Another Corporate ISP, LLC makes a local connectivity bill depend on upstream routes and field repair

Summary

• Another Corporate ISP, LLC is a specific legal identity, despite its generic wording. ARIN registers the San Francisco company as the holder of AS32329, whose network name is MONKEYBRAINS, while Oakland's June 2026 legislation explicitly identifies the company as doing business as Monkeybrains.
• The operating network is unusually tangible. The provider says more than 1,000 Bay Area buildings carry access points, customer antennas need unobstructed rooftop line of sight, and the access layer combines high-capacity wireless links with fibre. That density can spread site and transit costs across many accounts, but it also makes roof permission, building power, cabling and field response part of the service.
• Public routing views show a live internet edge, including a 100 Gbps SFMIX connection and observed connectivity to Hurricane Electric and Wave Broadband. They do not show whether the two upstream paths enter over separate fibre routes, whether every aggregation node has an alternate path, or how traffic performs during failover.
• Oakland's adopted 2026 authorization could materially alter the network by giving the company no-fee access to city fibre for ten years in exchange for free or low-cost service, including a commitment covering up to 2,500 affordable-housing units. It is an authorization and deployment commitment, not proof that all fibres, buildings or in-unit connections are already operating.
• The resulting network evidence grade is Medium. Current company, municipal, internet-number and routing records establish an operating regional ISP, but public evidence is insufficient to certify its ring topology, site-level backup power, physical route diversity, busy-hour headroom or measured repair performance.

The generic legal name resolves to a specific Bay Area network

The name Another Corporate ISP, LLC reads like a generic corporate description. In this case, it identifies a real legal and operating boundary. ARIN's current record for AS32329 names the autonomous system MONKEYBRAINS and places the registrant, Another Corporate ISP, LLC, at 933 Treat Avenue in San Francisco. The record was originally registered in April 2004, is active, and directs its technical contact to a Monkeybrains address. Oakland City Council file 26-0759 removes the remaining ambiguity: the city describes Another Corporate ISP, LLC as doing business as Monkeybrains and as a local internet service provider.

That identity link matters because the legal company, customer brand and routed network are three views of the same operating system. A bill may say Monkeybrains. Internet routes identify AS32329. A public contract names Another Corporate ISP, LLC. Each view answers a different question. The brand describes the retail relationship; the autonomous-system record shows who controls routing policy for a set of internet addresses; the legal name establishes who can sign a fibre licence, employ field crews and carry service obligations.

The public footprint is no longer especially thin once those names are joined correctly. Monkeybrains says it began as a dial-up provider in 1998 and now operates a hybrid fibre and high-capacity wireless network. Its current company history claims more than 25,000 accounts across the Bay Area, more than 50 technicians and 99.98% uptime. A 2023 filing by the company in a San Francisco Board of Appeals case described roughly 60 employees, 22,000 subscribers and free service to more than 8,000 subsidized-housing units at that time. The differing account figures are not necessarily contradictory: they were stated several years apart and use slightly different language. Both remain first-party representations rather than audited subscriber disclosures.

Public institutions provide stronger corroboration of actual work. San Francisco reported in its 2019-20 housing progress report that the city's Fiber to Housing partnership with Monkeybrains had connected 5,000 households across 36 housing communities. Oakland's 2026 agenda report says the provider operates a high-capacity wireless network serving more than 20,000 locations and has a history of deploying free or low-cost service. Those records do not validate every marketing number, but they establish that the company is more than an ASN registration or a retail website. It has delivered access networks, worked inside multi-unit housing and remained legible to two city governments.

The useful question is therefore not whether Another Corporate ISP, LLC exists. It is how the company turns a low local bill into a functioning path, and where that path can break.

A thousand rooftops substitute for one continuous last-mile cable

Monkeybrains describes itself as primarily a wireless internet service provider. Its technical service explanation says microwave links cover much of San Francisco and that access points sit on more than 1,000 buildings. A new location receives an outdoor antenna on the roof. From there, Ethernet cabling runs to a unit, telecommunications room or building patch panel, and the customer supplies a router to create the indoor Wi-Fi network. For a large apartment building, an installation can include both a receiving antenna and another antenna that extends the mesh.

