99% Is Now the Law: Engineering EV Charge Point Networks for Measured Reliability

Reliability is no longer a marketing claim — it is a reported number. And because regulators measure it from OCPP status messages, your compliance score is capped by your uplink. Here is what that means for the industrial cellular router for EV charging stations you specify.

Written by E-Lins Engineering Team

A charge point that is plugged in, powered up, and physically perfect can still fail its regulatory reliability target. Not because anything inside it broke — but because it stopped telling anyone it was fine. That single sentence is the most expensive misunderstanding in the EV charging industry right now, and it is the reason connectivity stopped being an IT line item and became a compliance one.

This article is written for charge point operators, site developers and the engineers who specify the equipment between the charger and the backend. It covers what the regulations actually require, why the measurement method matters more than the headline number, and how to design a site so that a cellular hiccup does not quietly cost you a compliance target. Requirements differ by jurisdiction and move quickly — verify the current text for your market before you commit to a design — but the underlying engineering logic does not change.

Reliability Became a Number, and Then It Became a Report

For most of the last decade, charge point reliability was a marketing claim. Operators published it if it was good and stayed quiet if it was not. That changed when three jurisdictions independently turned it into an obligation — and, crucially, defined how it is measured. That measurement rule is why charge point reliability reporting has become an engineering specification rather than a marketing statistic.

Jurisdiction	Requirement	How it is measured	Consequence
UK The Public Charge Point Regulations 2023	Network reliability averaged across the calendar year must reach 99%; contactless payment on points above 8 kW; payment roaming to at least one third-party platform	From OCPP status. A point reporting available, charging or reserved counts; one reporting unknown or blocked does not	Published on the operator’s website; annual report to the Secretary of State; penalties of up to £10,000 per charge point
EU AFIR — Regulation (EU) 2023/1804	All publicly accessible points digitally connected (from October 2024); ad-hoc payment without contract; price transparency; data sharing to National Access Points in DATEX II	Connection status and data availability reported to national access points	Directly applicable regulation; member states set the penalty framework
US NEVI program standards	Availability above 97%; minimum four ports at 150 kW each and 600 kW per site	Charger uptime reported through program compliance channels	Funding eligibility and ongoing compliance for federally funded sites

Read the middle column again, because that is where the engineering lives. The UK regulations do not ask “did the charger work?” They ask “did the charger tell the backend that it was working?” Those are different questions, and they have different failure modes.

A charger with a broken contactor reports Faulted — honest, visible, fixable, and arguably useful, because it tells you exactly what to dispatch. A charger whose router has lost its WAN path reports nothing at all. From the backend’s perspective the last known status goes stale and the point drops into an unknown state. It looks the same as a fault on a dashboard, but it is not a fault. It is silence. And silence is the thing you can fix cheapest, because there is nothing wrong with the charger.

The asymmetry worth internalising

Hardware faults are expensive to fix and easy to diagnose. Connectivity faults are cheap to fix and hard to diagnose — because the symptom is absence of data, and absence of data looks like a hundred other things.

There is a second-order effect that catches networks out: the UK requirement is measured across the network, averaged over a calendar year. That sounds more forgiving than a per-point target, and in some ways it is. But it also means one chronically under-connected site quietly consumes the allowance of every other site you own. We come back to that arithmetic in section 02, because the numbers are less comfortable than most operators expect.

Your Uptime Ceiling Is Your Uplink

Here is the arithmetic, stated plainly. Your industrial cellular router for EV charging stations sits in series with the charger, and in a series system the weakest element sets the ceiling — never the average.

3.65

Days of allowed downtime per point per year at 99%

11.0

Days at 97%, the NEVI threshold

87.6 h

Annual downtime budget per point at 99%

8,760

Point-hours of budget on a 100-point network

Now apply it to a real event. Take a motorway services site with ten rapid chargers. On a Friday evening the site’s single-carrier router loses its WAN path — maybe a local mast issue, maybe a SIM that has been quietly throttled, maybe nothing more exotic than a data allowance that ran out. Nobody notices until Monday because the site is unstaffed and the alarm went to a mailbox nobody reads over the weekend.

