Summary
- CRISP can show a long-lived enterprise fibre business, an ICE pole-sharing agreement, its own autonomous system, two visible upstream paths and 10 Gbps exchange ports, but none of those facts discloses whether a customer's primary and backup circuits share the same pole corridor, splice closure, powered site or international exit.
- The legal counterparty has changed materially: Livister acquired CRISP in 2025 and later regulatory records describe CRISP as absorbed by Ideas Gloris, so a resilience review now has to identify which company owns the cable, signs the service agreement, controls field repair and supplies upstream capacity.
- A credible redundancy claim would require site-specific route drawings, common-structure analysis, power-runtime evidence, port and backbone utilisation, tested failover records, repair hand-offs and restoration results; the public record supplies fragments of that picture, not the complete chain.
Beneath one ICE closure, two circuits can become one fault
The most revealing place to test CRISP's promise is not a sales presentation or an internet speed test. It is beneath an ICE pole on a wet road in Costa Rica's Greater Metropolitan Area, where two service cables may leave the same closure and run in the same direction. One cable can be labelled “primary” and the other “backup” in a customer diagram. They can terminate on different routers and carry traffic through different logical paths.
Yet if both fibres are lashed to the same sequence of poles, pass through the same splice enclosure, or enter the same powered aggregation site, a vehicle strike, falling tree, fire, pole replacement or failed power feed can remove both at once.
That example is a test, not a claim that every CRISP pair shares a route. The public record does not provide the coordinates needed to make that claim. What it does establish is that Costa Rica Internet Service Provider S.A., usually called CRISP and marketed as Luminet, signed a shared-use contract with the Instituto Costarricense de Electricidad for telecommunications pole infrastructure. The 2015 official notice identifies the parties and the contract file, but not the poles, kilometres, splice locations, route diagrams or restoration terms. It proves access to an important physical support system. It does not prove how extensively CRISP used it or whether any two customer circuits diverge.
The distinction matters because Luminet's own explanation of point-to-point fibre goes further than merely promising reserved bandwidth. It describes 1:1 service, symmetric performance, repairs measured in hours and, in its closing sales language, “real redundancy.” The same Luminet fibre comparison says point-to-point connections can exceed 1 Gbps and support service-level and mean-time-to-repair commitments, but it publishes no actual service-level schedule, route-separation standard, measured restoration distribution or excluded-cause list. The claims are directionally plausible. They are not a substitute for a circuit design.
Even the public availability process is site-specific. Luminet's coverage inquiry asks a prospect to place an address on a map and select a bandwidth band of 50–100 Mbps, 100–300 Mbps or more than 300 Mbps. That is evidence of a feasibility-led access business rather than a uniform national access product. The page's map is an address picker, not a route map: it does not reveal backbone rings, pole corridors, ducts, wireless fallbacks, aggregation nodes or protected zones.
There are therefore three different meanings of “diverse” in play. Commercial diversity means two purchased services or two invoices. Logical diversity means traffic can be routed through different devices, autonomous systems or exchange points. Physical diversity means the paths do not share structures, closures, entrances, power systems or repair dependencies that a single incident can disable. CRISP's public materials support the first two in some circumstances. They do not let an outside reader verify the third.
For an enterprise buyer, this is not semantic precision for its own sake. A nominal backup is valuable only across the failure modes the business is paying to survive. If a customer's two access circuits meet at the first pole, if two upstream sessions ride the same metro fibre, or if geographically separate routes depend on one unprotected router, the contract may still deliver two services while the physical network delivers one fate. The key question is not whether CRISP owns fibre somewhere along the path. It is where independence begins, where it ends, and who can restore each shared segment when it fails.
The CRISP name now sits on a changed legal boundary
CRISP's public identity is unusually easy to recognise and unusually difficult to map to the current operating counterparty. Its corporate site describes a Costa Rican fibre-connectivity provider serving the corporate market and gives a Río Segundo, Alajuela address. The site also contains unfinished generic text, which limits its value as operational disclosure. A separate contact page gives the full legal name, Costa Rica Internet Service Provider S.A., the same telephone number and an office in the airport-area complex. Luminet's company page presents the customer-facing brand, more than 15 years of experience and named commercial, engineering and service-desk roles. Together, these pages show a continuing public identity and a local support presence. They do not settle who now holds each contract, asset and operating duty.
