AVK-Computer Ltd makes a Lyubertsy connectivity bill depend on upstream routes and field repair

Summary

• The operating company behind the generic name is AVK-Computer Ltd. Its current company page describes a Russian telecom operator that says it has worked since 2000, serves more than 12,000 residential and organisational subscribers and expands its own fibre transport network in Moscow Oblast. Its contact page names the legal company, Lyubertsy office, tax number 7721237513, registration number 1027700040928 and a 24/7 subscriber line. This is evidence of a real regional operator, not of global service.
• The access network is physical and address-sensitive. AVK's current 1Gbps plan says a building may need infrastructure changes before speeds above 100Mbps are possible, that actual speed can vary with channel load and external factors, and that Wi-Fi performance is not guaranteed. Its private-sector plan says connection cost depends on site-specific factors. Historical first-party vacancies add the missing physical detail: fibre was run over poles, roofs and telephone ducts, into powered cabinets in technical rooms.
• AS48781 is live and well observed. A 10 July 2026 RIPEstat routing snapshot shows eight IPv4 announcements covering 18,944 addresses, visibility at 326 of 327 IPv4 RIS peers and 26 observed neighbours. RIPEstat routing consistency sees both registered upstreams, AS50509 and AS39156, in BGP. That is meaningful logical diversity, but no public record shows that the two upstream circuits leave Lyubertsy by separate ducts, buildings, power feeds or metro routes.
• Interconnection capacity is not customer capacity. PeeringDB reports regional scope, 10-20Gbps traffic and four operational 40Gbps PITER-IX ports, while a newer bgp.tools observation sees additional PITER-IX locations and two upstreams. Port speeds cannot be summed into a guaranteed 160Gbps or greater customer service pool, and multiple remote exchange sessions can still share one local transport path. PeeringDB lists no AVK interconnection facility and no IPv6 address on those four ports; RIPEstat sees no IPv6 announcement despite registered IPv6 resources.
• The network evidence grade is Medium. AVK has a live retail surface, licences, current tariffs, active routing, route-origin protection, local engineering roles and a long record of address-specific maintenance. The unresolved questions are the ones that decide recovery: access-ring topology, route and pole independence, node power and battery runtime, spare optics and active equipment, field crew depth, measured restoration times, busy-hour utilisation and the asset and operating boundary around a 2023 subscriber-contract change.

First, replace a generic label with a specific operator

The phrase "computer ltd" could describe almost anything. In this case, the network identifiers, company website and legal details converge on AVK-Computer Ltd. RIPE identifies AS48781 as AVK-COM-AS; PeeringDB's AS48781 record links the network to avk-com.ru; and the operator's contacts and legal-details page gives the same company name and a Lyubertsy customer office. The public identity is unusually coherent for a small regional provider.

The geographic claim needs an equally direct correction. AVK's own about page says it expands a fibre transport network into additional districts of Moscow Oblast. A preserved first-party news archive describes service in Lyubertsy, Nekrasovka, Ostrovtsy, Verkhnee Myachkovo, Malakhovka and Vereya, as well as later expansion around Gzhel and individual streets and housing developments. Those notices are historical, so they are not a current coverage map. They do establish that the operating story is local and regional. Nothing comparable supports an international retail access footprint.

"Global" can make sense as the scope of a website taxonomy. It does not describe this operator's service area. The network is registered in Russia, the office is in Lyubertsy, public projects cluster in and around Lyubertsy, and current third-party route observations place its operation in the Russian Federation. The responsible operating description is therefore a Lyubertsy and Moscow Oblast regional ISP with internet reach through upstream networks, not a global provider.

That distinction affects every resilience conclusion. A global network might distribute traffic, support teams and equipment across many metros. A regional ISP can know its streets and buildings better, dispatch locally and make construction decisions faster, but its customers may depend on a smaller set of metro exits, aggregation sites and specialist crews. AVK's public record supports the advantages and the concentration risk. It does not support the planned title if readers interpret "computer ltd" as the company's actual public name or "Global" as its physical footprint.

