What can be said with confidence is directional: both chosen arrangements convert a six-figure-a-year posted fixed cost into usage-priced services, and both surrender exactly the thing the June experiment briefly held — routes under the company's own number, at home.

One more computed line completes the picture of what the carriers are defending. The regulator's bulletin implies the average consumer broadband line yields about 35 yuan a month; the posted price of a single 100-megabit BGP-routed enterprise line, 66,300 yuan, therefore equals the monthly revenue of roughly 1,900 consumer households. A carrier that let enterprises slide from the second sheet to the first would be trading its densest revenue for its thinnest, which is why the consumer contract carries the no-resale clause and why no amount of bandwidth abundance — 95 percent of Chinese lines now run at 100 megabits or better, a third at gigabit — has ever been allowed to leak into the enterprise price. The scarcity being sold is not capacity. It is permission.

Against the studio's own economics, the choice prices itself. The flagship entity's registered capital is 18.14 million yuan; the 2021 round that reportedly ran to several hundred million yuan was raised to fund production industrialisation, with the founder telling interviewers the incremental investment in production technology alone would top a hundred million — money with a claimed return in output per week, not in routing tables. A registry-mirror snapshot shows the flagship entity reporting 25 insured employees against the group's public claim of six hundred professionals — a single-sourced figure treated here as a signal about group structure rather than headcount truth, since Chinese groups routinely distribute staff across affiliates like the film subsidiary that hires under its own name. But the juxtaposition frames the decision cleanly: for a company whose production arm was simultaneously reporting that AI tooling had collapsed a three-month deliverable into two weeks, 182,400 yuan a year of posted connectivity floor — approximately the annual cost of two junior animators in Wuhan, an inference from regional salary norms rather than any document — buys either ten megabits of sovereign routing or a measurable amount of additional output. The routing table records which way that comparison went.

The detour through Sacramento and Seattle

How the experiment actually ran, and where it went when it ended, is the most instructive part of the record, because at no point did the visible version of it touch the domestic tariff sheet at all. During the ten globally visible weeks, RIPE's route-replay data shows every path to the studio's announcement passing through a single upstream: autonomous system 7720, registered in APNIC's database as SKYWOLF-AS-AP, Skywolf Technology LLC, of a suite address in Sacramento, California — with the American backbone Lumen carrying it onward in almost every observed path. A Sacramento-registered LLC holding an Asia-Pacific network number and reselling transit is a recognisable species: the offshore intermediary layer that supplies Chinese networks with the global-side connectivity the domestic sheet prices out of reach. Nothing in the public record says what Skywolf charged, or where the session physically ran; what the record does say is that when a Wuhan studio first exercised its Chinese-issued network resources, the exercise was visible via California and never, in any collector's data, via a Chinese carrier's transit.

The single-upstream structure also explains, without needing any insider account, why the domestic half of the experiment never appeared. The alliance's own conditions require sessions with two licensed operators inside three months; the visible record shows one offshore reseller and a fading announcement at precisely the three-month mark. Whether the studio balked at the second 7,600-yuan line item, failed to negotiate one, or ran a domestic session invisible to foreign collectors cannot be determined from outside — Chinese announcements that never propagate past the national border can hide from the collectors' vantage points, a limit of the method this piece flags rather than waves away. What is not ambiguous is the direction of travel afterward.

The retreat was equally structured. From mid-August 2025 the /24's origination moved to Google's cloud — the standard bring-your-own-addresses arrangement, in which the cloud announces a customer's space from its own edge — and in mid-October to Amazon's. October was the pivot month on every axis at once: on the fifteenth, Unicom's backbone began originating the domestic /24 and Amazon the global one; on the twentieth, the studio's aut-num object was modified for the last time; on the twenty-sixth, the Wingmark number was registered; on the twenty-seventh, it went live. Then, on 26 October, the Wingmark vehicle acquired its own network number, registered not through the national alliance but directly with the regional registry under its own response-team object, and began originating the block from behind Amazon the next day, where it remains: every current path to 165.101.71.0/24 today shows Amazon's network immediately upstream of Wingmark's number. Ten days later the studio's PeeringDB entries were created — the peering industry's directory recording, with tidy irony, a restrictive-policy network with no exchange ports and nothing to peer, the bookkeeping arriving after the autonomy had already been re-housed.