The regulatory history does. SUTEL's April 2025 competition decision RCS-071-2025 authorised Livister Latam S.L.U. to acquire 100 per cent of CRISP without conditions. It describes CRISP as a specialist in point-to-point communications and fibre connectivity for corporate customers, operating under the Luminet brand and also selling managed SD-WAN, Wi-Fi, data-centre, cybersecurity and cloud services. It describes Livister as a UFINET Latam subsidiary and notes that the buyer's group already controlled UFINET Costa Rica, ADN Soluciones and Ideas Gloris, with Gold Data also considered in the competition analysis. The decision says CRISP focused on enterprise retail while the UFINET group focused on wholesale services. It also records the parties' expectation that CRISP would merge into Ideas Gloris in the short term, with Ideas Gloris surviving.
SUTEL's public concentration register marks the CRISP–Livister case concluded and accepted. Later records show that the proposed legal consolidation was not merely theoretical. In October 2025, the regulator considered the transfer of a block of 10,000 IP-telephony numbers and a special number after CRISP returned them because it would be absorbed by Ideas Gloris; the Council minutes emphasise continuity for end users and state that the technical interconnection arrangement was not changing. In February 2026, another set of Council minutes assessed how Ideas Gloris notified customers of contractual changes caused by the absorption. SUTEL found that the notice only partially met information requirements because it omitted the effective date of the contractual modification.
Those facts change the due-diligence question. “Does CRISP have a repair crew?” is no longer precise enough. A customer needs to know whether the contracting party is Ideas Gloris, whether the Luminet brand remains the operating front end, which company employs or dispatches the field team, who owns the fibre and optical equipment, and whether UFINET provides an affiliated wholesale segment. Those functions can sit in different companies without causing poor service. They can even improve resilience through scale.
But the allocation must be explicit if an escalation is to move rapidly from a service desk to a pole owner, fibre technician, equipment custodian or upstream carrier.
The public registers do not all update in a way that makes this transition obvious. SUTEL's current CRISP authorisation entry still shows the historic company name and the original RCS-489-2009 authorisation, later modifications and renewal. The regulator's operator list displays CRISP and Ideas Gloris as separate entries. Those entries are useful historic and administrative records, but they should not be read against the later absorption decisions as proof that two independent operating companies persist. They show why a buyer must ask for current corporate documents rather than infer legal responsibility from a search result.
The continuity is real as well. The 2013 official register ties corporate number 3-101-525475, CRISP and the Luminet name to RCS-489-2009. A 2024 SUTEL session containing RCS-212-2024 renewed the authorisation for another five-year period beginning in December 2024. A long-running licence, a durable brand and a merger into a larger group can all be true at once. The operational consequence is that historical performance under CRISP cannot automatically be assigned to every post-merger supplier, while current group scale cannot automatically be treated as physically diverse capacity on a particular customer route.
A national authorisation is not a national route map
CRISP's authorisation covers data transfer, internet access, virtual private networks, IP telephony and IPTV throughout Costa Rica. “Throughout” is a statement about the permitted service territory. It is not evidence that CRISP has installed fibre in every province, owns a nationwide backbone, has two entrances to every industrial park or can provision a protected circuit at any address.
The difference between authority and infrastructure is especially important for a regional enterprise provider. Such a provider can combine its own metro fibre, rented pole space, purchased transport, data-centre cross-connects, wireless access and partner tails. That mix can be economically rational. Owning every metre would impose large fixed costs on a business whose customers are geographically uneven and whose circuits often require individual engineering. The risk appears when the sales shorthand “own fibre network” is allowed to stand in for a route-specific bill of materials.
Luminet's services page says the company provides advanced IP connectivity, dedicated internet over fibre or wireless, data links and corporate voice through its own fibre network. It also describes managed point-to-point MPLS connectivity between a customer demarcation and the Luminet data centre. These statements establish a genuine operating surface: access fibre, wireless alternatives, packet routing, customer demarcations and at least one data-centre role. They do not quantify route-kilometres, owned versus leased spans, the number or location of aggregation sites, or the share of customers whose last mile uses a third party.
That missing map should not be filled with guesswork. The Alajuela address does not prove the network core is there. The centre point on the coverage form does not prove a node exists at that coordinate. A national authorisation does not show a cable. A list of IP prefixes does not show a cable either. Each is evidence of a different layer.