The legal company is not merely inferred from a route label. The current contact page gives tax number 7721237513, state registration number 1027700040928, a Moscow legal address and a Lyubertsy operating office. A current RBC company profile, drawing on Russian corporate and financial records, reports that the company was registered in March 2002, remained active at the end of June 2026 and had wired telecommunications as its main activity. That secondary profile is useful corroboration, but the operating analysis rests primarily on the company's live service pages, federal licence extracts and observable routing.

The asset is a regional fibre and cable network

AVK calls itself a universal communications operator. Its about page says the company provides internet access, telephony, leased communications channels, design, equipment supply, network construction and continuing project support. More importantly, it says the company continually expands its own fibre-optic transport network. The word "own" is a first-party ownership claim rather than an independently audited asset register, but it is more specific than a reseller presenting only another carrier's packages.

The commercial pages fit that fibre-centred picture. AVK's full tariff list offers apartment-oriented internet and television plans from 100Mbps to 1,000Mbps and separate private-sector plans from 30Mbps to 500Mbps. Its business menu includes internet, telephony, L2 channels and optical-fibre rental. The optical-fibre rental page says AVK can lease one or more strands in a fibre cable and can construct a fibre link to join a customer's offices. Its L2 service page offers a physical second-layer connection over dedicated copper, fibre or radio between offices and data-centre equipment.

That last reference to radio should not be inflated into a fixed-wireless access thesis. The current public retail pages are strongest on fibre, cable, building infrastructure and private-home connection work. They do not publish a tower list, licensed-radio footprint, sector map or fixed-wireless coverage checker. Radio is one possible bearer for an enterprise L2 circuit, not proof that AVK's residential network is tower-led. The evidence therefore narrows the primary asset to a regional fibre and cable access and transport network, with radio available in at least some bespoke links.

The access plant reaches several kinds of site. Apartment blocks create a density advantage: one building entry, riser or technical-room node can support many paying homes. Private houses are more expensive and variable because every extension can require more cable, pole or duct work and a longer drop. AVK's 30Mbps private-sector tariff says the cost of connecting a private home depends on many factors and is agreed individually. Its 500Mbps private-sector tariff repeats that qualification. That is the economics of distance, permissions and civil work appearing in the retail offer.

Historical company material makes the physical route clearer. An archived AVK vacancy page asked fibre installers to place optical cable on poles, across roofs and through telephone ducts, install metal cabinets in technical spaces and connect them to electricity. Because the page is old, it cannot establish the current construction mix or crew size. It does show the types of infrastructure the operator used and the dependencies that follow: pole access, roof access, duct availability, building technical rooms, local electricity and people who can splice and test fibre.

The current vacancies page confirms that the fibre work did not disappear from the public operating surface. AVK is advertising for a communications-line installer and a fibre-line installer-splicer. Duties include network construction, telecom equipment installation, fibre splicing, closure and patch-panel work, and commissioning active equipment. The posted schedules are five days a week, 08:00 to 17:00, with overtime for force-majeure circumstances or duty-roster work. A vacancy cannot prove filled headcount, but it is direct evidence that installation and repair labour remain part of the business.

A local bill pays for much more than internet transit

AVK's monthly broadband bill has to recover several cost layers. Transit and peering are the visible internet layer. Below them are metro transport, access fibre, building entries, pole or duct rights, powered cabinets, optical splitters or switches, customer drops, routers and television delivery. Then come labour, vehicles, splicing tools, spare optics, support staff, payment handling, licences, rent and the cost of returning to a site when the first repair does not hold.

The retail price ladder shows how the operator segments that stack. On the current tariff page, 100Mbps with television is advertised at 700 roubles per month, 500Mbps at 950 roubles and 1,000Mbps at 1,200 roubles. Private-sector plans are priced differently: the list shows 30Mbps at 600 roubles and 500Mbps at 1,750 roubles. Prices and promotions can change, and the pages use "up to" speeds. The important structural signal is that a sparse private-house connection costs more per delivered unit than an apartment connection sharing nearby plant.

