36 Years of Outages and the FAA Ground Stop: Same Strands, Different Technology

Posted on November 16, 2026

0


Jon W. Hansen, FCIPS | Procurement Insights | Hansen Models™


On Friday, 13 November 2026, at 3:36 p.m. Eastern, the FAA stopped every departure in the United States. By the weekend, thousands of flights had been canceled.

The next day, David Yarkin, the founder and CEO of Procurated and a former chief procurement officer of Pennsylvania, wrote about it after being stranded himself. His point was that vendor performance matters, and that it needs to be tracked over time, so that a slow decline is caught before it becomes a crisis.

He is right that vendor performance matters. But one detail from the government’s own briefing points past the vendor, and in my comment on his post I said so. This post is the longer version of that comment, because the detail turns out to be 36 years old.

That is the claim this post tests. I call it Invariant Physics™: the technology changes, but the strands that decide the outcome do not. If that is true, the strands behind Friday’s ground stop should show up in outages built on entirely different technology, decades apart. Let’s see whether they do.

What we know so far

These are preliminary statements from officials. The FAA’s root cause analysis has not been published.

  • An L3Harris crew was replacing a router during a copper-to-fiber upgrade at a remote FAA radio site in El Dorado, Arkansas. Transportation Secretary Sean Duffy called it “pretty straightforward work.”
  • The FAA says its initial investigation points to two sites as the source of the overload: the El Dorado site and a radar site in Amarillo, Texas. Together they sent about 1,000 connection attempts every four to five seconds. Instability began at 17 of the 22 centers that manage en route air traffic, and the overload lasted 45 minutes, according to FAA Administrator Bryan Bedford.
  • The FAA has since disabled a broadcasting function on about 60 similar routers as a precaution, and says it is still investigating the root cause.
  • Asked why the work was done on a Friday afternoon, Secretary Duffy said large airports switch over at night, while remote sites switch during the day. He added that “there will be reevaluation of when we do this work.”
  • Officials are still deciding whether the cause was user error, an installation error or a faulty router.

The strands that crossed

How do you find the strands? Not by guessing at causes. Hansen Strand Commonality™ starts from the outcome, a national ground stop, and traces back through every group that touched the cutover, asking each the same questions. Who did the work? Who coordinated it? Who set the timing rule? What did the site connect to? What pressure was the program under? Who staffed it? Who could have stopped it? Each question that has an owner and a public answer is a strand.

Start with the timing rule. It judges a site’s risk by how busy that site is. It does not judge risk by what the site is connected to. A routine job at a “remote” site, done in the middle of a Friday afternoon, became part of an overload that reached most of the national network.

That is one strand. Tracing back produced six more.

  • The vendor: L3Harris, which has run key parts of the FAA’s telecom network for more than two decades.
  • The integrator: Peraton, which coordinates the modernization program.
  • The clock: a Friday afternoon cutover, under a rule that sends remote sites through in daylight.
  • The coupling: a network in which traffic from two remote sites could destabilize 17 of 22 en route centers.
  • The pace: in a September 2026 report, the Government Accountability Office (GAO) warned that the FAA was “buying delivery speed at the risk of cost uncertainty and potential disruptions.”
  • The staffing: the FAA acknowledges its technician workforce is short.
  • The authority: who approved a daytime cutover at a site that could reach the national network, and who had the authority to stop it?

These are conditions under examination, drawn from public statements. None is established as a cause. That is what the FAA’s findings, and our baseline, will test.

Each strand makes sense on its own. Remote sites are quieter, so daytime work seems reasonable. The equipment is old, so modernizing quickly seems reasonable. An experienced contractor seems reasonable. None of them, on its own, predicts a national ground stop. The failure happened where they met.

A vendor scorecard watches the first strand. It cannot see the crossing.

It has happened before

Here is the test. If Invariant Physics holds, the same strands should appear in outages with different technology, different owners and different decades. They do. The same strands have crossed, with the same effect, through a different technology each time.

January 1990. A switch in AT&T’s New York long-distance network reset itself and came back online. Its messages triggered a one-line software bug, introduced in a routine update the month before, in the neighboring switches. Each switch’s recovery crashed the next. The failure moved across the network for about nine hours, and more than 50 million calls were blocked.