That is 48 hours × 10 charge points = 480 point-hours. Against a 100-point network’s annual budget of 8,760 point-hours, a single uneventful weekend has just consumed 5.5% of your entire year’s allowance — on chargers that were, physically, completely fine.

This is what we mean by an uptime ceiling. Your reported reliability is the product of two independent availabilities:

Reported availability = charger availability × connectivity availability

You can buy the most reliable charging hardware on the market and still report 94% if your link is at 94%. The hardware investment is capped by the connectivity investment, and no amount of spending on the charger moves that ceiling. Conversely — and this is the good news — connectivity is by far the cheapest variable in that equation to improve.

Three design consequences follow directly, and they are the practical core of this article:

• Eliminate single-carrier dependency. Dual-SIM failover on a charge point network is not a premium feature, it is the difference between a mast outage being a 30-second event and a three-day event. Two SIMs from different carriers, with continuous health-checking and automatic tunnel re-establishment.
• Make silence loud. The router should detect its own loss of backend reachability and alarm on it — not wait for the CPMS to notice a stale status. A device-level “I cannot reach my server” alert, correlated with the site ID, turns a weekend of silence into a 20-minute event.
• Alert on the right thing, to a human, at the right hour. The expensive part of the weekend story is not the outage, it is the 48 hours before anyone acted. Route availability alarms to whoever is on call, with escalation, and test the path.

The Day-1 Problem: the Grid Arrives, the Fibre Never Does

Ask any site developer what delays a charging hub and you will hear about grid connections, wayleaves, and DNO lead times. Almost nobody mentions data — until commissioning week, when the chargers are energised, the payment terminal needs to be certified live, and there is no backhaul.

The reason is structural. A new charging site gets power because power is the point of the site. Data is not. A fixed-line circuit is a separate order, with a separate provider, a separate wayleave, and a separate lead time that frequently exceeds the entire civil programme. On motorway and retail sites, trenching across a live car park can cost more than the chargers.

So the realistic commissioning sequence is: energise on cellular, and treat fixed line as an upgrade that may or may not ever arrive. That has two implications for equipment selection:

• Cellular is the primary path, not a backup. Design the site so that full commercial operation — OCPP, payment authorisation, firmware, remote diagnostics — works over the cellular link with no degradation. If your design assumes fibre and treats cellular as a stopgap, commissioning slips to the fibre date.
• Keep a wired WAN port free. Where a circuit does eventually arrive, or where the site has an existing Ethernet segment from the site operator, you want to be able to plug it in as an additional WAN path with automatic failover — not re-architect the site.

On multi-tenant sites — retail car parks, fleet depots, workplace hubs — there is a related trap. The site owner offers “we have internet, just connect to our network.” Resist it as your primary path. You do not control their change windows, their firewall policy, or their incident response, and your compliance number is now downstream of someone else’s IT department. Take it as a secondary path; keep your own cellular as primary.

OCPP Is Not Just a Protocol — It Is Your Compliance Meter

OCPP (Open Charge Point Protocol) is the language between the charge point and the charge point management system. It is not mandated by AFIR — the regulation requires digital connectivity and smart charging capability without naming a protocol — but in practice OCPP is universal, and the OCPP 1.6J vs 2.0.1 decision for charge point management is a design choice rather than a default: 1.6J over WebSocket is the most widely deployed variant today, with 2.0.1 gaining ground.

Most people treat OCPP as plumbing. It is not. It is the instrument your regulator reads.

What the backend actually sees

A charge point sends a BootNotification when it powers up, periodic Heartbeat messages at a configured interval, and a StatusNotification whenever its state changes — Available, Preparing, Charging, SuspendedEV, Finishing, Reserved, Unavailable, Faulted. The CPMS maintains a picture of the estate from those messages.

When the network path drops, no StatusNotification arrives. After enough missed heartbeats the CPMS marks the point unknown — and under the UK measurement rules, unknown is not available. That is the entire mechanism by which a connectivity fault becomes a reliability miss.