AVK's 1Gbps tariff detail contains the most useful caveats. It says a connection can become chargeable if a building's network infrastructure cannot support the advertised rate. It warns that the customer's adapter and other equipment must handle more than 100Mbps, that maximum speed can vary widely with channel load and external factors, and that 1,000Mbps over a wireless connection is not guaranteed. It also says some cases require changes to the subscriber line. These are not minor terms. They map the boundary between a marketed tier and a usable service.

The operator says it has more than 12,000 subscribers. That figure remains on the current about page, but no date or audited definition accompanies it. It may count residential and organisational accounts, active lines, historical customers or another counting method. It should be treated as scale evidence, not a precise 2026 denominator. The RBC company profile reports 2024 revenue of 246.36 million roubles, net profit of 14.363 million roubles and an average headcount of 32.

Those figures make a functioning small operator economically plausible, but they do not reveal how many of the 32 people were field technicians, network engineers, support representatives or staff in unrelated project work.

Density can improve both economics and recovery. A technician working in one apartment building may restore many customers by replacing an aggregation switch or repairing one feeder. The same concentration increases the blast radius: one powered cabinet, feeder or inaccessible technical room can become a common point of failure for the building. Private homes reverse the pattern. A fault may affect fewer customers, but longer drops and scattered sites can consume more travel and repair time per account. AVK publishes enough to see this trade-off, not enough to calculate it by neighbourhood.

Installed capacity is not usable capacity

Three public numbers can easily be misread: a 1Gbps retail tier, 40Gbps exchange ports and 18,944 announced IPv4 addresses. They measure different things. The retail tier is a maximum access rate offered at a qualified premises. The exchange port is a physical or virtual interconnection interface into an exchange environment. The address count is routed number space. None is a direct measurement of the throughput available to one customer at the busiest hour.

At the internet edge, PeeringDB reports that AS48781 has regional scope, a Cable/DSL/ISP network type, a mostly inbound traffic ratio and a self-reported traffic level of 10-20Gbps. The same entry lists four operational 40Gbps connections to PITER-IX in Helsinki, Riga, Tallinn and Saint Petersburg. Each is marked as a route-server peer; none has an IPv6 address or BFD support recorded, and PeeringDB lists no AVK facility. Four 40Gbps interfaces should not be presented as 160Gbps of independent production capacity.

Traffic may be far below port speed, sessions may be delivered remotely, and the local tails may converge before reaching the exchange fabric.

A newer bgp.tools snapshot sees PITER-IX interfaces in Frankfurt and Moscow as well as multiple Saint Petersburg entries, alongside the four locations in PeeringDB. That observation suggests the interconnection surface has changed or that the two services expose different records. It does not identify contract capacity, physical handoff sites, common transport or whether every observed interface is part of normal forwarding. The defensible statement is that AVK has broad PITER-IX reach and many public neighbours, while the physical path from its local network to those sessions remains undisclosed.

RIPEstat gives a firmer measurement of the routed edge. Its AS overview marked AS48781 announced on 10 July 2026. The routing-status snapshot counted eight IPv4 prefixes covering 18,944 addresses, 26 observed neighbours and visibility at 326 of 327 full-feed IPv4 RIS peers. Its announced-prefixes view lists 91.107.0.0/18, its two component /19 routes, 95.128.240.0/21, two component /22 routes and two /24 routes from 185.33.244.0/24 and 185.33.246.0/24.

The aggregate and component routes overlap, so their nominal address sizes must not be added. RIPEstat already reports the unique announced space as 18,944 IPv4 addresses. The addresses also cannot be equated with customers or line capacity. A residential connection may use dynamic or shared addressing; a business plan may receive a real or static address; network equipment consumes addresses; and one subscriber may use several over time. AVK's 100Mbps plan says a real IP address is included, but that commercial feature does not turn IP inventory into a subscriber count.

IPv6 exposes another installed-versus-usable gap. PeeringDB reports one IPv6 prefix, and RIPE's consistency data lists 2a01:a9c0::/32 and a /34 in registry material. Yet the 10 July routing snapshot sees no announced IPv6 prefix and no IPv6 visibility. Registered address space is installed administrative capacity; a globally visible route is usable routing capacity; working IPv6 at a subscriber premise is a third condition. The public evidence reaches the first condition, not the last two.