A useful customer map would start at both demarcation points and trace each circuit through every splice closure, pole or duct sequence, handhole, building entrance, aggregation chassis, powered shelter, cross-connect, exchange and upstream hand-off. It would identify which spans are CRISP or Ideas Gloris fibre, which use ICE structures, which are leased from an affiliated or unaffiliated carrier, and which technicians have authority to open or move them. For a protected service, the drawing should mark every shared-risk link group rather than merely colour two lines differently.
The absence of a public map is not itself a sign of poor engineering. Detailed telecom routes are often commercially and physically sensitive. What matters is whether the provider can disclose enough under a confidentiality agreement to support a customer's resilience claim. A redacted drawing can omit street addresses while still showing separate corridors, separate entries, separate active sites and ownership boundaries. A provider can also furnish pole or duct identifiers for the customer's access zone without releasing its entire national plant.
The current public evidence leaves four map questions open. First, how much of the advertised own fibre is aerial and how much is underground? Second, where does CRISP's plant meet UFINET, Columbus Networks or other wholesale transport? Third, can two enterprise circuits be kept on different ICE pole sequences or different infrastructure owners from the customer site to the first resilient node? Fourth, are Costa Rica and Miami exchange appearances reached through physically separate international systems or through capacity that converges before leaving the country?
Until those questions are answered for a site, “national” should be read as regulatory reach, not physical ubiquity.
The fibre may be CRISP's while the structures and permissions are not
Shared poles split responsibility in a way that is easy to miss. The telecommunications operator can own the cable, closures and optical signal while ICE owns the pole and gives the cable a permitted position. That arrangement does not make CRISP a virtual provider. It makes the service dependent on two maintenance domains that must coordinate under safety rules, access terms and a commercial agreement.
Costa Rica's shared-infrastructure regulation makes the division concrete. Infrastructure owners must maintain the scarce resource so shared use can continue. Each telecom operator remains responsible for identifying its network elements; cables on poles are to be labelled at least every 250 metres and at transitions between aerial and underground plant. A request for shared use must specify individually identified infrastructure, a georeferenced route and any needs such as power, security and staff access. The resulting contract is expected to state the type and quantity of shared infrastructure, its geography and identifiers, the agreed route and additional facilities. In other words, the information needed to test CRISP's physical diversity should exist between the relevant parties even though the 2015 public notice does not reproduce it.
The regulation also shows why the commercial boundary affects restoration. If parties cannot agree on shared use, SUTEL can intervene. Infrastructure charges are negotiated under a cost-oriented method, and the infrastructure owner is entitled to compensation. A 2025 SUTEL pole-charge methodology formalises annual charges using recognised pole value, useful life, operation and maintenance costs and the number of users. Pole sharing is therefore both an engineering dependency and a recurring input cost. Route diversity that consumes more structures or a second owner's corridor may cost more than two fibres placed together.
ICE's infrastructure-access page says shared use covers poles, ducts and conduits, depends on technical feasibility, and requires either agreement on compensation and conditions or a SUTEL order. It also states that the primary and priority use of the infrastructure is ICE's provision of electricity and telecommunications. That priority is sensible for a utility asset, but it matters after a collision or storm: restoring or making safe the electrical structure may precede a telecom operator's final cable work.
The full ICE shared-use offer makes feasibility conditional on pole type, height, loading, planned reconstruction or replacement, existing cables, safety clearances and the need not to degrade ICE's services. It provides for charges by pole position and other infrastructure units. It also allocates tasks around modifications and pole substitutions. Those are not administrative details outside the network. They determine whether an access build can take the intended route, who must act when a pole is replaced, and how quickly a telecom crew can complete its part.
ICE separately explains that network interconnection requires a prior agreement and is based on its existing network structure in its interconnection offer. Physical support and traffic interconnection are different products, but both demonstrate a general point: an enterprise circuit can depend on agreements beyond the retail provider even when the provider owns meaningful infrastructure of its own.
The implication is not that shared poles are inherently unreliable. They are a common and efficient way to expand fibre. ICE's broad footprint can reduce deployment time and avoid duplicative civil works. The implication is that two cables on the same structure are exposed to the same structure, and a cable on someone else's structure inherits that owner's safety and work sequence. Resilience must therefore be described as a chain of separately controlled dependencies, not as a single percentage printed beside a provider's name.
