Cell Sites Out of Service: The Percentage Moves When the Reporting Area Does

Within a day of a landfall or a large fire, a number begins to circulate: this county has a given percentage of its cell sites out of service. It is not a reporter's estimate but a federal report published while the event runs, filed by the carriers themselves. What that percentage counts, and what it leaves out, was decided by rules written years before any specific event.

Those rules sit in Title 47, Part 4 of the Code of Federal Regulations, and Part 4 does not create one outage reporting system. It creates two, differing in what triggers a filing, what unit it is measured in, how fast it is due, and whether anything is published at all. The quoted percentage comes from only one of them. The other is presumed confidential, and during a declared disaster it is frequently switched off in exactly the area everyone is asking about.

Two Systems That Count Different Things

The first is the Network Outage Reporting System, governed by 47 CFR § 4.9. It is event-driven: a provider files when an outage crosses a threshold, and the thresholds are alternatives rather than a single test. Under FCC, 47 CFR § 4.9 Outage reporting requirements—threshold criteria, "All wireless service providers shall submit electronically a Notification to the Commission within 120 minutes of discovering that they have experienced on any facilities that they own, operate, lease, or otherwise utilize, an outage of at least 30 minutes duration:". Four numbered gates follow: a Mobile Switching Center failure, a user-minute threshold scoped further down, 667 OC3 minutes, and a 911 or 988 special facility, with clause (iv) reserved. Any one opens the door, and an outage clearing none of them generates no filing.

The second, the Disaster Information Reporting System, is geography-driven rather than event-driven. Under FCC, 47 CFR § 4.18 Mandatory Disaster Information Reporting System (DIRS) reporting, facilities-based providers "shall be required to report their infrastructure status information each day in the Disaster Information Reporting System (DIRS) when the Commission activates DIRS in geographic areas in which they provide service, even when their reportable infrastructure has not changed compared to the prior day." The obligation runs from "the start of DIRS activation until DIRS has been deactivated."

The published county percentage is a DIRS product: the Commission drew a boundary and required every facilities-based provider inside it to state a status once a day. Nothing in that process measures how many people could place a call.

The Chain

The sequence from a dark antenna to a printed percentage runs in a fixed order, and each step discards something.

First, a site stops serving. Section 4.5 defines an outage in terms of significant degradation in an end user's ability to establish and maintain a channel of communications, so the trigger is a service condition, not broken equipment.

Second, the Commission decides which counties are inside the activation. The daily reports introduce the county list with the sentence: "The following counties in Florida are in the current geographic area that is part of the DIRS activation area (the “disaster area”) for today's report." That decision fixes the denominator for every percentage published during the event.

Third, each facilities-based provider counts its own sites inside that boundary and files a daily status. Sites served is the provider's own count, not audited in the report.

Fourth, the Commission sums the county rows and publishes them; the individual filings do not come with it. FCC, 47 CFR § 4.2 Availability of reports states that "Reports filed under this part will be presumed to be confidential under § 0.457(d)(1) of this chapter." The rest of § 4.2 sets terms for agencies granted direct access under a Certification Form, including the condition in paragraph (d) that information such an agency releases "must be aggregated from at least four service providers and must be sufficiently anonymized so that it is not possible to identify any service providers by name or in substance." Those terms are addressed to those agencies. On what basis the Commission publishes its own daily report, and whether that floor reaches it, is not stated in § 4.2 or in any document consulted here.

Fifth, the aggregate is divided and rounded to one decimal. That is the number that travels.

The Switch That Turns One System Off

Section 4.18(b) is the hinge between the two paths, and the clause most easily quoted short. It says facilities-based providers who file in DIRS "are not required to make submissions in the Network Outage Reporting System (NORS) under this chapter pertaining to any outage that occurs in an area in which the Commission has activated DIRS, as long as the first daily DIRS report for the activation is due before the NORS submission under section 4.9 of this chapter would be due for the outage, and the outage is timely reported in DIRS."

