Summary

  • Reportability under Part 4 is a union, not a threshold: at least 900,000 user-minutes of telephony, at least 667 OC3-minutes, blocked or lost call counts on tandem and SS7 networks, and a list of named facilities that enters the record with no arithmetic attached.
  • Two of the rule's affected-user counts are constructions of the rule itself: simultaneous call capacity multiplied by a concentration ratio of 8 for mobile interconnected VoIP, and disabled macro cell sites multiplied by average users per site for wireless.
  • Above the arithmetic sits 47 CFR 4.2. Filings are presumed confidential, and the only permitted public aggregate must come from at least four providers and must not identify any provider by name or in substance, a permission the Commission declined to use itself while it publishes daily aggregated status reports during disasters.

On the morning of 22 November 2026 I read the outage definition in 47 CFR 4.5 twice, because the first pass gives one rule and the second gives four. Paragraph (a) is almost humane: an outage is a significant degradation in an end user's ability to establish and maintain a channel of communications, resulting from failure or degradation in a communications provider's network. Everything after that paragraph is bookkeeping.

Paragraphs (b) through (f) do not measure the degradation. They sort it. A special office or facility is an entity enrolled in the Telecommunications Service Priority Program at priority Level 1 or 2, which the rule illustrates with major military installations, key government facilities, nuclear power plants and current primary airports in the FAA's integrated airport plan. A 911 special facility gets four separate triggers, and a 988 special facility gets one of its own. None of these needs a headcount before it enters the record.

One rule, four counted objects

Read the thresholds together and the rule stops looking like a single formula.

  • 900,000 user-minutes of telephony. Cable providers must notify within 120 minutes of discovering an outage of at least 30 minutes that potentially affects at least 900,000 user minutes of telephony service (47 CFR 4.9).
  • 667 OC3-minutes, as an alternative to the same paragraph, counted in circuit-minutes instead of people: the unit is defined in 47 CFR 4.7.
  • Call counts. An IXC or LEC tandem outage is reportable at 30 minutes with at least 90,000 calls blocked or 667 OC3-minutes lost, and SS7 providers report at 90,000 blocked calls on real-time data or 30,000 lost calls on historic carried loads (47 CFR 4.9).
  • Named addresses. Any special office or facility, any 911 special facility, any 988 special facility, with no arithmetic attached at all (47 CFR 4.5).

These are alternatives. They are never summed, and they cannot be: a circuit-minute is not a user-minute, and a blocked call is neither. The rule prints four different questions and one duty to answer any of them.

The floors and clocks carry the same precision. Thirty minutes everywhere. A 120-minute notification clock for most providers, 240 minutes or 24 hours for interconnected VoIP depending on the trigger, an initial report within 72 hours, a final report within 30 days, and a call to the designated 911 facility official within 30 minutes (47 CFR 4.9).

Two of the counts are not people

Here is where the arithmetic becomes a construction rather than a census. For telephony, user minutes are assigned telephone number minutes: duration multiplied by the sum of assigned and administrative numbers potentially affected (47 CFR 4.7). A working number counts as a user, which is generous and defensible and not the same thing as a person.

For other communications the rule multiplies duration by the number of end users potentially affected, then explains what that number is for two classes. For interconnected VoIP to mobile users it is the simultaneous call capacity of the affected equipment multiplied by a concentration ratio of 8 (47 CFR 4.7). For a wireless switch failure it is disabled macro cell sites multiplied by the average number of users served per site, and that average is total users divided by total macro cell sites (47 CFR 4.9).

Read those two sentences and the number that crosses 900,000 stops being a count of anybody. It is a call path times eight. It is a cell site times an average the carrier supplies. Both multipliers are the Commission's own writing, and neither is a measurement of people. That is not a scandal; it is what a rule has to do when it needs one scalar for a network it cannot see. It is worth naming because every downstream sentence about how many users were affected inherits the construction.

The 911 class is not one thing either

The emergency-calling provisions look like the rule's exception, and on a first reading I assumed they were: a class of harm so serious that it bypasses the product. That is not the printed text. All four 911 triggers require at least 30 minutes. Three of them also require at least 900,000 user-minutes. One, the loss of 911 call processing capability in one or more E-911 tandems or selective routers, carries only the duration (47 CFR 4.5).

So the emergency-calling list contains a member with no product condition and three members with one. A 988 lifeline outage is treated as potentially affecting a 988 special facility only where it also potentially affects at least 900,000 user-minutes for at least 30 minutes (47 CFR 4.5). The tightest way to put it: emergency calling was not carved out of the arithmetic. A list was written, and one member of that list is arithmetic-free.

The second threshold

Everything above governs which failures become rows. 47 CFR 4.2 governs who may see them, and it is where this piece turns. Reports filed under Part 4 are presumed confidential under 47 CFR 0.457(d)(1), and inspection runs through 0.461(d)(3) unless the Chief of the Public Safety and Homeland Security Bureau grants a federal, state, territorial or Tribal agency direct access on certification.