This is a distributed access plant. Instead of digging a dedicated strand from a street conduit to every house, the provider reuses elevation, line of sight and short building cable runs. Some rooftops are customer endpoints; others also forward or distribute capacity. A dense city helps because one elevated node can see many nearby roofs, and every new suitable building can create another place from which the network reaches around an obstruction.

The physical economy is attractive. Trenching streets, obtaining rights of way and building fibre drops are expensive and slow. A compact outdoor radio, mount, power supply and cable can connect a building with much less civil construction. The provider says typical receiver equipment for a large multi-unit building consumes 10 to 20 watts, with less than 100 watts for larger bandwidth subscriptions. The small load lowers the cost of keeping a radio running, although it does not say how long batteries at any particular roof can support it.

The same architecture converts real estate into network infrastructure. A roof must be safe to enter. The building owner or manager must permit an installation in common space. A mount must survive wind and remain aligned. The radio must see another node without a building, tree or roof structure blocking the path. Cable must reach the communications room or apartment. Electricity must be available at the correct point. These are not ancillary details. They decide whether the provider can accept an order and whether a technician can restore it.

Monkeybrains states those limits plainly. Its service page lists excessive distance, obstructed line of sight, unsafe or unreliable roof access, steep roofs and inadequate mounting locations as reasons a building may not be serviceable. The residential offer likewise says every inquiry needs review and that management approval is required. Its broad coverage image is therefore an area of possibility, not a promise to every address inside the colour. The provider's current retail page estimates 100 Mbps for single-family homes and small apartment buildings and says larger buildings may receive up to 1 Gbps after assessment. Business service is advertised at up to 10 Gbps, but that top tier cannot be assigned to a building without a site and capacity decision.

The network is hybrid rather than purely wireless. The company says it operates fibre alongside wireless links, San Francisco's housing programme uses city fibre, and Oakland now proposes city-fibre integration. Fibre can carry high aggregate capacity between key locations while rooftop microwave handles the building reach. It can also create a resilient alternative to an aerial path. But the word hybrid does not reveal topology. It does not show which rooftops have fibre, which rely on one wireless backhaul, whether loops close into rings, or whether two apparently different links converge in the same conduit or facility.

That gap is central. More than 1,000 rooftop nodes indicate reach and operational experience. They do not automatically indicate 1,000 independent routes. A mesh can route around a failed radio only where alternative links have adequate signal, routing policy and spare capacity. A chain of roofs can look dense on a map while many buildings still depend on the same aggregation site. The installed count is evidence of scale; the resilience value depends on connections between the nodes.

The low bill is a density calculation

The company's home page advertises residential service at $35 a month, month to month, with no data cap, plus an installation fee. The number is commercially striking in San Francisco, but it is not a complete description of cost. It works because the provider's design attempts to share expensive components across many nearby subscribers.

One roof agreement can bring an antenna to a building containing dozens or hundreds of homes. One high-capacity backhaul can aggregate many access nodes. One SFMIX port and a small set of transit relationships can carry traffic for the whole customer base. A local field team can move between sites without crossing a state. Density turns a capital-intensive connectivity business into a sequence of repeatable building projects.

Multi-unit housing is especially important. The radio and roof visit are relatively fixed costs; the revenue or public benefit rises with every unit reached through the building's internal wiring. That is why the distinction between homes, accounts, locations and units matters when reading scale claims. Twenty-five thousand accounts are not necessarily 25,000 independent rooftop links. Thousands of subsidized units can share a smaller number of property connections. A network with fewer external access links than subscribers can still be large, but its common points need enough capacity and a credible repair plan.

The Oakland programme makes this economic mechanism explicit. The city's May 2026 agenda report says smaller providers told Oakland that access to municipal fibre and vertical assets could extend reach and reduce deployment cost. Regional providers also said low-income areas can be hard to enter because customer acquisition does not always recover the required capital. Under the authorized arrangement, Monkeybrains would receive city-fibre access at no charge while delivering and maintaining free or low-cost broadband in the project areas. The public asset reduces the private backhaul cost; the provider contributes operations, building connections and ongoing service.