Route-origin security is better evidenced. RIPEstat's RPKI validation endpoint reports a valid origin state for AS48781 on the main 91.107.0.0/18 route. Checks on 95.128.240.0/21, 185.33.244.0/24 and 185.33.246.0/24 also return valid. That reduces one class of accidental or malicious route-origin problem. It does not prevent fibre cuts, router failure, power loss, congestion or a correctly originated route from becoming unreachable behind a failed local link.

Two upstreams are useful, but their physical independence is unknown

AVK's internet edge is not a single-neighbour stub. RIPE's registered policy and current routing observations identify two upstreams: AS50509 and AS39156. RIPEstat's ASN neighbours view sees both on the upstream-facing side of AS48781, plus a wider set of peer-like or uncertain adjacencies and two observed downstream paths. bgp.tools also reports two upstreams and a much larger peer count. This is stronger evidence than a website claim of "redundancy" with no visible BGP support.

Still, BGP diversity is not civil-engineering diversity. Two sessions can terminate on separate routers while travelling through the same building entry. Two providers can lease fibres in the same duct. Separate metropolitan circuits can cross the same bridge, railway corridor or utility pole line. An upstream can be logically distinct but purchased through a common intermediary. Public route collectors see AS paths; they do not see cable sheaths, splice chambers, power panels or the point where two local tails become one.

PITER-IX reach adds route options and can keep popular traffic away from paid transit. Peering through exchange route servers can shorten paths, reduce transit demand and give AS48781 many reachable networks. The mostly inbound traffic ratio is consistent with an access provider whose users download more than they upload. But four or seven PITER-IX observations are not four or seven independent ways out of Lyubertsy. If the remote exchange sessions are delivered over one metro transport service, one local failure can remove several at once.

Nor does peering replace full transit. An exchange provides routes from participating networks or route-server customers; it does not necessarily provide a default path to the entire internet. The two upstreams remain important when a destination is not available over peering, when an exchange route server fails, when an exchange entity withdraws routes or when policy prefers transit. A resilient design would combine upstream diversity, exchange reach and physically separate local exits. The first two are visible. The third is not.

The current evidence also cannot establish headroom. PeeringDB's 10-20Gbps band is self-reported and broad. It may describe typical traffic, peak traffic or the network's chosen disclosure range. No public graph shows busy-hour utilisation by transit, exchange, access ring or neighbourhood node. A network can have two healthy upstreams and still slow down because an aggregation uplink, PON segment, building switch or radio L2 link is full. The retail pages openly acknowledge that channel load can move actual speed; they do not say where the bottleneck usually sits.

The first failure path is the access network

For most AVK customers, an upstream outage is not the only or even the most likely loss mechanism. The service must travel from a home or office through a drop, building or street node, local feeder and aggregation network before AS48781 can choose an internet route. A cut or failed switch anywhere in that chain can make two healthy upstreams irrelevant.

AVK's own notices show address-level and neighbourhood-level blast radii. A July 2023 maintenance notice said work on a main line serving the Samolet Lyubertsy development, including Vertolyotnaya and Druzhby streets, could interrupt service for five hours. The archived news page records earlier node upgrades affecting specified groups of buildings and streets, work on a third-party main cable that was expected to cause partial degradation, and one 2013 notice saying planned work by a backbone operator would leave customers without internet during the stated period.

These are scheduled notices, not an outage-performance history. They nevertheless reveal topology through impact. A "main line" modernisation that names a housing development implies a shared feeder or aggregation dependency. A node-equipment upgrade that lists several buildings implies a common node. Work by another backbone or cable owner that degrades AVK service shows that the operator boundary can include third-party infrastructure even when AVK owns much of its local fibre.

The old vacancy description adds the repair locations: poles, roofs, telephone ducts and technical rooms. Each has a different access constraint. A pole route can be exposed to vehicle damage, utility work, weather and attachment disputes. A roof route can require building or landlord permission. A duct route may be protected from weather but difficult to enter, trace and repair. A cabinet in a locked technical room can remain unreachable even when the technician and replacement switch are ready.

