They Descended From 80,000 Feet For Nine Days. Then Someone Rolled Back The Patch.
A defect ticket, a build number and a rollback account for the radar. The supervisor mocked on television had diagnosed it on day three, on the right form. It explains nothing the aircrew reported.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
SAN DIEGO, Calif. — Discrepancy report 11-4472. Priority 3. Filed at 1120 local on 10 November from a console eighty nautical miles off San Diego, by the supervisor of the watch.
It describes an object falling from 80,000 feet to 50 feet in under a second.
It describes it as a fault.
It was closed eleven weeks later with one line: fixed in 7.3.3.
What $240 bought
The descents are famous. The build number is not, and it has been sitting on a public procurement portal since 2007.
Freedom of information got us the discrepancy log, after two appeals and a redaction fight we half lost. It took twenty-eight months. But the software configuration record — which build was running on which set, to the minute — is in Kestrel Marine Systems' own delivery documentation, and Kestrel filed that documentation as a contract deliverable, and the portal charges $240 for it.
The portal's access counter says four people have ever downloaded it. One of them is this newspaper.
Nine days
Descents appear on nine consecutive days: 8 to 16 November. Between four and thirty-one a day. Not one on a track carrying a correlated identification return.
Before 8 November: none in the preceding six weeks.
After 16 November: none, on either set, ever.
Build 7.3.2 went live at 2140 the night of the seventh. The first descent is logged at 0655 the following morning.
What 7.3.2 changed
One paragraph of the release note, and it is not a paragraph anybody was hiding.
Until 7.3.2, the system needed several consecutive returns inside a range gate before it would open a track file. 7.3.2 lowered that threshold so a track could be initiated from a single unresolved return — and it took the new track's altitude from whatever the first elevation estimate happened to be.
When that first estimate was rubbish — a sidelobe, a duct, a bit of sea clutter arriving at the wrong angle — the system opened a file at an absurd height. On the next good return it corrected the altitude. In one frame.
Dr. Wilhelmina Achterberg-Osei, given the release note and forty-one days of track history without being told what we were hoping for, wrote it in a sentence.
The rollback
Tomasz Weerakoon-Hale was the contractor's field service engineer. On the afternoon of the sixteenth he put both sets back on 7.3.1. The configuration record gives thirty-four minutes and carries his signature.
The last descent in the log is 1338. The rollback completed at 1352.
There is nothing after it.
We also checked the sky
Because a software fault on two ships is still two ships, we pulled the Federal Aviation Administration primary-return anomaly log for the overlapping sector and the same nine days — PX-1330, in the Vault.
Civil radar in that sector logged the ordinary number of unresolved primaries for November. Nothing correlates with a single one of the descents.
That is consistent with a fault aboard. It is not proof of one, and Dr. Achterberg-Osei's report contains a paragraph explaining, at some length, that our story is smaller than we think it is.
Ten minutes later
At 1352 the radar was on the previous build.
At 1402 four aircrew in two aircraft reported a visual.
The set was clean. The descents were over. Whatever those four men described, they described it to a system that had stopped doing the thing this story explains, ten minutes earlier, and had a signature in a log to prove it.
What this desk takes from it
A man has spent twenty-two years being called either a witness or a liar, and he was neither. He was a competent operator who watched a track-initiation regression, diagnosed it correctly, and filed it on the right form with a priority code.
The impossible descents have a ticket number.
That is the whole of what we established, and we want to be blunt about the size of it, because a story this tidy invites a conclusion it has not earned. We have explained a radar. We have not explained an afternoon.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We stopped asking what the radar saw and asked what the radar was running. The software configuration record says which build was on which set and at what minute; the discrepancy log says what the watch floor thought of it at the time. We laid one against the other, day by day, and then gave both to a radar engineer with no connection to the contractor and no idea what we wanted her to find.
Who we spoke to
- Senior Chief Petty Officer (ret.) Halvard Ocampo-Reese, Watch supervisor, air-search radar, at the time. Interviewed in person twice, in Chula Vista; read the configuration record before he answered anything March and May 2026 His reply to this newspaper is printed in full and is the best thing on this page.
- Tomasz Weerakoon-Hale, Field service engineer, Kestrel Marine Systems, 1998–2011. Three telephone interviews; confirmed his own signature on the rollback entry April 2026 Confirmed the thirty-four minutes. Declined to discuss anything that happened after 1352 and said plainly that he was ashore by then.