Three configuration mistakes that cost points

• Heartbeat interval too long. Set it to an hour and a short outage can be missed entirely — which sounds good until you realise it also means your status data is up to an hour stale, and a genuine fault is invisible for just as long. Somewhere in the 60–300 second range is the usual balance between responsiveness and data volume. Whatever you choose, make sure it is shorter than your CPMS’s own staleness threshold, or you have two clocks disagreeing about what “offline” means.
• A dead WebSocket that nobody notices. This is the nastiest failure mode, because it is asymmetric. A NAT or firewall timer silently kills the idle connection; the charger’s TCP stack thinks the socket is open and keeps writing into a void; the CPMS has already given up. Both sides believe the other is at fault. The fix is at the router: TCP keepalive, plus a router-level reachability check that re-establishes the tunnel on failure rather than waiting for the application to notice.
• Maintenance windows that look like faults. If you take a point offline for planned work by powering it down, you generate unknown states. If you can command it to Unavailable first, through the CPMS, the backend has an intentional status rather than a silence. Check how your CPMS maps each status before you assume this helps — implementations differ — but deliberate status beats ambiguous silence in almost every case.

Security profiles, briefly

OCPP 2.0.1 formalises security profiles. The weak ones rely on basic authentication over TLS; the strong one — profile 3 — adds client-side certificates, so the charge point proves its identity to the backend cryptographically rather than with a shared username and password. If you are building now, specify profile 3 or its equivalent. It costs you a PKI and saves you the shared-credential problem described in section 08.

One detail that belongs in the design document: OCPP over an untrusted network should always terminate in an encrypted tunnel. Whether you run IPsec or WireGuard from the site router to your concentrator, the OCPP session should never traverse the public internet in the clear — and with profile 3, the certificate check gives you device identity even if someone does get onto the transport.05

One Router, Three Trust Zones

A modern rapid charging site carries three categories of traffic that have no business seeing each other, and usually one physical router.

• Payment. The contactless terminal and its acquiring traffic. Regulated, cardholder-data-adjacent, and the reason your compliance team will ask pointed questions.
• OCPP and operations. Charge point to CPMS: status, transactions, smart charging setpoints, firmware, remote diagnostics.
• Site OT. Energy meters, the dynamic load management controller, battery storage, CCTV, and whatever building systems the landlord has wired in.

The motivation for separating them is not tidiness — it is PCI DSS scope on an EV charger payment terminal. PCI DSS Requirement 1 is about network segmentation, and the reason it exists is practical: if the payment terminal shares a flat network with everything else on site, the audit scope tends to expand to include all of it. Segmentation is the control that keeps scope — and therefore audit cost, and therefore the size of the problem when something goes wrong — contained.

The implementation is the same tool used in every other industrial network: VLAN segmentation for EV charging sites — 802.1Q on the site router, with firewall rules between VLANs. Concretely:

• Payment VLAN. Egress to the acquirer only. No route to the OCPP VLAN or the site OT VLAN. Client isolation on.
• OCPP VLAN. Encrypted tunnel to the CPMS concentrator. Inbound-initiated connections denied; the site initiates outbound.
• Site OT VLAN. Meters, DLM controller, storage, CCTV. Local traffic stays local — the DLM loop should never make a round trip to a cloud server to decide a setpoint.
• Management VLAN. Router, NMS, syslog, out-of-band. Reachable only from your operations network.

There is an operational dividend too. When three contractors own the charger, the payment terminal and the CCTV, a partitioned architecture lets each have their own credentials, their own VLAN and their own log stream, with none able to touch another’s configuration. On sites where things go wrong, that separation is usually the difference between a two-day diagnosis and a two-month dispute.

The PKI You Did Not Budget For

This section is the one that will matter most in eighteen months and is least present in today’s tender documents.

Commission Delegated Regulation (EU) 2025/656 adds technical specifications under AFIR. From 1 January 2027, newly installed or renovated publicly accessible AC and DC charge points must comply at least with EN ISO 15118-20:2022. In doing so it turns ISO 15118-20 Plug & Charge certificate management into a site requirement rather than a vehicle feature. ISO 15118-20 is the vehicle-to-grid communication standard that carries Plug & Charge, advanced smart charging, and bidirectional power transfer.