Historical notices also show that access permission can become the limiting resource. In December 2015, the archived site told subscribers that property managers had restricted entry to attics, basements and technical spaces over the holiday period, sharply limiting repair and connection work. The notice is too old to describe current agreements, but the mechanism is durable: the ISP does not control every room or structure that its cable crosses. Recovery depends on the person who has the key as well as the person who has the fusion splicer.

Power can turn a healthy fibre into a dead service

Optical fibre does not need electricity along every metre, but the network equipment around it does. Building switches, optical line terminals, media converters, radio links, routers, video systems and customer-premises devices all need local power. The current public site does not publish battery runtime, generator coverage, dual-feed arrangements or the number of unpowered passive segments. That absence prevents a strong power-resilience rating.

The historical vacancy page explicitly said installers placed metal boxes in technical rooms and connected them to electricity. The old news archive also carried a notice that municipal power repair at named addresses would cause internet interruptions. Those historical records do not prove that present nodes lack batteries. They do establish that local electricity has been part of AVK's failure chain.

Backup power must be assessed end to end. A battery in the core does not help if the building switch loses power. A generator at an office does not power a street cabinet. A protected access node does not keep a customer's router and optical terminal alive during an apartment outage. A business may have an uninterruptible power supply for its firewall but no power at the provider's handoff one floor below. The proper question is not "does AVK have backup power?" but "which devices on this particular service path remain powered, and for how long?"

The customer contract tier also matters. AVK's public pages advertise 24/7 service and support but do not publish a residential uptime guarantee, power SLA or restoration percentile. A corporate leased channel may be negotiated differently, especially where AVK provides a bespoke L2 circuit or fibre strand. Without a public service specification, no general guarantee can be transferred from the availability of the support phone to the availability of every powered network element.

Field repair is a capability and a queue

AVK visibly maintains local engineering capability. The current vacancy for a fibre installer-splicer requires fibre construction and maintenance experience, splicing, closure and patch-panel work, and active-equipment commissioning. The communications-line installer role covers network construction, structured cabling, access-control systems and video surveillance. AVK's corporate services page also markets design, construction and maintenance of security systems alongside business internet. Those skills overlap with the practical work of reaching, testing and repairing access plant.

The project pages show execution at local sites. AVK says it surveyed, designed, supplied, installed and then maintained systems at a Lyubertsy housing development, where it lists 93 cameras across entrance, outdoor and indoor placements. A Tomilino park project lists 14 cameras plus subsequent maintenance. A Panki and Ukhtomskaya station project lists 25 cameras integrated with the regional safety system. These are first-party case studies, not audited broadband repairs, but they support a local field-service capability beyond telephone support.

Capability is not the same as immediate availability. The current job adverts state a normal five-day daytime schedule and allow overtime or duty-roster work. The contact page calls the subscriber department 24/7. A person answering a call at 03:00 can identify an outage and open a ticket; that does not prove a splicer, driver, building key holder and spare switch can all arrive at 03:30. Public sources do not provide response-time commitments, crew rosters, mean time to repair or geographic dispatch coverage.

A 2018 holiday notice in AVK's archived news said subscriber service would remain available around the clock while one duty specialist covered main-line emergencies over the holiday. That is valuable historical evidence because it distinguishes the service desk from the field. It is not a current staffing statement. A current assessment would need the number of simultaneous on-call crews, skill mix, vehicles, splicers, optical test equipment, spare cable and active devices, and escalation arrangements with landlords, utilities and upstream carriers.

The regional-ISP labour problem is therefore a queueing problem. One crew can restore a high-impact feeder quickly if the fault is clear and access is open. The same crew can become a bottleneck when a storm, construction incident or power event creates several faults at once. A specialist may spend hours tracing intermittent optical loss while new tickets accumulate. The monthly bill must fund enough idle capacity for rare peaks, yet carrying too many idle specialists harms a small operator's economics. AVK's postings show that it hires the right kinds of people; they do not reveal where the company sits on that trade-off.