Everything after "as long as" is the operative part: the relief is conditional on timing. A NORS notification for a wireless outage is due within 120 minutes, and a DIRS activation that begins after that clock has started does not retroactively excuse the filing. Interconnected VoIP providers, also covered by § 4.18(b), work to 240 minutes or 24 hours under § 4.9(g), so the gap is not the same width for every class. The result is a seam: outages beginning before an activation and after it are documented in different systems on different clocks. A reader of the daily percentage sees only the DIRS side of that seam, because § 4.2 presumes the NORS filings confidential.

What the User-Minute Gate Is Written to Catch

The 900,000 figure is the one number in § 4.9(e)(1) that looks like it might measure a modern wireless outage, and the scope limiting it sits mid-sentence. Clause (ii) reaches an outage "That potentially affects at least 900,000 user minutes of either telephony and associated data (2nd generation or lower) service or paging service;". The word "either" joins two named service classes: second-generation-or-lower telephony with its associated data, and paging. By its own terms the clause does not reach LTE or 5G service.

That leaves three live gates for a wireless outage outside those service classes: failure of a Mobile Switching Center, 667 OC3 minutes, and a 911 or 988 special facility. None is a count of dark macro cell sites, and nothing in § 4.9(e) states a threshold expressed as a number or share of macro cell sites out of service.

The unit itself is not what the phrase suggests. FCC, 47 CFR § 4.7 Definitions of metrics used to determine thresholds splits the definition in two. User minutes are "(1) Assigned telephone number minutes (as defined in paragraph (c) of this section), for telephony, including non-mobile interconnected VoIP telephony, and for those paging networks in which each individual user is assigned a telephone number;" and "(2) The mathematical result of multiplying the duration of an outage, expressed in minutes, by the number of end users potentially affected by the outage, for all other forms of communications." For the telephony branch the count is telephone numbers, not people: paragraph (c) builds those minutes from assigned numbers plus administrative numbers, or alternatively from working telephone numbers.

Within whichever branch applies the threshold is a product, so it behaves as a curve. Dividing 900,000 by the duration gives the count required.

Outage duration	Telephone numbers or end users needed to reach 900,000 user minutes
30 minutes	30,000
60 minutes	15,000
120 minutes	7,500
240 minutes	3,750
480 minutes	1,875
1,440 minutes (24 hours)	625

Thirty minutes is the left edge because § 4.9(e)(1) sets a minimum duration of its own. A short outage over a large city and a long one over a small town can land on the same side of the gate, but only if the affected service is one the clause names.

One place in the rule does convert cell sites into people, and it is narrower than usually taken. Paragraph (e)(2) opens "In determining the number of users potentially affected by a failure of a switch," and then directs a wireless provider to "multiply the number of macro cell sites disabled in the outage by the average number of users served per site, which is calculated as the total number of users for the provider divided by the total number of the provider's macro cell sites." The rule states the formula in full; what it does not do is make either input visible. The affected-user figure is computed from a provider-level ratio rather than counted, which is the same distinction that separates a measured quantity and a number read off a curve, where only one of the two is observed at the instrument.

What One County Row Actually Contains

The daily status reports carry one row per county. In the report issued at 9:00 a.m. EDT on 14 October 2024, each row carries Cell Sites Served, Cell Sites Out and Percent Out, then three cause columns for damage, transport and power, then Cell Sites Up but On Backup Power. There is no column labelled other or unknown.

The total row reads 15,878 sites served and 341 out, which the report prints as "2.2% (down from 4.4% yesterday) of the cell sites in the affected area are out of service." The 52 county rows add up to exactly those two totals. The three cause columns on the total row read 8, 52 and 153, summing to 213. Against 341 sites out, that leaves 128 sites, or 37.536657 percent of the outages, carrying no cause in any published column.

The last column is larger than the outage column: 691 sites recorded as up but on backup power, against 341 out. A site on a generator or a battery string counts on the working side of the ledger, with a runtime the table does not report. The two together are 6.499559 percent of sites in the activation area either dark or on stored power, against the 2.2 percent published.