That trade can support a low customer price, but it does not eliminate operating expense. Radios fail. Roofs are replaced. Fibre gets cut. Switches, optics and power supplies age. Transit, insurance, vehicles, storage, colocation, monitoring and salaries continue every month. A $35 plan leaves less room for repeated truck rolls than a premium service, so first-visit diagnosis, standardized equipment and route-level fault isolation matter. The provider's ability to keep the bill low is inseparable from its ability to avoid unnecessary field work and to concentrate repairs where one intervention restores many users.

Affordability also changes the cost of failure. For a household using a low-price plan because larger providers are unaffordable, a prolonged outage may not have an equivalent fixed alternative. For a free housing connection, the subscriber may not hold an individual retail contract that makes remedies obvious. Public-private programmes therefore need operating measures, not only connection counts: building uptime, peak-period performance, time to acknowledge faults, time to restore service and a clear demarcation between city, housing provider and ISP responsibilities.

The ownership boundary changes from roof to roof

The path begins inside premises the ISP does not own. A customer generally supplies the indoor Wi-Fi router. Monkeybrains supplies or installs the external receiver and cable path, but a building owner controls roof access and may control risers, telecom closets and patch panels. The electric utility supplies commercial power. In a city-fibre partnership, the municipality may own the backbone while a housing provider owns internal wiring. At the internet edge, a colocation operator owns the facility and upstream carriers own their long-haul routes.

Each boundary affects restoration. If the outdoor receiver has power and signal but a customer's router fails, the access network is healthy even though the user sees no service. The company's support guidance asks customers to compare performance directly at the supplied cable or power injector with performance over their own Wi-Fi. That is a sensible demarcation test. It isolates the provider path from indoor radio interference and an underpowered home router.

At a multi-unit building, the division is harder. The rooftop link can work while an old Ethernet run, switch, patch panel or unit cable fails. Oakland's report acknowledges this by including site-specific capital improvements, in-building wiring, network equipment and in-unit connections in the municipal project. Fibre reaching a property is only an installed upstream asset; it becomes usable service after the building distribution layer has been surveyed, upgraded, powered and tested.

Building access law can improve entry without making every roof unconditional. San Francisco's Article 52 gives occupants of qualifying multi-occupancy buildings a framework for choosing communications providers. Oakland maintains a reporting process for denials of internet choice under its own ordinance. Federal rules also protect certain antennas. In 2021 the FCC expanded its over-the-air reception devices rule to cover qualifying hub and relay antennas, recognizing that modern fixed-wireless networks use dense small antennas to overcome obstruction and distance limits.

Those protections have limits. The company's own explanation notes that landlords or associations may regulate common roof areas, while safety and historic-preservation restrictions can still apply under federal rules. A provider may have a legal path to request entry but still need a safe mount, insurance documentation, a wiring route and coordinated maintenance. Rights reduce arbitrary exclusion; they do not turn another party's building into an ISP-owned tower.

The same logic applies in Oakland. City-owned fibre remains a public asset. The June authorization allows a licence and integration; it does not transfer ownership of the municipal network. Housing organizations control their properties. Sonic would maintain the separately leased dark-fibre segment that fills a gap in the city's backbone. Monkeybrains would deliver and maintain service over the resulting access arrangement. A fault may therefore need one organization to identify it, another to grant access and a third to repair the failed asset.

Resilience is strongest when those boundaries are written before failure: who monitors optical loss, who carries spares, who may enter a roof after hours, who powers the building switch, who communicates with residents, and what restoration time each party accepts. The published records describe broad responsibilities but do not expose a complete fault matrix or service-level schedule.

AS32329 has a visible edge, but logical diversity is not trench diversity

An independent ISP needs more than radios. It needs a routing edge that can exchange traffic with the rest of the internet. AS32329 is that public edge for Another Corporate ISP, LLC. ARIN registration establishes administrative control. Current route observers show the network originating address space and exchanging paths with other autonomous systems.