ISO 15118-2 already defined Plug & Charge: the driver plugs in, the car and charger exchange certificates, the contract is identified, the session starts and bills itself. No RFID card, no app, no registration. It is excellent for drivers and it converts your charge point into a device with a certificate lifecycle.

What that means concretely at the site:

• The charger holds a certificate chain. The SECC — supply equipment communication controller — needs its own valid certificate, signed under a V2G root, and it needs to be renewed before it expires.
• Contract certificates must be validated. The vehicle presents a contract certificate; the charger (or the backend on its behalf) checks it, typically including an OCSP or equivalent revocation check against an external responder.
• Trust lists must be maintained. Which mobility operators and roots do you trust? That list changes, and it has to reach the site.
• Time must be right. Certificate validation is time-dependent. A charge point whose clock has drifted will reject valid certificates. Reliable NTP is not a nicety here; it is a functional dependency.

Notice what those four have in common: every one of them depends on the charger having a working, correctly-timed, outbound internet connection. Plug & Charge adds external dependencies that an RFID-based charger simply did not have. If the OCSP responder is unreachable, or the clock has drifted, or the trust list is stale, sessions fail — and the failure looks, to the driver, exactly like a broken charger.

Design question worth asking your supplier today

What does your charge point do when it cannot reach its certificate validation service? Does it fail open and allow the session, fail closed and refuse it, or fall back to RFID or app authorisation? Every one of those is a defensible answer. Not knowing which one you have is not.

From the connectivity side, the requirements are unglamorous and specific: stable outbound TLS, accurate time via NTP with a fallback source, enough headroom for certificate and revocation traffic, and GNSS or network time for sites where NTP may be unreachable. Our M400 is often used here precisely because it combines serial and GNSS in an 83 × 53 × 25 mm enclosure — small enough to sit alongside the charger controller and give it a position and a time source.

Dynamic Load Management: Connectivity as a Grid Asset

Ask a site developer what limits a rapid charging hub and the answer is almost never the chargers. It is the grid connection. A site with ten 150 kW chargers and a 400 kW supply is a 400 kW site that happens to have ten chargers, and the interesting engineering is in how you share that 400 kW.

That is what dynamic load management for EV charging does: measure the site’s real consumption continuously, and adjust each charger’s available current in real time so the total stays inside the connection limit. Do it well and a 400 kW connection can serve a site that would otherwise need 1 MW — which is frequently the difference between a viable project and one that dies at the DNO quotation stage.

It also makes connectivity load-bearing in a way it was not before, for two reasons.

First, the control loop has to be local. If the setpoint decision round-trips to a cloud service, you have put a variable-latency network inside a control loop that is protecting a physical fuse. Cellular latency is normally fine — tens of milliseconds — but “normally” is doing a lot of work in that sentence. The right architecture is a local controller that closes the loop on site, reading the meter over RS485 and writing setpoints to the chargers over the local Ethernet, with the cloud doing supervision, scheduling and reporting rather than real-time control. Keep the loop local; let the network be the supervisor.

Second, decide what happens when the link fails. This is the question nobody writes down and everybody should:

Fail open or fail closed?

If your load management loses its controller, its meter, or its uplink, does each charger revert to full rated power — fail open — or drop to a conservative static limit — fail closed? Fail open risks tripping the site supply and taking every charger down. Fail closed costs throughput during the outage and keeps the site alive. For almost every site we have worked on, fail closed to a safe static limit is the correct answer, and it should be an explicit line in the design rather than a firmware default nobody has read.

The practical BOM that comes out of this: a site controller, an MID-certified or equivalent revenue-grade energy meter read over RS485 Modbus for EV charging revenue metering, and a router that can both read that meter locally and carry site data to the backend. That is exactly the shape of the H820 — five Gigabit Ethernet ports for the chargers and controller, a DB9 RS232/RS485 port for the meter, and PoE as either PD or PSE if you need to power something in the cabinet.

A Public-Facing Computer With a Card Reader Attached

A rapid charger is a networked Linux computer, in a public place, at ground level, unlocked or behind a service panel that a determined person opens in under a minute, connected to a payment terminal and to a grid-connected load. The threat model is not subtle.