Dedicated bandwidth does not create a separate physical path
Dedicated internet answers an important performance question: whether the contracted access capacity is reserved for one customer rather than shared with nearby subscribers at the access layer. Luminet's explanation of dedicated internet promises specified bandwidth for one company and presents stable speed, lower latency and improved security as benefits. Those attributes may be valuable. They do not answer the resilience question.
A customer can have a fully dedicated 500 Mbps Ethernet circuit carried over a single fibre. It can also have two 500 Mbps circuits that share a pole route and one aggregation chassis. Conversely, a customer can have physically separated access paths that eventually share an upstream port. Bandwidth contention, physical failure and logical congestion are different risks. The word “dedicated” resolves only part of the first.
The same care is needed with point-to-point terminology. A point-to-point service describes the relationship between two endpoints or the way a circuit is provisioned. It does not guarantee that the path is geographically direct, entirely owned by one company or free from shared closures. An MPLS service can keep customer traffic logically separated while travelling through common fibre and routers. A separate virtual routing instance does not survive a severed feeder cable merely because its forwarding state is isolated.
CRISP's product language also spans fibre and wireless. Wireless can be a useful access alternative when it reaches a separate tower or rooftop over a different physical corridor. It can be a weak backup if its base station returns through the same fibre aggregation site, draws from the same electrical feeder or loses line of sight in the same weather event. The medium alone does not establish independence.
For a buyer, a defensible dedicated-circuit specification has to name the layer at which the commitment applies. Committed information rate describes throughput. A latency objective describes packet delay. A restoration target describes time after a fault is accepted. An availability percentage describes allowed downtime over a defined period, subject to exclusions. A route-diversity clause describes shared physical structures and sites. A dual-power clause describes feeds, batteries, generators and maintenance bypass. One phrase cannot do all of that work.
This is where CRISP's sparse public service-level disclosure becomes important. The Luminet company page displays headings for a guaranteed annual percentage and years of experience without supplying values next to them. The point-to-point article says repairs occur in hours but gives no threshold, measurement clock, service window or remedy. The lack of public detail does not mean customers receive no contractual schedule; enterprise terms are often negotiated privately. It means outsiders cannot use the website to compare the strength of the promise with the physical design.
The appropriate conclusion is deliberately narrow. Luminet markets dedicated enterprise fibre and has evidence of a real network and operating history. Its public descriptions are not sufficient to verify “real redundancy.” That phrase becomes meaningful only when attached to a particular pair of circuits and a written inventory of every element they do and do not share.
Two upstreams diversify routing, not necessarily the road out
The internet-facing record gives CRISP more substance than its marketing pages alone. LACNIC's public registration for AS28022 identifies CRISP S.A. as the registrant of an active, directly allocated autonomous system first registered in December 2008. This is a durable routing identity, not merely a reseller label.
At the time of review on 17 July 2026, RIPE NCC's announced-prefix view showed sixteen IPv4 /24 announcements from 190.106.64.0 through 190.106.79.0 and one IPv6 /32, 2800:510::/32. Its routing-status view showed the origin active, with 4,096 announced IPv4 addresses, one IPv6 /32 expressed as 65,536 /48 units, broad visibility among RIPE RIS peers and a first-seen date in February 2009. Address space and routing longevity support the case that CRISP operates a genuine internet network. They still say nothing about available gigabits, fibre routes or power autonomy.
The observed upstream picture is useful. RIPE NCC's autonomous-system neighbour view showed two networks on the upstream side: AS23520, associated with Columbus Networks, and AS52468, UFINET Panama. It also saw downstream or customer-side relationships with AS12222, AS264607 and AS28110. The direction is an inference from observed paths, not a complete contract inventory. A third private interconnection can exist without being visible in the same way, and a visible neighbour can be reached through a shared transport segment.
The ownership change makes one neighbour especially relevant. UFINET Panama sits in the same broader UFINET group as the Livister buyer described by SUTEL. That can produce operational advantages: aligned escalation, regional backbone access and purchasing scale. It also means the two visible upstream names do not necessarily represent two economically independent groups after the acquisition. Corporate affiliation is not physical convergence, but it changes how a customer should think about supplier concentration and escalation.