PeeringDB's current record for AS32329 classifies Monkeybrains as a regional cable, DSL or ISP network. It lists one public exchange connection: an operational 100 Gbps port at SFMIX, with both IPv4 and IPv6 addresses, at Digital Realty's 200 Paul facility in San Francisco. The record describes an open peering policy and a traffic range of 50 to 100 Gbps. Because PeeringDB entries are maintained by network participants, they are strong evidence of intended interconnection and facility presence, although they are not utilization measurements or a contractual audit.

The provider's colocation page independently identifies 200 Paul Avenue as one of its two San Francisco colocation locations and says it peers with more than 50 networks there through SFMIX and AMS-IX. It describes more than 50 Gbps of capacity at 200 Paul, compared with a 10 Gbps fibre link at its Mission District facility. These are first-party capacity statements. The page does not give a current circuit inventory, and its comparison table should not be read as proof that every access route can use all of that capacity.

Public BGP views add another layer. Hurricane Electric's view of AS32329 shows 35 originated IPv4 prefixes, one originated IPv6 prefix, no RPKI-invalid originated routes in its current summary, and observed connections including Hurricane Electric AS6939 and Wave Broadband AS11404. BGP.tools similarly identifies Hurricane Electric and Wave among visible upstream paths and shows the SFMIX exchange connection. Cloudflare Radar reports active announcements and estimates a customer population of about 14,000 users, a methodology-based figure that should not replace the provider's own account count.

Together, the records support a live and independently routed network with public peering and more than one visible upstream organization. That is materially stronger than an ISP name attached to a dormant ASN. It still leaves the crucial physical questions unanswered.

Two upstream AS numbers do not guarantee two building entrances. Separate carriers can lease fibres in the same conduit, cross the Bay on the same vulnerable route, terminate in the same room or depend on the same power plant. Public route collectors show which networks can appear next in an AS path; they generally do not show street routes, conduit ownership, optical protection, handoff addresses or whether a second path has enough capacity for a full failover. A 100 Gbps exchange port also does not provide universal internet transit by itself. Peering reaches participating networks under routing policies; transit is still needed for destinations not covered by direct or exchange relationships.

Nor does a large port settle local bottlenecks. A customer can be limited by a 100 Mbps retail tier, a shared rooftop sector, a lower-capacity mesh hop or a fibre uplink long before traffic reaches SFMIX. Conversely, a roof can have excellent signal while a congested or failed edge makes the internet unusable. Capacity must be traced end to end, and redundancy must be tested at every aggregation point rather than inferred from the strongest component.

The 99.98% claim needs a denominator

Monkeybrains currently presents 99.98% as an uptime performance figure. If it represented every minute of a full year under a conventional availability calculation, the unavailable allowance would be about 105 minutes. The company does not publish enough methodology on the page to know whether that is the intended denominator, the measured period, the population included, or the exclusions.

An honest availability figure needs to say what is counted. Is it core reachability, an average across building links, the share of accounts online, or the ability to reach a probe from outside? Are scheduled maintenance, customer power loss, failed home routers and unsafe roof access excluded? Does one building with 300 units count as one endpoint or 300? Does degraded service below the purchased rate count as unavailable? Different choices can move the number substantially while all being labelled uptime.

The architecture makes a single percentage especially difficult. A routing edge can stay available while one rooftop cluster fails. Most buildings can remain reachable while a fibre cut removes an affordable-housing site. A customer antenna can be online while a saturated sector makes video calls unusable. A robust report would therefore separate edge availability, building-link availability and performance during busy hours.

The public record does not supply that report. The company's support page does offer an outage map for entire buildings and notes that a connection down for less than 15 minutes may not appear. That is useful operational transparency, but it also defines a visibility threshold and excludes individual-unit failures from the map. The map at a single moment cannot validate a long-term percentage.

For this reason, 99.98% should be treated as a first-party performance claim rather than a verified resilience result. It suggests the company measures or markets availability and sets a high expectation. Evidence that would strengthen it includes a published calculation, at least twelve months of building-level history, separate planned and unplanned downtime, packet-loss and latency thresholds, and the distribution of repair times rather than only an average.