The choice of registry for the second number is a detail with content. The studio's resources came through the national alliance — CNNIC paperwork, CNNIC maintenance, Chinese fee schedule, and the implicit legibility to domestic authority that channel carries. Wingmark's number was registered directly with the regional registry, with its own incident-response object and no national intermediary: the channel used by Chinese networks that intend to operate on the global side of the wall, where what matters is standing with APNIC and the clouds rather than with the ministry. Same office, same engineer, two registries — a company keeping one foot in each numbering regime, which is the under-storey's survival posture expressed in database objects.

Wingmark Matrix itself rewards a closer look. Its public storefront — Chinese legal name 翎迹天算(武汉)计算机系统有限公司, address on the same Guanshan Avenue — sells elastic compute instances, application hosting, a security layer called CloudInsight, a web firewall and a house CDN, claiming a hundred-plus partners and 4,582 services in stable operation, under Microsoft partner-network branding. Those claims are the company's own and unverified; an attempt to pull its file from the national enterprise-credit system was refused at the gate from this vantage point, an outcome stated here plainly. But the technical linkage to the studio is directly observable in the public DNS: the studio's own websites resolve through hostnames carrying the CloudInsight label on Wingmark's furrydns.net infrastructure before terminating, for overseas visitors, on Amazon's CloudFront edge; and Wingmark's flagship chat service — a character-persona platform in the furry idiom, of all things — resolves to a node literally labelled "wh210" (Wuhan, two-ten) on a China Unicom Hubei address. One building, two legal entities, one technical team: the licensed, filed, domestic-facing studio rides Unicom lines and Huawei-cloud distribution at home, while the young computer-systems affiliate holds the offshore-facing resources, the direct registry membership, and the Amazon origination.

That division of labour is the real answer to the assignment's question about what space the triopoly leaves. It is not a market segment; it is a seam. A private network firm in China cannot out-buy the incumbents at home — the tariff sheet sees to that — and cannot lawfully carry traffic across the border on its own account. What it can do is arrange itself so that everything requiring a licence, a filing or a domestic route is bought retail from the carriers, while everything requiring autonomy — address portability, origination control, global delivery — is exercised through clouds and registries beyond the sheet's reach, with an affiliate's name on the paperwork. The studio's ten weeks were the experiment; the current architecture is the finding. The under-storey survives not by occupying the room the three leave, but by straddling its wall — thin, legal in each part, and dependent at every point on suppliers who are also, on one side, its price-setters.

Signals from around the seam

The soft evidence around this record points in one direction: a technically ambitious group improvising infrastructure faster than its paperwork, in a sector whose economics were shifting under it. The scale claims are the loudest signal. Six hundred professionals, a thousand-strong talent community, twenty billion cumulative views, per the studio's own materials, against 25 insured staff at the flagship entity in a mirror snapshot and a funding-round census of five hundred in 2021 — a spread that would be settled instantly by consolidated group filings, and that until then reads as a marketing perimeter drawn around a network of affiliates, standard practice in Chinese creative industries and commercially meaningful mainly as a sign that the listable entity, whenever one is assembled, has yet to be drawn. The listing itself is a second signal: 2021 coverage spoke of a 2022 flotation, provincial programmes named the company a listing-reserve "golden seed", and five years later no prospectus has surfaced in any search performed for this piece — consistent with the wider freeze in Chinese consumer-content listings, and a reminder that the group's equity cushion, however many hundreds of millions it raised, has had no public mark-to-market since Pop Mart's cheque.

Continuity signals cut the other way, against any reading of the group as fragile. The domain has been registered since March 2007 and is paid through 2033; the web archive holds captures of the studio's site from September 2007 onward without a gap worth noting; the content filing has been renewed on schedule since 2017; the affiliates keep hiring through mainstream recruitment platforms rather than shedding. Whatever the insured-headcount snapshot means, this is a nineteen-year-old company that pays its renewal fees early — the behaviour of an owner who expects to still be operating the brand a decade out, and a useful prior when weighing whether the network experiment was a lark or a first draft.