CRISP also has verifiable exchange presence. The public PeeringDB network record identifies CRISP, also known as Luminet, and reports a selective peering policy, a self-reported traffic band of 20–50 Gbps, twenty IPv4 prefixes and thirty-two IPv6 prefixes, plus 10 Gbps ports at CRIX in Costa Rica and Equinix Miami. Some of those inventory counts differ from the prefixes currently visible through RIPE NCC. That is not surprising: a self-maintained directory can lag routing changes or count holdings differently. For current announcements, the observed routing view is the stronger measure.
The 10 Gbps entries are installed interface speeds, not proof of 10 Gbps of unused capacity. A port can be partly occupied, oversubscribed upstream, limited by transport into the exchange or used primarily for bilateral and route-server traffic. The 20–50 Gbps band is self-reported aggregate traffic, not a measured customer headroom figure. Neither number reveals whether the Costa Rica and Miami appearances are reached on separate cable systems.
CRIX itself is independently visible. Packet Clearing House lists the Costa Rica Neutral Internet Exchange Point as an active Ethernet exchange in San José managed by NIC Costa Rica and established in 2014. The exchange's member directory lists Luminet, AS28022, with a joining date of 28 May 2014. Local peering can shorten paths to participating networks and keep some Costa Rican traffic from making unnecessary international trips. It cannot protect traffic to destinations that still need a failed upstream, nor can it help if the customer's access path cannot reach the exchange.
A LACNIC-commissioned 2022 study of country-level interconnection provides historical context for CRIX and Costa Rica's dependence on local and international relationships. It is useful for understanding the ecosystem, not for measuring CRISP in 2026. The current public evidence therefore supports a carefully worded conclusion: AS28022 has long-lived address space, local exchange participation, a Miami exchange presence and at least two observed upstream paths. It does not reveal whether those paths leave the CRISP network through different buildings, ducts, poles, border routes, terrestrial systems or submarine cables.
Routing security adds one more positive but distinct signal. RIPE NCC's RPKI validation result showed a valid route-origin authorisation for AS28022 covering 190.106.64.0/20 with a maximum length of /24. That helps other networks reject an unauthorised origin for the covered routes. It reduces one class of routing error or hijack. It does not protect a cable, a power feed or a router chassis from failure.
Installed, lit and usable capacity are different quantities
Capacity claims become misleading when three different quantities are merged. Installed capacity is the rating of equipment and interfaces that have been put in place. Lit capacity is the portion activated with optics, transport and configuration. Usable capacity is what can be carried at the relevant hour after existing traffic, resilience reserves, traffic-engineering constraints and failure conditions are taken into account.
The public evidence supplies fragments of the first two quantities. PeeringDB lists two 10 Gbps exchange ports. Luminet advertises access tiers above 300 Mbps and says point-to-point fibre can readily support more than 1 Gbps per port. The visible address space supports thousands of IPv4 endpoints and a substantial IPv6 allocation. None of these facts yields a backbone total. Adding the two exchange ports would be wrong because they serve different locations and traffic sets.
Multiplying address counts by a retail speed would be worse because addresses are not subscriptions and subscriptions do not all transmit at their contracted rate simultaneously.
The self-reported 20–50 Gbps traffic band is the closest public figure to aggregate traffic scale, but it is broad, undated in the response and not independently measured. Even if perfectly current, it would not identify a peak, a percentile, an average or the direction of traffic. It would not show how much of the load crosses CRIX, Miami or an upstream transit link. It would not disclose the largest failure load the remaining paths can accept.
For a protected enterprise service, failure-state capacity is the quantity that matters. A network may perform comfortably with two upstreams active and congest when one disappears. A 10 Gbps interface can have 4 Gbps of normal headroom but only 500 Mbps after traffic is rerouted through a more constrained transport segment. A backup wireless access can pass a health check at low load and fail to carry the customer's committed rate during an outage. An availability design is incomplete unless capacity is tested in the state it is meant to survive.
CRISP's public pages do not disclose backbone interface inventories, optical channel counts, metro-ring utilisation, upstream commits, burst terms, oversubscription ratios, peak traffic, reserved restoration capacity or failure-state tests. SUTEL's competition decision contains market analysis but redacts company-specific shares and commercially sensitive quantities. Those omissions are understandable in public documents. They nonetheless leave a buyer unable to reconcile advertised access rates with the network's usable headroom.