Failure path one: the roof loses line of sight, access or power

A rooftop network can fail without a dramatic regional incident. A new structure can obstruct a path. A mount can shift in wind. Water can enter a connector. A roofer can disconnect cable. A building switch can lose power. A property manager can change locks or require a new access procedure. Any of these can isolate a building while the core remains healthy.

Line of sight is a design constraint and a maintenance condition. A good installation includes fade margin and a clear path, but cities change. Construction cranes, rooftop plant, trees at lower edges and neighbouring development can alter a path after activation. Where the network has another visible node with spare capacity, technicians may repoint the link. Where it does not, recovery can require a new relay roof, a higher mount or a wired alternative.

Local power creates a layered failure. The outdoor radio and building network equipment need electricity; so does the customer's router. A battery at the provider radio cannot power an entire apartment building. A generator at a data centre cannot keep an unprotected roof switch alive. CISA's communications dependency guidance treats electricity, information technology and transport for generator fuel as linked dependencies. California's regulator similarly recommends backup power, redundant networks, hardening, temporary facilities, coordination and staffing in its communications resiliency programme.

Those standards do not prove that every urban rooftop requires a 72-hour generator, and the CPUC's priority rules focus particularly on high-fire-threat districts and minimum communications. They do show why a vague statement about backup is insufficient. A provider needs to know which nodes aggregate the most users, what load their radios and switches draw, how long installed batteries last at end of life, and whether a technician can safely swap or recharge them during a wider outage.

The company's colocation disclosures illustrate how site-specific the answer can be. It says 200 Paul has substantial diesel generation and redundant power, while the Mission District colocation location has only three to five hours of short-term backup. Those statements concern the two colocation environments, not the 1,000-plus rooftop access points. No public inventory gives backup runtime for the rooftop layer. Core resilience should not be projected onto access sites without evidence.

The field consequence is straightforward: restoration time begins with detection, but it is governed by access. A remotely diagnosed radio cannot be replaced until a trained person, a spare and roof permission arrive at the same place. During wind, lightning or unsafe conditions, a responsible operator may have to delay elevated work. The customer experiences the delay as an outage even when the provider already knows the cause.

Failure path two: fibre or upstream connectivity disappears

Wireless last mile avoids a cable cut between each home and the street, but the wider network still uses fibre. The provider's hybrid core, colocation links, city-fibre partnerships and carrier handoffs all create cut exposure. One excavation can remove a path serving many wireless nodes if their backhaul converges before the damaged segment.

Oakland's own planning supplies a concrete warning. The 2026 agenda report says a cable deployed along the Bus Rapid Transit corridor had experienced major outages from fibre cuts and left the route unstable. The city chose a 3.1-mile Sonic dark-fibre lease along International Boulevard to complete a missing backbone segment; Sonic would maintain it for the 15-year term. That is useful evidence of a route problem and a selected remedy. It is not evidence that the finished Oakland network has a complete ring or that every Monkeybrains building receives two physically separated paths.

At the internet edge, the visible Hurricane Electric and Wave relationships reduce dependence on a single upstream organization. Their resilience value depends on implementation. If both handoffs are at one facility, a building power event or common router can still remove them. If one circuit is sized mainly as emergency capacity, a failover can preserve reachability but create congestion. If routes are filtered incorrectly or automatic preference is not tested, a healthy backup can remain unused.

CISA's network resiliency guide makes the distinction explicit: services bought from different carriers can share a central office, point of presence or physical asset, defeating the expected redundancy. The guidance is written for public-safety communications, but the engineering principle applies to a regional ISP. Independence has to be proved at the conduit, entrance, facility, power and routing levels.

The public routing evidence is therefore encouraging but incomplete. Active prefixes, RPKI-valid routes, a major exchange port and two observed upstreams establish a serious edge. They do not settle route geography. The decisive records would be simplified physical-path attestations, two independently powered edge locations, tested failover results, and utilization data showing that the surviving path can carry peak demand.