Why the Highest Percentage Is Usually the Smallest County

Percent out of service is a ratio, and across the 52 counties the denominator runs from 20 sites to 1,858. On 14 October 2024 the highest county figure was 10.0 percent, from a county with 20 sites served and 2 out. The county with the largest outage, 113 sites down, printed 6.8 percent because it had 1,667 sites served.

The arithmetic consequence is direct. In a county with 20 sites, a single site is worth 5.0 percentage points; in a county with 1,667 sites, 0.059988 percentage points. One site moves the published figure 83.35 times as far in the first county as in the second, so rank by percentage and rank by number of outages are different orderings of the same data.

The Denominator Is Redrawn Every Activation

Because the denominator is the set of counties the Commission names, it changes from event to event, and the published percentage moves with it even when the dark-site count does not. Three published Florida reports show the range: a 30-county activation with 8,097 sites served and 153 out, printed as 1.9 percent; a 33-county activation with 11,450 served and 103 out, printed as 0.9 percent; a 52-county activation with 15,878 served and 341 out, printed as 2.2 percent.

Held against each other, the arithmetic is unforgiving. The 153 sites out in the first report would print as 0.963597 percent against the 15,878-site denominator of the third; the 341 in the third would print as 4.211436 percent against the 8,097-site denominator of the first. Only the boundary changed. This is structurally the same effect as the one where a published outage average changes when one category of day is taken out of it, except that here the exclusion is spatial rather than temporal.

Rounding adds a smaller version of the same effect. Percent out is printed to one decimal, and the reports state no rounding convention. Two counties had 4 of 104 and 15 of 390 sites out, both 3.846154 percent to six decimal places, and both printed as 3.9 percent, while a county at 2 of 1,858, or 0.107643 percent, printed as 0.1 percent. The 104-site and 390-site rows are the only two in the table where the printed value sits a tenth above the ordinary rounding of its own printed counts, and the direction of the last digit is not documented. That is the same class of question as one where the same reading can land in a wide band or a narrow one depending on where the breakpoints sit.

Where This Doesn't Apply

The percentage is not meant to describe coverage, and the Commission says so on the face of the reports: "The number of cell site outages in a specific area does not necessarily correspond to the availability of wireless service to consumers in that area." The 14 October 2024 report adds the reason: "Wireless networks are often designed with numerous, overlapping cell sites that provide maximum capacity and continuity of service even when an individual site is inoperable." Where sectors overlap, a county at 6.8 percent can hold service across most of its area; where they do not, a county at 3 percent can have a corridor with nothing. The published figure cannot distinguish the two, and is not designed to.

Nor does the published table appear to suppress small counties: rows for counties with 20, 23 and 25 sites served are printed alongside rows in the four figures. Whatever aggregation rule the Commission applies to its own daily report, it is not visibly withholding the smallest denominators, and what that rule is remains unstated in the documents consulted here.

Commission documents are not consistent about the system either. FCC, Disaster Information Reporting System (DIRS) describes DIRS in these terms: "A voluntary, web-based system through which the Commission collects operational status and restoration information from communications providers during major disasters." The same page states that "The infrastructure status information reported in DIRS is presumed confidential." Section 4.18 is titled as mandatory reporting and requires daily filing during an activation. The rule text governs, but a reader comparing the two documents will find them at odds.

None of this makes the published percentage wrong. It makes it a specific quantity: the share of provider-reported cell sites, inside a boundary drawn for this activation, reported as not serving at one moment each day, with sites on generator counted as working and more than a third of the outages carrying no stated cause. Read as that, it is informative. Read as the fraction of people who cannot make a call, it is a different number, and FCC, Communications Status Report for Areas Impacted by Hurricane Ian and FCC, Communications Status Report for Areas Impacted by Hurricane Milton both say on their face that it is not that number.