There is also an economic reason to separate installed from usable capacity. A regional provider pays for poles, fibre construction, optics, routers, colocation, upstream transit, exchange ports, field staff and spares before every unit is sold. Keeping enough empty capacity to absorb a failure has an opportunity cost. The larger UFINET group may improve access to wholesale capacity and regional infrastructure, but shared group supply can also concentrate procurement. Neither effect can be quantified from the public record.
The minimum useful disclosure would be a dated, redacted capacity table for the service path. It would show the customer committed rate; access-port speed; normal and 95th-percentile utilisation on each relevant uplink; upstream committed and physical limits; the capacity available after the largest single failure; and the date of the last loaded failover. Figures can be expressed in ranges to protect commercial information. What cannot be replaced by a range is the relationship between demand and surviving capacity.
Until such evidence is supplied, the defensible statement is that CRISP has installed 10 Gbps exchange interfaces and publicly reports traffic in the tens of gigabits per second. It is not defensible to infer total backbone capacity, spare capacity or the maximum protected customer load from those facts.
Power turns passive fibre into an active service
Glass fibre between two points does not need electricity to carry light, but every useful service around it does. The optical transmitter, receiver, access switch, router, wireless radio, monitoring system and customer equipment require power. A route can be physically intact and still disappear because a battery has exhausted, a rectifier has failed or a generator cannot be refuelled.
CRISP's product surface implies several powered layers. Managed MPLS, SD-WAN, Wi-Fi, corporate voice, data-centre services, cybersecurity and cloud access all depend on active equipment and management systems. The AS28022 exchange appearances require routers and optical transport. Customer circuits require powered termination at the customer site and somewhere in the provider network. The public record does not identify those powered locations, their utility feeds, battery runtime, generator coverage, fuel arrangements or maintenance history.
The pole relationship introduces a subtle coupling. ICE is both a telecommunications actor and a major electricity-infrastructure owner. Its shared-use offer makes electricity and ICE telecommunications the priority uses of the pole asset. That does not mean CRISP necessarily buys electric service from ICE at every site, or that a pole event always removes power. It means a physical incident can trigger a safety sequence controlled by the infrastructure owner before telecommunications attachments are restored.
If an active cabinet or nearby site is supplied by the same damaged distribution section, the circuit may face both a cable fault and a power fault.
Power diversity must therefore be tested node by node. Two utility feeds are only independent if they arrive from distinct feeders and do not share a transformer or upstream substation exposure relevant to the customer's scenario. A generator is useful only if it starts under load, has enough fuel and can be reached during the event. Batteries are useful only for their measured runtime at current load and temperature, not their original catalogue rating. A dual-corded router is not protected if both cords enter one failed power strip.
The customer end matters as much as the provider end. A company can buy a highly resilient carrier circuit and connect both carrier devices to the same uninterruptible power supply. It can place dual routers in one room with one cooling unit. It can lose its local-area network while the provider remains healthy. A rigorous review should label the demarcation: which power and environmental risks belong to the provider, which to the customer, and which depend on a landlord or data-centre operator.
For CRISP, the missing public evidence is straightforward. There is no inventory of active field cabinets, aggregation points or data-centre nodes; no stated minimum battery runtime; no generator coverage ratio; no record of black-start tests; no fuel replenishment plan; and no published power-related incident history. That absence should not be converted into a claim that backup power is absent. It should be converted into a request for dated runtime evidence on the exact service path.
A useful proof package would show a one-line power diagram for each critical node, the most recent battery discharge or conductance test, generator automatic-transfer results, fuel duration at measured load, maintenance bypass, alarm escalation and the dependency of any wireless backup site on fixed-network return capacity. The provider should then demonstrate a failover with the primary utility supply removed, not merely a logical shutdown of the primary router.
Power evidence often feels less marketable than a bandwidth figure, but it decides the tail of the outage distribution. Fibre can be repaired in hours and still deliver a long interruption if an active site remains dark. Conversely, strong local power can keep an alternate path working while a pole corridor is rebuilt. The promise of resilience is credible only when route and power independence are evaluated together.