The infrastructure signals are odder and more interesting. A furry-fandom chat platform running on a Unicom line in the studio's building; a DNS estate scattered across whimsical domains; a cloud storefront with no visible content-filing number of its own; a network number registered one day and originating routes the next — these are the traces of engineers building at side-project speed inside a funded company, not of a compliance department building a carrier. The pattern suggests the group's network ambitions are real but young, driven by product needs (global delivery of heavy media, AI-era compute, a consumer chat service) rather than by any plan to sell connectivity. What would settle the reading: a value-added licence upgrade in Wingmark's name appearing in the ministry's public registry, which would signal a genuine hosting business formalising onshore; or the quiet lapse of the CNNIC resources at renewal, which would signal the experiment written off. The August 2025 production-cycle testimony — three months to two weeks — cuts both ways: it explains both why a studio suddenly needs cloud-scale infrastructure and why every yuan of fixed network cost now competes with tooling that multiplies output directly.

The facts that would move this judgement

This piece reads TWOTENNET as a priced retreat: evidence that the room beneath China's triopoly is too expensive to occupy in its official form even for a well-funded enterprise with a real need, and that the practical under-storey now runs through clouds and affiliates instead. Several observable facts would move that judgement. If autonomous system 139187 reappears in the global table with two domestic carriers as upstreams, the studio will have bought compliant autonomy at something like the posted floor, and the conclusion about the wall's height softens accordingly. If Wingmark's name surfaces in the ministry's licence registry with data-centre or access classes, the group is becoming a licensed operator in earnest, and this profile's subject changes character from enterprise to carrier. If the group files for listing anywhere, audited financials would replace every mirror-sourced figure used here, and the unit-economics passage should be rebuilt from the prospectus. If the regulator extends settlement reform downward — publishing, for instance, a wholesale BGP-access rate for licensed resellers the way it once published the 80,000-yuan inter-backbone rate — the entire spread this piece measures would compress, and with it the rationale for the offshore seam. And if enforcement attention ever turns to Chinese-allocated address space originated from foreign clouds, the seam itself closes, which would be visible within days in the very routing archives this piece is built on. Two quieter facts would also matter. Renewal behaviour at the registry — whether the studio keeps paying the annual fees on a number it no longer uses — will distinguish an abandoned experiment from a held option; the alliance's published grace rules mean a lapse would surface in the database within months of a missed payment. And any audited disclosure of the group's connectivity spending, in a prospectus or a bond filing, would replace this piece's list-price floor with an actual transaction price for the seam architecture, the one number the public record cannot currently supply. Each of these is checkable from public records; none requires anyone's cooperation. That is the one respect in which the subject makes an analyst's life easy: a company that expresses its strategy in the global routing table publishes its next move by definition.

Evidence register

APNIC's record system and whois carry the two autonomous-system registrations, their dates, holders, shared address and contact, and the studio's portable /23. RIPE's routing archive, route-replay and live looking-glass data carry the announcement chronology: the June–August 2025 visibility arc, the Skywolf-and-Lumen path structure, the Google and Amazon originations, the Unicom origination of the second /24, and the current silence of AS139187. PeeringDB carries the organisation and network entries and their November 2025 creation dates. The industry ministry's 2025 statistical bulletin carries the sector revenue, subscriber and broadband-revenue figures; the 2020 settlement notice, in the education network's republication, carries the 80,000-yuan standard and the peering reforms, with Caixin's report corroborating the sums. The national registry's charging policy and application rules carry the alliance fees, allocation cap and multi-homing conditions, with an independent community analysis corroborating the schedule as executed. The archived Hubei Telecom tariff standard carries every dedicated-line price cited; the Guangdong tariff filing carries the 290-yuan consumer price and no-resale clause; the 2021 Government Work Report excerpt carries the mandated line-price cut. Corporate-registry mirrors carry the incorporation date, capital, legal representative and insured-headcount snapshot; the filing-database mirror carries the ICP trail; the company's sites carry the licence number, client roll and scale claims. Funding coverage from 36Kr's syndication and 21st Century Business Herald carries the Pop Mart round; The Paper carries the AI production-cycle testimony; Wingmark's storefront and the public DNS carry the affiliate's services and the technical linkage between the two companies. Query attempts that failed — the ministry's licence-lookup shell, the enterprise-credit gate, an encyclopedia mirror — are noted in the text where their absence bears on confidence.