The structure we recommend is the same IEC 62443 zones and conduits model used elsewhere in industrial control, applied to a charging site:

• Operations zone. CPMS, NMS, PKI, certificate authorities. Hardened, jump-hosted, MFA.
• DMZ zone. Where site traffic terminates: VPN concentrators, OCPP brokers, update servers, syslog collectors.
• Site zone. The chargers. Physically exposed, individually untrusted, each authenticating with its own certificate.

Then the specific controls that matter here:

• No shared credentials across the estate. One PSK across 500 charge points means one opened cabinet compromises all 500, and revocation means a fleet-wide rekey that in practice never happens. Per-device certificates, issued at staging and rotated, with a revocation path.
• Signed firmware only. The router and the charger should refuse unsigned images, and firmware should ship in rings — canary, then a small percentage, then the estate — with automatic rollback on failure.
• Deny inbound by default. The site initiates outbound tunnels. Nothing on the public internet initiates a connection into a charger.
• Physical tamper as an alarm. Cabinet door switch into a digital input. Costs nothing — the input is already there — and converts an invisible event into a logged, alertable one. Our pole and site routers expose four DI/DO channels; wire one to the door and one to the supply contactor state.
• Central syslog with synchronised time. When something does happen, the difference between a two-hour and a two-week investigation is whether you have site, tunnel and platform logs with consistent timestamps.

Commissioning and Operating at Fleet Scale

Commissioning ten sites is a networking project. Commissioning five hundred is a logistics project that happens to involve networking. The constraint is never bandwidth — it is that configuration must not depend on which engineer did which site.

The pattern that works, and that we see mature operators converge on:

• Identity at the depot, not on the site. Each site router gets its certificate, its SIM or eSIM profile, and its configuration template bound to a site ID before it leaves the workshop. The installer’s job becomes physical: mount, terminate, scan, close. With eSIM and eUICC, carrier selection can even happen after installation.
• Zero-touch bootstrap. On first power, the router brings up the link, authenticates with its certificate, reaches a bootstrap endpoint, and pulls its configuration — VLANs, tunnel endpoints, VLAN-to-port mapping, NTP sources, alerting thresholds. TR-069 plus an NMS that auto-registers the device against the expected site ID, and flags any mismatch for human review.
• Ring-based remote firmware updates for EV chargers. Canary sites first, hold, then 5%, then 25%, then the estate. Automatic abort and dual-bank rollback, so a failed update reverts itself rather than bricking a site.
• Remote diagnostics good enough to avoid the visit. This is the point of the whole exercise: being able to tell, from the NOC, whether a failed session was the charger, the meter, the carrier, or the backend. Every ambiguous alarm you resolve remotely is a visit you did not pay for.

The platform matters as much as the device. A router that cannot be managed in bulk will be managed one at a time, which means it will not be managed at all. That is the whole argument for centralised NMS across an EV charging router fleet. Look for configuration templates, group policy, bulk firmware, per-device certificate lifecycle, per-site data usage reporting, and an API — because your CPMS vendor will want to integrate with it.

The Motorway Truck Roll

We have written before about the economics of the truck roll, but truck roll cost on an EV charging network has a multiplier that other industries do not: the sites are often on motorway corridors or in live retail car parks, and attending them can require traffic management.

A routine visit to an urban charge point might be a couple of hundred euros. The same visit at a motorway services area, with lane closure or a barrier-separated work area, out-of-hours access restrictions, and a technician with the right accreditation, is comfortably several times that. And the visits tend to be urgent, because a down rapid charger on a corridor is a customer-facing incident.