A pole break creates a multi-party recovery chain
Consider a vehicle striking a shared ICE pole carrying a CRISP cable. The first task may be to secure an electrical hazard and make the area safe. ICE may need to assess or replace the pole. CRISP or its current operating successor must identify its cable, determine whether fibres are broken, obtain access, move or re-lash the attachment, splice the cable, test optical levels, restore routing and verify the customer service. If the span is leased, an affiliate or third carrier may enter the chain. If traffic is rerouted, upstream or data-centre staff may also be involved.
Every hand-off creates a clock that a single retail restoration promise can conceal. The service desk clock starts when the customer reports the fault, but the field clock may start only after triage. A telecom crew may be ready while the site remains unsafe. A new pole may be standing while the cable owner awaits a lift or splice team. The optical path may test clean while a router session fails to return. The customer may see service before the network has returned to its protected state.
Costa Rica's current service-quality regulation requires operators to provide service efficiently and continuously around the clock, report network faults that interrupt or degrade service for more than an hour, describe affected services and locations, and give a repair or estimated restoration period. It also provides for customer compensation, while allowing exclusions when the provider proves force majeure, an unforeseeable event or a third-party act. Those rules establish accountability to the end user. They do not automatically compress a two-owner repair sequence.
The difference between restoration and recovery is crucial. Restoration means traffic is flowing again, perhaps over a reduced-capacity or temporary route. Recovery means the primary infrastructure has been repaired, capacity has returned to normal, alarms have cleared, the backup is again available and the customer is no longer one failure away from another outage. Reporting only the first time hides the duration of degraded resilience.
The legal transition to Ideas Gloris adds another potential hand-off. If the Luminet service desk takes the call but the contract, network staff, spares or wholesale supply sit elsewhere in the group, escalation rights must be clear. Group integration could shorten the chain if one operations centre commands all relevant resources. It could lengthen it if responsibilities are split and the customer lacks a single accountable incident commander. Public records do not establish which outcome applies.
Local labour is therefore part of infrastructure capacity. The named service-desk and engineering roles on Luminet's site show that the company has presented a local team, but names and titles do not reveal shift coverage, on-call depth, regional depots, fusion-splicer availability, lift access, spare optics or simultaneous-fault capacity. A provider can meet ordinary repair demand and still be overwhelmed by a storm that breaks many spans.
A serious service review would request incident records in a consistent form: detection time, customer notification, dispatch time, arrival time, infrastructure-owner clearance, cable repair, traffic restoration, full resilience restoration and root cause. It would distinguish customer-equipment, provider-equipment, fibre, pole, power, upstream and third-party incidents. Percentiles matter more than a best case. The median can look good while the longest ten per cent of repairs determine whether a factory loses a shift.
It would also test authority before an incident. Does the retail provider have round-the-clock contact with ICE's fault centre? Can it dispatch a crew while commercial offices are closed? Which party carries replacement pole hardware, fibre, closures and optics? Are permits or police traffic control needed at common routes? Which executive can authorise temporary capacity or a third-party tail? Recovery speed comes from answers prepared in advance, not from a generic commitment to attentive support.
The customers at risk are concentrated at the handoff
CRISP's stated market is corporate rather than mass residential service. That changes the shape of harm. A lost enterprise circuit can isolate a branch, interrupt voice service, stop card authorisations, break remote access, disable cloud applications or separate two data-centre environments. The number of affected subscriptions may be small while the economic impact is large.
The handoff concentrates that risk. CRISP's network can be healthy nationally while one building entrance or access closure removes the customer. Two branches can each have working local access while a shared data-centre interconnection fails. Local peering can remain available while the application the customer needs sits beyond an international path. Availability must therefore be defined from the user's application location, not from the provider's core.
There is no public CRISP customer list, access-circuit count, revenue concentration, trouble-ticket series or province-level performance table. SUTEL's 2025 fixed-internet quality report evaluated four large providers accounting for 83 per cent of the market—Kölbi, Liberty, Telecable and Tigo—using probes across the country. CRISP was not one of the four named providers. That omission is not a negative score and should not be represented as one. It means the regulator's public mass-market comparison does not provide an independent performance baseline for this enterprise-focused network.
Enterprise contracts may contain better evidence than public consumer reports. A customer can examine its own latency, loss, utilisation and incidents. But a single customer's monitoring still has blind spots. If both circuits are idle during a test, it may not expose failure-state congestion. If probes sit behind the same customer firewall, they may confuse local failure with carrier failure. If the provider closes a ticket when any traffic returns, the customer may not capture the time until full protection is restored.