Failure path three: the surviving network runs out of usable capacity

An outage does not need to make every light go dark. Congestion, packet loss and unstable latency can make a connection functionally unavailable for work, telehealth or voice while basic pages still load. Shared wireless networks are particularly sensitive to where capacity is aggregated.

The advertised number at the home is a ceiling for a particular service configuration, not a reserve commitment. A 100 Mbps estimate does not state how much capacity remains at the rooftop sector in the evening. A 1 Gbps large-building possibility does not reveal the uplink shared by that building. A 10 Gbps enterprise offer does not mean every roof is connected at 10 Gbps. The company correctly says each location requires assessment; external analysis should preserve that condition.

Failure can increase load on the paths that remain. If one mesh link drops, traffic may move through another roof. If one upstream fails, all internet traffic may shift to the other. If a fibre-fed building becomes a relay for more wireless sites, its backhaul can become the new bottleneck. Resilience requires spare capacity at the moment of failure, not merely unused ports on a normal day.

The FCC's fixed broadband availability guidance helps define the lower bar: providers report locations where facilities are built and a standard installation can be completed, and fixed-wireless coverage can rely on propagation calculations with specified loading assumptions. Availability does not certify measured busy-hour throughput, failover headroom or repair performance. A serviceable map and a resilient network answer different questions.

Useful evidence would show the distribution of speeds and latency by building during peak hours, sector or aggregation utilization, oversubscription policy, and performance while one major backhaul is deliberately removed. No such current results are public. The 100 Gbps SFMIX port shows that exchange capacity need not be the smallest link, but it says nothing about the many links before it.

Oakland adds fibre, public obligations and new demarcations

The most important current development is OaklandConnect. In July 2024, the California Public Utilities Commission awarded the city up to $14,026,946 for its last-mile project. Oakland added a $2 million local match, producing a programme of roughly $16 million. The city's 2025 broadband plan calls for a publicly owned hybrid middle-mile and last-mile network, more than 12 miles of new municipal fibre and connections to affordable housing and the state's middle-mile system.

On 16 June 2026, Oakland's council adopted an authorization covering three related pieces. First, the city can lease the Sonic dark fibre needed to close the International Boulevard gap. Second, it can connect housing partners and fund the property-level improvements required for reliable in-home service. Third, it can enter a ten-year, zero-dollar fibre licence with Another Corporate ISP, LLC, doing business as Monkeybrains, allowing the provider to integrate city fibres into its service network.

The exchange is not simply free infrastructure for a private company. City staff says Monkeybrains would deliver and maintain free or low-cost high-speed service across areas including East Oakland, West Oakland, Fruitvale and Downtown, with free fibre internet for up to 2,500 affordable-housing units. The city selected housing partners and identified priority properties, while retaining public ownership of the fibre. The provider's operating experience in San Francisco's Fiber to Housing programme is part of the city's justification.

This can improve both economics and resilience. Municipal fibre can remove expensive leased backhaul, reach buildings that do not have a useful rooftop path, and provide high-capacity aggregation for nearby wireless links. A city route can diversify a building previously dependent on one aerial chain. Public investment can pay for the internal wiring that turns a fibre handoff into in-unit service.

It can also create new common dependencies. If many buildings aggregate onto one city segment, a cut there becomes more consequential. If the city, Sonic, housing partner and ISP divide maintenance, fault escalation has more handoffs. If grant-funded construction installs equipment without funding timely replacement, usable capacity can degrade after the build. A durable partnership needs lifecycle budgets, spare ownership, monitoring access and clear restoration authority.

Most importantly, authorization is not completion. The council action permits agreements and sets terms. It does not show that the licence has been executed, every construction milestone has been met, every priority property has passed a survey, or 2,500 units are online. The CPUC award finances a planned network; it is not a speed test. The right way to track progress is through executed agreements, construction reports, accepted fibre segments, activated property counts and measured service after residents connect.