The listing is categorical rather than ad hoc. 47 CFR 0.457(d)(1)(vi) places outage reports filed under Part 4 among the materials not routinely available for public inspection, on the basis of 5 U.S.C. 552(b)(4) and 18 U.S.C. 1905.

And here is the only route out, verbatim:

“Agencies granted direct access to information collections may disseminate aggregated and anonymized information to the public. Such information 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.” (47 CFR 4.2)

Three things are in that paragraph. It says may, not shall. It puts a floor of four providers under the aggregate. And it sets the anonymisation test at substance rather than at name, which is a harder standard than redaction.

The final rule that adopted it in 2021 is explicit in both directions. NORS filings are presumed confidential and withheld from routine public inspection. The Commission says it publicly shares limited analyses of aggregated and anonymized data for industry-wide reliability work. And the Commission declined to have the Commission itself produce aggregated and anonymized reports for participating agencies, describing that course as impractical and administratively burdensome. The same order notes that even four-provider aggregation may be insufficient where one provider's data is an overwhelming share of the whole.

The Commission's NORS page repeats the presumption and offers a filer login, while describing how the Cybersecurity and Communications Reliability Division routinely analyses NORS data internally. The NORS User Manual is a filing guide, Version 3 effective 28 October 2021, which restates the presumption. An undated staff technical page supplies the sentence about the other audience: aggregate reliability statistics based on NORS data are reviewed with industry on a regular basis. Reviewed with industry. Not published.

What a hurricane prints and a routine year does not

The contrast inside the same agency is printed rather than inferred. The DIRS page says the Commission relies on that system both to give network status information to emergency management officials and to publish daily public communications status reports with aggregated service status information, and that DIRS reporting became mandatory for cable, wireless, wireline and interconnected VoIP providers under 47 CFR 4.18 effective 20 February 2025. The network reliability resources page says it again: providers submit DIRS reports during emergencies or disasters, and the Commission publishes reports aggregating the information.

So the standing public aggregate in this area is event-driven and daily, and the routine series behind it is closed. A 2013 Bureau report shows the ad hoc version of the same practice: 534 NORS reports for 29 June to 6 July 2012, presented in aggregate, with the confidentiality presumption stated in the report itself. One dated release is not a statistical series and I am not treating it as one.

No reviewed document states why the disaster-time aggregate is printed while the routine record is not. The asymmetry is real, the explanation is missing, and it stays missing here.

What the constant admits, in my arithmetic

Every figure below is mine, derived from the printed constants. None of it is the Commission's, and none of it describes an actual outage.

  • 900,000 divided by 30 minutes is 30,000 users.
  • The same constant at 60 minutes is 15,000 users; at 600 minutes, 1,500; at 1,440 minutes, 625 people out for a full day.
  • 667 OC3-minutes over a 30-minute floor is 22.23 OC3 circuits, which at 155.52 Mbit/s per OC3, a rate the rule does not print, is roughly 3.46 Gbit/s.
  • 90,000 blocked calls in 30 minutes is 3,000 blocked calls a minute, if the outage ran exactly at the floor.
  • The SS7 real-time blocked-call threshold is three times the historic lost-call threshold.

Those five lines say one thing in plain English: the same reportable event can be thirty minutes for thirty thousand people or a full day for six hundred and twenty-five. Which is the worse outage is not a question the rule answers, and I am not answering it either.

Ten boundaries

  1. No raw NORS filing, export or row-level public dataset was obtained, and none turned up in the documents reviewed on 22 November 2026.
  2. No count of failures below any threshold was located, so the size of the excluded population is unknown and is not asserted here.
  3. No evidence was located that any agency has used the 4.2(d) permission, and no published aggregated NORS statistical series turned up in the reviewed set.
  4. The 2013 derecho report is one dated aggregate release, not a recurring publication.
  5. Only the paragraphs quoted here were inspected in 47 CFR 4.9; intervening text was elided in the retrieved extract, eCFR is authoritative but unofficial, and its source note reflects amendments through 89 FR 2513 of 16 January 2024.
  6. The NORS User Manual was read through indexed text of the Commission's own file after a direct fetch returned HTTP 403, so pagination and complete field definitions are unverified; it states Version 3 effective 28 October 2021 while some third-party documents cite the same address as the 2018 version, and that discrepancy stands unresolved.
  7. The eCFR Part 4 landing page returned an automated-access block and was not read; nothing here rests on it.
  8. The 2021 order predates the operational launch of participating-agency access, and the Commission's outage information sharing page was not obtained, so this describes the rule as written and dated rather than administrative practice on any given day.
  9. No reviewed document gives the Commission's reasons for publishing daily DIRS status reports while presuming NORS filings confidential.
  10. No reviewed document says how many agencies hold direct access, what any of them has published, or whether anything published has since been withdrawn.

The arithmetic is public and the rows are not. Anyone quoting an American outage trend is quoting the first and borrowing the second on trust. That trust may be well placed. It is still a loan, and the rule prints the rate and the collateral requirement without ever printing the balance.