- Software configuration record and discrepancy log, two air-search sets, Obtained under freedom of information after two appeals. Requested October 2023; released in three tranches, redacted; published here entire, redactions included 2023–2026
- Kestrel Marine Systems build baseline delivery documentation, Purchased from a public procurement portal. Bought for $240 with a credit card. It was never withheld from anybody. January 2026 The portal's own access counter records four downloads since 2007, of which one is ours.
- Dr. Wilhelmina Achterberg-Osei, Radar systems engineer; no connection to the contractor. Given the release note and forty-one days of track history with no indication of what we hoped for June 2026 Her report is published in full, including the paragraph in which she says our story is smaller than we think it is.
Documents
- PX-1330 — Federal Aviation Administration primary-return anomaly log, Sector 41, extract accepted
What we could not confirm
- What four aircrew reported at 1402 on 16 November. The set had been on the previous build for ten minutes and was behaving normally. Nothing in the discrepancy log speaks to it, this story does not explain it, and we are not going to let the neatness of the rest of the page suggest otherwise.
- Whether the two sets failed independently. Both took build 7.3.2 within an hour of each other and shared a track-management feed, so a common-mode software fault and a real object over a formation produce the same log. This record cannot separate them and Dr. Achterberg-Osei says so in her report.
- Why the ticket took eleven weeks to close, and who closed it. The closure text is one line. The originator field is redacted under a personnel exemption we appealed twice and lost twice.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
Signal & LedgerKestrel Pulls Air-Search Build After Fleet Regression
Reported the rollback in its defence-software column at the time, correctly, in 180 words.
Had that we did not
The build numbers, the rollback, the eleven-week fix. Everything.
Left out
That the fleet in question was the one everyone would still be arguing about in 2026. The trade press had the answer for twenty-two years and nobody read the trade press.
-
The Hollow PostTHEY ROLLED BACK THE SOFTWARE TO HIDE WHAT THE RADAR SAW
Treated the rollback as suppression: the descents stopped because the system was blinded.
Had that we did not
The rollback, the date, and the fact that the descents stopped that hour. All true.
Left out
The ticket. A watch supervisor asked for that rollback six days earlier, in writing, and their piece does not mention that a request exists.
-
They Buriedthis newspaperThey Descended From 80,000 Feet For Nine Days. Then Someone Rolled Back The Patch.
Matched the software configuration record against the track history, day by day.
Had that we did not
The configuration record, the discrepancy log, and an independent engineer's reading of both.
Left out
Our headline says 'someone rolled back the patch', which makes an ordinary field action sound furtive. A named engineer did it in the afternoon and signed the log, and our own timeline says so four rows down. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Senior Chief Petty Officer (ret.) Halvard Ocampo-Reese read the full draft, the configuration record and Dr. Achterberg-Osei's report; given eighteen days and no word limit. Replied on 2 August 2026 by email, in one message, at 0430 his time. Printed unedited.
I want to thank you and then I want to tell you what you have done.
You are the first outlet in twenty-two years to read the ticket. Everybody else read the numbers off the ticket, put them in a headline, and then called me either a hero or a fantasist depending on what they needed that week. So thank you. Genuinely. My wife has read your draft twice.
Now the other part.
You have written a story in which I am vindicated, and the shape of that story is that I was right about the radar. I was right about the radar. It took me two and a half days and I have never once claimed it took skill. Any supervisor on any watch floor who saw an initiation walk a track from angels eighty to fifty feet in one frame would have written the same ticket, because that is not an aircraft, that is arithmetic having a bad afternoon.
Here is my difficulty with your page. For twenty-two years those descents have been quoted as the evidence. People who have never sat a watch have read my ticket numbers out on television as though they were a sighting. I have said, on three separate programmes, that the radar behaviour was a fault and that I had reported it as a fault, and I was cut every time, because it is a boring thing to say and it ruins the segment.
So now you have printed it and I am grateful and I would ask you to notice what happens next. Every person who has been using my descents to argue for something is going to have to give them up. Good. And every person who has been using my descents to argue against something is going to say the whole thing is closed. That is the part that is not good, and your paper is going to be the citation for it, and you should sit with that.
At 1402 on the sixteenth the set was on 7.3.1 and had been for ten minutes and it was clean. Four men in two aircraft said what they said. I was on the floor. I did not see anything, I want to be precise about that, I saw a display behaving properly and I heard four voices. I have no explanation and I have never offered one and I resent that this is the first time in twenty-two years anybody has let me say it without laughing.
One correction to your draft, which is small and which I would like made. You have me filing on day three. I filed on day three because the second day was a Sunday and the form goes through a chief who was not aboard. If he had been aboard it would have been day two. I would rather not be thanked for a delay caused by a duty roster.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.