The operator boundary is visible but not complete

AVK's legal and network surfaces line up in several places. The current contact page names AVK-Computer Ltd and its registration details. The licences page lists communications-channel, data, local telephony, cable broadcasting and other authorisations. A linked federal register extract for channel services identifies the same legal company and says the licence runs to May 2028. RIPE registration and PeeringDB attach AS48781 to AVK-Computer.

One first-party notice complicates the customer-facing boundary. In September 2023, AVK published a change-of-contracting-party notice saying that the name and payment details of the company providing services under existing contracts would change from 30 September, while quality, tariffs and contract conditions would stay the same. It directed questions to a smartorange.ru support address. The notice does not, on its face, explain which company owned the fibre, which held each subscriber contract after the change, which employed field staff or which controlled AS48781.

The current evidence then swings back toward continuity. The 2026 contact page still presents AVK-Computer Ltd, the current vacancies use AVK's domain, and the RBC profile says AVK-Computer owns a SMARTORANGE trademark valid to 2030. That makes a brand or contracting reorganisation plausible, but it is not enough to reconstruct the exact legal and asset transfer. The safe conclusion is continuity of the AVK service and network identity with an incompletely disclosed subscriber-contract boundary, not a claim of acquisition or ownership change.

This uncertainty matters during failure. The customer needs to know who accepts the ticket, who can enter the building, who owns the failed switch, who calls the upstream carrier, who carries the spare and who owes any service credit. Those roles can sit with one company or several. Public service pages show a single front door; they do not publish a responsibility matrix. A strong resilience assessment would obtain the current standard contract and escalation schedule before assigning recovery obligations.

Who is affected when the network fails

AVK's customer surface extends beyond home web browsing. Residential plans combine broadband with IPTV, and the site also offers telephony and smart-home services. A feeder or node outage can therefore remove television and voice alongside internet access. A home that depends on cloud cameras, remote work, school services or app-controlled devices experiences a wider interruption than a speed-test failure.

Business customers face different dependencies. AVK markets continuous dedicated internet access, static addresses, L2 channels, fibre rental and telephony. Its corporate page specifically connects internet service with card-payment acceptance and cash-register connectivity. A lost line can interrupt point-of-sale authorisation, remote access, voice calls, cloud applications and inter-office traffic. A business buying an L2 path to data-centre equipment may lose a private network service even when public internet routes remain healthy.

Public-safety and municipal systems add another layer, though they should not be confused with the residential broadband network. AVK's business video page says it designs, approves, installs and commissions systems for the regional safety programme. Its public project list includes parks, housing, roads and railway stations. These pages show that AVK works on locally important communications systems; they do not establish that every camera traverses AS48781 or shares a broadband feeder. The appropriate conclusion is potential operational relevance, not a universal common-mode claim.

Failure impact depends on topology. If a housing development's internet, IPTV and building cameras share a feeder or powered room, one cut could affect all three. If their circuits and power are separate, the same physical incident may have a smaller effect. Public project descriptions list installations and maintenance but not circuit diagrams. Readers should resist joining systems merely because one company worked on them.

The two observed downstream AS paths also broaden the theoretical blast radius of AS48781 routing trouble. RIPEstat sees AS51686 and AS212095 on the downstream-facing side, while bgp.tools reports two downstreams. These AS numbers are routing evidence, not a basis for describing new corporate relationships. They show that an AS48781 routing failure may affect traffic originated beyond AVK's own prefixes. The scale, contracts and physical paths are not public.

Customer reviews are clues, not outage statistics

Unofficial market signals are unusually mixed. A current Yandex Maps review page includes customers praising same-day help, long periods of stable service and useful support. It also includes complaints about intermittent connections, packet loss, delays in field visits, equipment that could not initially support a requested tier and a weekend outage where the reviewer said a technician would not arrive until Monday. A 2GIS review page likewise contains both positive and negative experiences.

Those reports suggest sensible questions. Does experience differ by building, private-house area or access node? Are high-speed tiers unavailable on older in-building equipment? Are weekend restoration resources thinner than call-centre coverage? Do intermittent faults cluster around power, weather, overloaded access ports or customer routers? The reviews cannot answer those questions because the sample is self-selected, addresses and service tiers are often missing, customer equipment is uncontrolled and the operator's ticket records are unavailable.