Oakland's high-speed internet request for information had already identified the useful ingredients: city rooftops, buildings, poles and fibre, plus open-access middle-mile investment. The 2026 decision moves those ingredients toward a specific operating arrangement. It raises the evidence grade for the company's institutional role, while leaving the final topology and operational result to be demonstrated.

Field repair is not overhead; it is part of the product

The company's architecture reduces trenching, but it does not remove labour. Its own wireless field-technician description calls for rooftop equipment installation, communications wiring, troubleshooting, ladder climbing, lifting masts and concrete blocks, detailed site reports and occasional evening or weekend rapid response. Those duties map directly to the likely failure points.

The labor has several specializations. An installer can replace a customer radio or cable. A network engineer can change routing or diagnose packet loss. A fibre technician can splice a cut strand and test optical levels. A qualified electrician can address a damaged power service. A building representative can unlock a roof or telecom room. Large incidents often require several of them in sequence.

National labor data underline the constraint. The Bureau of Labor Statistics says telecommunications technicians install and repair internet routers, radio equipment and fibre, may work on rooftops or towers, and face unusually high injury rates. Training takes months or years. A provider cannot instantly expand this workforce when a storm, heat event or construction accident creates simultaneous faults.

Local scale can be an advantage. A team based in San Francisco and West Oakland can know roof entrances, recurring obstructions, building wiring and the fastest way to reach a node. Standardized radios and preconfigured spares can turn a long diagnosis into a swap. A dense route schedule can lower travel time. The company's claim of more than 50 technicians suggests meaningful field capacity, but it does not reveal shift coverage, specialist mix, contractor dependence or the number of open repairs at peak demand.

Repair quality should therefore be measured in distributions. Median restoration time can look good while a small set of inaccessible roofs remains down for days. Averages can hide whether affordable-housing properties or small buildings wait longer. The useful figures are the 50th, 90th and 99th percentile acknowledgment and restoration times, separated by customer-premises, building access, radio, fibre, power and upstream causes.

Spare policy matters too. A regional provider needs compatible outdoor radios, power injectors, surge protectors, switches, optics, cable, mounts and weatherproofing close to the service area. It needs configuration backups and a process for replacing a failed node without introducing a routing loop or security error. The public record establishes a field workforce and a local headquarters; it does not disclose spare coverage or repair-time performance.

Who is affected when a shared node fails

The impact follows the aggregation tree. A failed customer router affects one household. A failed building switch may affect every connected unit. A lost rooftop relay can isolate other roofs behind it. A fibre cut near an aggregation point can remove several neighbourhood clusters. An edge or upstream failure can reach the full network.

The customer mix makes those failures economically and socially different. Residential users lose work, school, entertainment and communication. Small businesses can lose card payments, hosted applications and voice. Event customers may lose ticketing or production links. Colocation customers depend on facility power and external paths. Affordable-housing residents receiving city-supported service may rely on it precisely because another fixed subscription is not affordable.

San Francisco's recent budget material says its Fiber to Housing programme provides free in-unit internet across more than 7,400 affordable-housing units at 210 sites, while Oakland plans another large property set. Those figures describe public programmes and should not all be assigned to the company's retail count. They do show why building-level continuity matters. One shared fault can affect many people whose service has become part of housing infrastructure.

Not every affected user needs the same design. A best-effort home plan may be acceptable without a formal restoration guarantee. A clinic, emergency shelter, business or large housing property may need a second provider or a physically distinct backup path. Customers considering resilience should ask where the demarcation sits, whether the alternate connection shares the same building power and fibre entrance, and how long their own router and local equipment remain powered.

The provider can reduce impact by publishing building-level status through a separately hosted channel, giving realistic restoration windows, and stating when customer equipment is the cause. Its outage map is a useful start. The stronger form is a history that lets customers and public partners see how often faults occur, how broadly they spread and how long they take to repair.

Customer reports are signals, not an uptime audit

Unofficial accounts broadly fit the engineering picture, but they cannot settle it. A May 2026 Oakland discussion among customers includes praise for price, speed and local support alongside reports that reliability can vary by neighbourhood and that wind, rain or a PG&E outage have coincided with interruptions. Other posts report very few failures over long periods. The sample is self-selected, addresses and service configurations are not verified, and several comments explicitly rely on memory.