September 1991. During a heat wave, AT&T switched its Manhattan facility to its own generators at the utility’s request. A power surge tripped the backup batteries, which ran down. The alarm sirens malfunctioned and the visual warnings went unnoticed for more than five hours. According to the telephone workers’ union, technicians who would have responded were attending a class on new alarm systems. Air traffic controllers in New York lost contact with planes.

November 2009. A faulty card in a router in Salt Lake City failed, and the router did not switch to its backup. The replacement part was in a storage closet, and the only person with access had not yet arrived at work. Commercial flight plans could not be filed nationwide for hours. Harris, now L3Harris, ran the network.

January 2023. Contract staff working to fix a sync problem between the live and backup databases of the FAA’s NOTAM system unintentionally deleted files. Flights were halted across the country.

Telephone switches, power systems, routers, databases and fiber. A different technology each time, over 36 years. Yet each time:

  • The trigger was routine. A software update, a power switchover, a database fix, a cable upgrade.
  • The spread came from coupling. One node could reach everything, and the backup either failed or shared the same fault.
  • People sat at the crossing. Technicians in a classroom, one person with a key, a workforce that is short today.

What doesn’t transfer

The test has limits, and they should be stated. The 1990 and 1991 outages happened on AT&T’s telephone network, not the FAA’s. The owners, the regulators and the economics were different, and so were the specific failures. Only 2009 is a close analog: the same vendor lineage and a national effect on flight operations. What carries across the cases is not the details. It is the pattern: a routine trigger, coupling that let one point reach everything, and people at the crossing.

In 1990, one switch’s recovery messages crashed its neighbors. In 2026, connection attempts from two remote sites overloaded theirs. The technology is 36 years apart. The behavior is the same.

That is the result of the test: five outages, a different technology each time, and the same three strands. What Strand Commonality adds is the practice of finding where those strands cross, before the crossing finds you.

The strands are not a closed set, though. New strands join over time, and AI is the latest. The FAA rolled out an AI tool for air traffic controllers the day before the ground stop. Officials say it played no part, and there is no evidence it did, but it is now a strand in that system, and the next routine change will cross it. New strands don’t replace the old ones. They add new places for the old ones to cross.

Why nobody looks at the seams

If the strands are this consistent, why does it keep happening? It is not for lack of warning. As Fast Company has reported, the Department of Transportation’s Inspector General warned in 2009 of a continuing risk of critical outages. In 2023, an expert panel commissioned by the FAA warned that its facilities and equipment were “elevating system risk to unsustainable levels.” And in its September 2026 report, the GAO warned that schedule pressure could lead to “more system outages, ground stops, flight delays, and safety incidents.”

The problem is structural. Every strand has an owner. The contractor owns its work, the integrator owns the schedule, operations owns the timing rules and finance owns the budget. Each is hired, measured and rewarded for its own strand. Nobody is assigned to the seams between them. There is no shared map that puts the strands side by side. And a “routine” change, by definition, gets no pause to ask what it touches.

So five groups can each do everything right, and the system can still fail. The failure is not in any one strand. It is in the seams.

What we are doing about it

The FAA’s root cause analysis has not been published. So today, Hansen AI Labs™ locked a baseline: a dated, fingerprinted record of which strands we expect the official findings to name, and which we expect them to leave out.

When the findings are released, an independent model on our panel will score each expectation. The ones we get wrong will be studied as closely as the ones we get right. Anyone can explain a failure after the cause is announced. Writing down the strands before it is announced is the test.

The baseline also states what would prove our reading wrong. If the FAA’s findings name the crossings, including why two remote sites could reach 17 of 22 en route centers and who had the authority to stop the work, and if the fix changes the design rather than a single device, then the official analysis saw the strands. If that happens, we will say so.

History gives one more reason to watch. After the 2009 outage, the first account named a faulty card. According to Fast Company’s reporting, the Department of Transportation’s Inspector General later found configuration and procedural errors as well. The first explanation is rarely the whole one.

The question

If you manage a network, a supply chain or a program of any kind, the question is not whether your vendor is performing. It is this:

What does your next “routine” change connect to, what clock is it running on, and who has the authority to stop it?

Answering it means widening the aperture of your strand understanding. Most of us watch the strands we know best: our own function, our own vendor, our own schedule. But the strands you don’t see are still there, whether you see them or not, and they are the ones that will surprise you. Think beyond the strands you are most familiar with. That is where the next “routine” change will find the crossing you missed.

Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™

-30-

Posted in: Commentary