The acquisition also changes who might benefit and who might be exposed. Access to UFINET group transport and data-centre assets could broaden route choices, improve purchasing power and deepen spare inventories. Corporate integration could also move traffic onto common group infrastructure, reducing supplier independence even as the list of products grows. Both are plausible inferences. Neither can be selected without post-acquisition route and incident evidence.
Customer concentration matters to repair priority as well. A regional enterprise provider can deliver personal attention because it serves fewer, higher-value sites than a national mass-market carrier. Yet the same scale can limit simultaneous dispatch when several pole corridors fail. A customer should ask how priority is assigned, whether premium restoration reserves a crew or merely places a ticket earlier in a queue, and whether the provider has mutual-aid arrangements for widespread events.
The right unit of analysis is the business process at the far side of the handoff. A 99.9 per cent annual availability figure permits about 8 hours and 46 minutes of downtime, but even that arithmetic is incomplete without measurement rules and exclusions. Eight hours overnight at an office may be tolerable; eight minutes during a time-sensitive transaction window may not be. A circuit pair should be designed against the customer's maximum tolerable outage, data-loss tolerance, traffic load during failover and manual recovery burden.
CRISP's public profile supports a reasonable case that it has served Costa Rican businesses for many years and built local technical capability. It does not allow an outside reader to quantify how many businesses share each access corridor or active site, how incidents are prioritised, or how post-merger operations have changed. Those are the questions that determine who experiences the hidden common failure.
Proof, not adjectives, should define the redundancy claim
CRISP's public evidence is stronger than that of a provider with only a sales page. It has a long-standing authorisation, an autonomous system, visible address space, local exchange membership, international peering presence, two observed upstream paths, an official pole-sharing relationship and a regulatory record that describes its enterprise services. The same evidence also reveals the limits of what can be known from outside.
The most important unresolved item is a route-separation statement for each protected customer pair. It should identify common poles, ducts, bridges, building entrances, closures, aggregation sites, cross-connect rooms and long-haul exits. It should say where diversity starts. “Diverse after our core” is materially different from “diverse from the customer demarcation.”
The second item is a current responsibility matrix. It should name the contracting company after the CRISP–Ideas Gloris absorption, the owner of each fibre span and active device, the wholesale suppliers, ICE's role, the party that can authorise work, and the escalation contact for each dependency. It should explain whether the Luminet brand, AS28022 operations and field support sit in the same company or are shared across the UFINET group.
The third item is a capacity disclosure that separates installed interfaces from lit transport and surviving headroom. The two 10 Gbps exchange records and the 20–50 Gbps self-reported traffic band are useful starting points. They must be reconciled with current observed prefixes, peak utilisation, upstream commits and the load carried after the largest credible single failure. A failover test should run at the customer's expected peak, not with an almost empty circuit.
The fourth item is power evidence. For every active node on both routes, the provider should show feed separation, measured battery runtime, generator coverage, the last loaded start and any shared cooling or distribution equipment. The customer should do the same on its side of the demarcation. A protected optical path with an unprotected aggregation router remains one service.
The fifth item is recovery evidence. A sample of recent incidents should disclose detection, dispatch, infrastructure-owner access, repair, traffic restoration and full protection restoration. It should include at least one pole or aerial-cable event if those structures serve the customer, one upstream failover and one power loss. Names of other customers can be removed; the timing and dependency sequence cannot.
Finally, the contract should translate the evidence into enforceable language. It should define a shared-risk event, require notice when a route is moved onto common infrastructure, specify normal and degraded capacity, distinguish temporary restoration from full recovery, and provide a remedy that reflects the business impact. It should not treat a third-party structure as outside the service merely because another company owns it; the retail provider remains the party the customer hired to manage the chain.
None of this requires CRISP to publish its whole network. It requires the current operator to prove the attributes it sells to the customers who rely on them. A confidential route pack, a responsibility schedule, redacted utilisation ranges and witnessed failover results can protect security and commercial interests while making the resilience claim auditable.
The central finding is therefore neither that CRISP lacks redundancy nor that its fibre is unreliable. The evidence does not support either assertion. The finding is that logical diversity is visible while physical and operational independence remain unresolved. Beneath the ICE pole line, the value of a second circuit depends on whether it truly leaves the first circuit's fate behind.