What the discussion suggests is variation by building and route, which is plausible for a rooftop access network. It does not prove that weather caused a particular fault, that the entire network is less reliable than fibre, or that current performance at one Oakland address predicts another. Nor can customer praise validate 99.98% uptime.

The evidence that would settle those questions is comparatively simple: timestamped building-level outage history, cause codes, weather and utility correlation, and measured packet loss and latency from consenting customer endpoints. Until then, customer reports are useful for forming questions about local failure domains, not for calculating network availability.

What would prove the bill is resilient

Another Corporate ISP, LLC now has enough public evidence to support a concrete verification standard. The company does not need to reveal security-sensitive maps or customer locations. It can demonstrate resilience through bounded operational facts.

First, the access layer needs a topology account. For representative building types, the provider could state how many independent backhaul paths exist, whether alternate nodes use a different roof and power source, and what percentage of accounts can be rerouted after one aggregation-node failure. A ring claim should identify where the ring closes and whether both sides have sufficient peak capacity. Merely counting radios or buildings is not enough.

Second, the power account should be site-specific. Core facilities, major aggregation roofs and ordinary customer endpoints have different requirements. Publishing backup categories, tested runtime bands and the share of accounts behind each category would be more informative than one network-wide statement. The three-to-five-hour Mission colocation disclosure shows that the company can communicate limits plainly; the rooftop layer needs comparable clarity.

Third, upstream diversity should include physical separation. The public record already shows Hurricane Electric, Wave and SFMIX. The missing fact is whether critical handoffs use distinct entrances, conduits, facilities, edge routers and power. A concise third-party attestation could answer that without disclosing exact routes. A scheduled failover should then show convergence time, packet loss and peak utilization on the surviving path.

Fourth, capacity should be measured where users encounter it. Building and neighbourhood distributions for busy-hour throughput, latency and packet loss would distinguish an installed link from a usable one. Results during maintenance or a failed backhaul would reveal whether redundancy preserves only reachability or also preserves service quality.

Fifth, repair needs public operating measures. The provider could report fault counts by cause, median and tail restoration times, repeat truck rolls, spare-related delays and the share of incidents waiting on third-party access. Oakland and San Francisco could require property-level service results for publicly supported connections while protecting household privacy.

Finally, the Oakland expansion needs milestone discipline. Executed licences, accepted fibre segments, completed building surveys, activated units, in-building wiring acceptance and ongoing performance should be reported separately. That prevents a grant award, a passed resolution or a fibre route on a plan from being mistaken for working in-unit capacity.

These measures would also clarify the economics. If a $35 service sustains strong busy-hour performance and rapid repair across a dense rooftop network, the operating achievement is more significant than the low price alone. If outages cluster at access-controlled roofs or one shared upstream corridor, the same data identifies where the next dollar should go.

A credible operator with an incompletely visible resilience design

Another Corporate ISP, LLC is a rare case in which a deliberately generic legal name conceals a locally distinctive network. The company behind it has operated for decades, controls AS32329, connects at SFMIX, combines fibre with rooftop microwave, employs a substantial local team and participates in major housing-connectivity programmes. Current municipal and routing records support continued operation.

The network's strength is the same feature that creates its obligations. More than 1,000 building access points can turn Bay Area density into affordable service and route around some street-level construction costs. They also create more roofs, power supplies, cables, permissions and physical interventions to manage. Public fibre can lower backhaul cost and improve reach, but only after buildings are wired and operating duties are assigned. Two visible upstreams improve logical choice, but only physically separate routes deliver full failure independence.

The appropriate evidence grade is Medium. The identity, access technology, service geography, public programmes and live routing edge are well supported. The grade cannot be Strong because no public evidence demonstrates end-to-end ring topology, independent upstream trenches, rooftop backup runtime, failover headroom, spare coverage or measured repair times. Those are not decorative details around the service. They decide whether a local connectivity bill buys a path that remains usable when one of its shared physical links fails.