Set that against the device decisions:

Design choice	Unit delta	What it buys	Verdict
Industrial router with a 10-year design life over a commercial-grade unit	+ tens of euros per site	Avoids a replacement cycle across the estate	Decisive
Dual-SIM multi-carrier failover	Small	Every carrier-related outage becomes a non-event, and every “is it the network?” visit disappears	Decisive
Device-level reachability alerting	Zero — configuration	Turns a weekend of silence into a 20-minute event	Free win
Door switch wired to a DI	Zero — the input exists	Converts silent tamper into a dispatchable alarm	Free win
Dual-bank firmware with watchdog rollback	Included	A failed update reverts itself instead of bricking a site	Decisive

Field note

“Our reliability number was 96 and change and we could not find the reason. The chargers were fine. It turned out to be four sites on one carrier with a marginal mast, dropping for a few minutes at a time, several times a day. Nobody saw it because nothing ever stayed down long enough to trigger an alarm — it just never stayed up long enough to look healthy either.”

Network operations lead, European CPO — shared with permission, details withheld

That quote is worth sitting with. The failure mode was not an outage. It was flapping — repeated short drops that individually look trivial and collectively destroy an availability percentage. It is invisible to alarms tuned for “down” and completely visible to a system that tracks availability as a percentage over time. If your monitoring is binary, you will not find this. Track the percentage.

Heat, Ingress, and Where to Put the Box

A charging site cabinet is a hostile place to put electronics. Rapid chargers reject kilowatts of heat as a matter of normal operation; the cabinet also holds contactors, power supplies and possibly a battery; and it sits outdoors in full sun. Internal temperatures above 60 °C are ordinary in Southern Europe and the Gulf in summer.

The specification to write into the tender is −35 °C to +75 °C ambient operating range, and note that the cold end matters as much as the hot one: Nordic sites need equipment that cold-starts in January without a heater. Commercial-grade hardware rated 0–40 °C does not belong in this tender.

Then there is the placement question, which has three valid answers depending on the site:

• Inside the charger cabinet or a site cabinet. Use a fanless metal-enclosure router with convection cooling, mounted high rather than low in the enclosure, away from the power electronics, with cable glands at the bottom and drip loops on every cable.
• In a roadside or exposed enclosure with no shelter. Use a genuinely sealed device — this is precisely what IP68 outdoor routers for exposed charge points are built for. The H820QO is IP68 with built-in high-gain cellular and Wi-Fi antennas, reserved connectors for external N-type antennas, and pole mount as a designed installation option rather than an improvised bracket.
• Inside the charger itself. For OEM integration, size is everything. The H685frc is a 5G RedCap router at 100 × 60 × 21 mm and 220 g, with two Gigabit Ethernet ports, four DI/DO channels, optional serial and GNSS, and PoE-PD — small enough to disappear into a charger head or a pole-mounted unit.

Product Picks for Charging Sites

There is no single right device, because a motorway hub, a retail car park and a fleet depot have different constraints. What matters is matching the router to where it physically goes and to what else shares the cabinet.

Model	Radio	Why it fits a charging site	Envelope
H820 Site gateway — the default	4G LTE	5× Gigabit Ethernet for chargers and the local controller, DB9 RS232/RS485 for the revenue meter, PoE as PD or PSE, VLAN 802.1Q and VRRP. The natural choice wherever a meter and a load-management controller share the cabinet.	−35~75 °C 168 × 104 × 25 mm
H900pf Where the cabinet must supply power	5G SA+NSA, dual SIM	PoE PSE for charging site CCTV and terminals on two LAN ports at 802.3af/at/bt (48 V, 1 A / 1.5 A / 2 A) — powers CCTV, sensors or a terminal without a separate PoE switch. Also carries RS232/RS485.	−35~75 °C 168 × 104 × 25 mm
H900f High-throughput hub	5G, dual SIM	For large hubs where several chargers, payment terminals and CCTV share one uplink and you want headroom. Serial on a terminal socket; DI/DO × 4.	−35~75 °C 168 × 104 × 25 mm
H685frc In-charger / OEM integration	5G RedCap + LTE fallback	100 × 60 × 21 mm, 220 g. 2× GbE, DI/DO × 4, optional serial and GNSS, PoE-PD, dual power inputs. Fits inside a charger head or a pole unit; RedCap keeps module cost and power down across large fleets.	−35~75 °C 5–40 V DC (5–60 V opt)
H820QO Exposed / roadside	4G LTE	IP68, built-in high-gain antennas with external N-type option, pole mount as a designed option, PoE PD. For sites with no cabinet and no shelter.	−35~75 °C 185 × 172 × 82 mm
M400 Timing, positioning, serial pipe	4G LTE	83 × 53 × 25 mm with RS232/RS485/USB and GPS/BeiDou/GNSS. Gives a charger controller a time and position source for ISO 15118 certificate validation, or acts as a compact serial-to-cellular pipe for a meter.	−35~75 °C DB9 / DB15 / terminal block