An older 2IP provider review surface offers the same caution. Individual users report both satisfactory long-term service and recurring speed or interruption problems. No review proves a network-wide condition. A positive review cannot establish route diversity; a negative review cannot locate the fault inside AVK rather than Wi-Fi, customer equipment, a landlord's power or a remote service.

The evidence that would settle the signal is operational and address-specific: timestamped incident records, affected-node identifiers, ticket open and restore times, access technology, measured optical levels, port utilisation, CPE tests, power alarms and the reason for closure. Aggregated by neighbourhood and service tier, those records would show whether complaints are isolated customer-edge faults or repeat failures in shared plant. Until then, reviews belong in the question set, not the reliability score.

What meaningful redundancy would look like

At the internet edge, AVK can already point to two observed upstreams and extensive peering. The next proof would be physical: two handoffs on separate routers and power feeds, entering the network at separate sites over paths that do not share a local duct, bridge, pole line or carrier tail. A drawing need not reveal sensitive coordinates; an attestation of route and facility separation, plus failover tests, would materially improve confidence.

Inside the access network, a protected ring needs evidence that traffic can reverse around a cut. A fibre loop on a marketing map is not enough if both sides share one cable sheath or if the switching design does not reconverge. The useful measures are route separation, automatic protection behaviour, reconvergence time and the number of customers behind unprotected spurs. Building and private-house networks may have different answers.

At active sites, recovery requires power and spares. AVK could disclose classes of node, battery design targets, test intervals and the proportion covered by generators or mobile backup without publishing a security-sensitive inventory. Spare optics, switches, power supplies, splitters, closures and pre-positioned cable determine whether a technician completes a repair in one visit. Supplier lead time matters when the installed platform is old or uncommon.

For field operations, the decisive measures are not the call-centre hours but response and restoration distributions. The operator should know the median and 95th-percentile time to acknowledge, dispatch, access, diagnose and restore for feeder cuts, building-node failures, private drops, power events and upstream faults. It should also know how those times change at night, on weekends and during multiple simultaneous incidents.

At the customer edge, resilience becomes a product choice. A household may accept best-effort recovery and use mobile backup. A shop that cannot take payments may need dual access from providers with genuinely separate plant, plus a powered router that can fail over. A camera or L2 customer may require a protected path and explicit restoration terms. Buying two AVK circuits helps only if their common components are understood.

The operating verdict is real, regional and only partly transparent

AVK-Computer Ltd clears the basic operating test comfortably. It has a current retail site, current tariffs, a 24/7 subscriber contact, federal licence extracts, an active legal profile, live AS48781 routing, valid sampled route-origin authorisations, two observed upstreams, broad peering and visible demand for fibre construction and splicing labour. It also publishes local projects and maintenance notices detailed enough to connect the network to real streets and technical work.

The initial thin-footprint concern should therefore be upgraded to a confirmed operating footprint. The evidence grade should stop at Medium. The company's public materials do not show a route map, protected-ring coverage, independent upstream tails, exchange-delivery architecture, node power protection, spare inventory, current field-team depth, restoration performance or busy-hour utilisation. The 2023 contracting-party notice leaves a further question about who carries each operational obligation.

The regional-ISP category is supported, as are all three topics. Regional ISP economics appears in apartment density, individually priced private-house connections and infrastructure-dependent high-speed tiers. Local support labour appears in current installer and splicer roles, local project maintenance, building access and duty-roster language. Peering and transit appears in two observed upstreams, many neighbours and the PITER-IX footprint. None of those facts supports a global operating region.

The company-specific conclusion is simple. An AVK customer does not buy an abstract gigabit. The customer buys one qualified path through a particular building or street, powered equipment, local fibre, AS48781, two logical upstream choices and a repair organisation that must reach the fault. The bill is resilient only when those components fail independently enough, and when the right person, permission and spare arrive before the customer's tolerance runs out.