One honest exclusion: the H820QO has no serial port. If an exposed site needs both IP68 and an RS485 meter, the answer is an H820 or H900f inside a sealed cabinet, not an H820QO on a pole. Plan that before you order.

Frequently Asked Questions

Q1：Does AFIR require OCPP specifically?

No. AFIR requires publicly accessible points to be digitally connected and to support smart recharging, but it does not mandate a specific charge-point-to-backend protocol. OCPP is the de facto market standard and is what most national implementations assume, so in practice you will use it — but the obligation is about connectivity and capability, not about a named protocol.

Q2：How is the UK 99% reliability figure actually calculated?

It is an average across the operator’s network over a calendar year, determined from the charge point’s reported status. States such as available, charging and reserved count positively; unknown or blocked states are what erode the figure. Because the measurement is derived from reported status rather than from a physical test, a point that is working but silent counts against you. Operators must publish the figure and report annually.

Q3：Can we use the site owner’s existing internet instead of cellular?

You can use it as a secondary path, and you should where it is free. Do not make it your primary. You do not control their change windows, firewall policy or incident response, and your reported reliability becomes downstream of somebody else’s IT department. Keep your own cellular link as primary with automatic failover to the wired path.

Q4：What heartbeat interval should we configure?

Something in the 60–300 second range is typical, but the important rule is that it must be comfortably shorter than the staleness threshold your CPMS uses to declare a point unknown. If those two timers are close, you get flapping: the backend declares the point dead, the point comes back, and your availability trace fills with short dropouts that are artefacts rather than outages.

Q5：Why do short repeated drops hurt more than one long outage?

Because percentage availability does not care whether downtime is contiguous. Four hundred three-minute drops cost exactly as much as one twenty-hour outage. Worse, short drops are usually invisible to alarms tuned for “device down”, so they can persist for months. Track availability as a percentage per site over time, not as a binary up/down state.

Q6：Do we need per-device certificates if we already use a VPN?

Yes, for different jobs. A VPN protects the transport; a device certificate establishes identity. A shared pre-shared key across an estate means one opened cabinet compromises every site, and revocation requires a fleet-wide rekey. Certificates give you per-device identity and a working revocation path. OCPP security profile 3 is the formal version of this idea within the protocol itself.

Q7：What does ISO 15118-20 change for us in practice?

From 1 January 2027, newly installed or renovated publicly accessible charge points in the EU must comply at least with EN ISO 15118-20:2022. Beyond the compliance date, the practical change is that your charge points become PKI clients: they need valid certificates, a way to validate driver contract certificates, maintained trust lists, and accurate time. All four depend on outbound connectivity.

The charger was never the unreliable part.

Nobody sets out to build an unreliable charging network. What happens instead is that reliability gets treated as a hardware property — buy better chargers, get a better number — while the thing actually being measured is whether the charger was able to say so. Once regulators started measuring reliability from reported status, connectivity stopped being the plumbing behind the product and became part of the product. A silent charger and a broken charger score the same.

The uncomfortable version of that: your reported availability is a product, not an average, and the weaker of charger availability and connectivity availability sets the ceiling. The comfortable version: connectivity is the cheapest factor in that product to improve. Dual SIM across carriers, device-level reachability alerting, a heartbeat interval shorter than your staleness threshold, per-device certificates instead of a shared key, a load-management loop that fails closed, and a router with the thermal range to last a decade beside a 150 kW power supply. None of it is expensive. All of it shows up in the number you have to publish.

Building or upgrading a charging site?

Send us your site archetypes, grid capacity and target availability. We will come back with a per-site connectivity BOM, a failover design and a commissioning plan you can put into a tender.