Skip to main content
← Flight Data Recovery

What A 403 Actually Means

The mistake​

This site published that N102DZ's FlightRadar24 page was "A REAL REMOVAL, DOCUMENTED." The evidence for that statement was one number: the URL flightradar24.com/data/aircraft/n102dz returned HTTP 403 when a script asked for it.

The 403 was real. The conclusion drawn from it was wrong, and it has been retracted on the overview page.

Here is exactly how the error was made. A script walked five tracking sites for each of the sixteen aircraft in this investigation and recorded the status code each site returned today. FlightRadar24 returned 403 for all sixteen. So did radarbox.com. So did planespotters.net. FlightAware and ADS-B Exchange returned 200 for all sixteen. Eighty probes, and the pattern looked like three sites had shut the door on this fleet.

What was never asked was whether those three sites serve anybody at all. They do not — not to a script. The number that would have caught this took eleven minutes to collect and nobody collected it until after the claim was published.

The three things a 403 can mean​

403 Forbidden is the server saying "I am not giving you this." It does not say why. From outside, these three situations are indistinguishable:

What is actually happeningWhat you see
The page was removed — the aircraft was taken off the site, blocked, or delisted403
The site blocks robots — a bot filter, a CDN challenge, or a missing browser fingerprint403
The site is rate-limiting — too many requests from one address, too fast403

Only the first one is a finding. The other two are the weather. And the response body is no help either: FlightRadar24's 403 is 5,749 bytes of challenge page for N102DZ, and 5,749 bytes of the same challenge page for a Ryanair 737. Identical, to the byte.

This is exactly why the control test is not optional. A status code on its own is a measurement with no baseline. It becomes evidence only when you have something to compare it against.

The control test​

The rule is one sentence, and it is the whole method:

A non-200 is only evidence about an aircraft if the same site serves aircraft that have nothing to do with the case.

So the identical script, from the identical address, seconds apart, was pointed at five aircraft chosen for irrelevance — two mainline airliners, one regional airliner, and two business jets with no connection to this case, this investigation, or anyone in it:

Control aircraftWhat it isWhy it was chosen
EI-DCLRyanair Boeing 737-800A European short-haul workhorse flying six sectors a day. Nothing rarer or better-documented exists.
D-AIMALufthansa Airbus A380A flagship widebody. If any page on earth is public, this one is.
N904XJDelta Connection CRJ-900, operated by Endeavor AirA US regional jet — ordinary domestic traffic, not a flagship.
N102UWAmerican Airlines Airbus A320A US mainline narrowbody, and a tail number one character from a subject.
N1KENike corporate Gulfstream G650The closest match to our subjects that has nothing to do with the case — a privately-owned long-range business jet that sits on the ground most days.

That last one carries most of the weight. The subjects in this investigation are private jets. If a private jet unconnected to any of this behaves the same way, the behaviour is about private jets, not about this case.

What came back​

Five sites, five controls, five subjects — fifty requests in one run:

Tracking siteControls servedSubjects servedWhat a non-200 from this site means
flightradar24.com0 / 50 / 5Nothing. The site blocks all scripted requests.
radarbox.com0 / 50 / 5Nothing. The site blocks all scripted requests.
planespotters.net0 / 50 / 5Nothing. The site blocks all scripted requests.
flightaware.com5 / 55 / 5Read it per aircraft — this site answers scripts.
globe.adsbexchange.com5 / 55 / 5Read it per aircraft — this site answers scripts.

Three of the five sites refused every single request, control and subject alike. The Lufthansa A380 is as forbidden as the Kirk family Gulfstream. Whatever else that shows, it cannot show anything about N102DZ.

The two sites that do answer a script — FlightAware and ADS-B Exchange — served every one of the sixteen aircraft in this investigation, HTTP 200. Not one refusal on either.

The full run, every URL and every byte count, is in page_control_probe.json.

The browser test​

The control test says the 403 is meaningless. It does not say what the page actually contains. For that the pages have to be opened the way a person opens them — in a real browser, where the bot check is satisfied by the browser itself.

All sixteen subjects and four controls were loaded from a logged-out FlightRadar24 session on 24 August 2026. Every column below is what the page rendered:

TailKindHTTPRows rendered"could not find data"Type
N904XJCONTROL200699noCRJ9
EI-DCLCONTROL200683noB738
N2100LSUBJECT200378noGLEX
N872RASUBJECT200294noBE40
D-AIMACONTROL200196noA388
N59906SUBJECT200110noPA31
SU-BTTSUBJECT20056noFA7X
SU-BTUSUBJECT20056noFA7X
N40JDSUBJECT20056noPRM1
SU-BTVSUBJECT20054noFA7X
N560TWSUBJECT20028noC56X
N1098LSUBJECT20028noGLEX
SU-BNDSUBJECT20014noGLF4
N102DZSUBJECT2000yesGLF5
N888KGSUBJECT2000yesCL30
N582MMSUBJECT2000yesLJ60
SU-BGMSUBJECT2000yesGLF4
T7-ELLSUBJECT2000yesGLEX
N55906SUBJECT2000yesP28R
N1KECONTROL2000yesGLF6

Twenty pages. Twenty HTTP 200s. Every one loaded with its full identity block — N102DZ as Gulfstream V, GLF5, Mode S A00C85. Nothing has been removed from FlightRadar24, for any aircraft in this investigation.

The pattern in that table is recency, not relevance​

Read the bottom seven rows against the top five. The empty pages are N102DZ, N888KG, N582MM, SU-BGM, T7-ELL, N55906 — and N1KE, which is a control. Nike's Gulfstream, an aircraft with no connection to anything on this site, produced character-for-character the same result as the tail number at the centre of the dispute, on the same day, in the same session.

Every one of those seven is a privately-owned aircraft that had not flown in the previous week. Every aircraft that had flown that week shows rows, and the row count tracks how hard it was flying: the Ryanair 737 doing six sectors a day shows 683, the Delta Connection CRJ 699, the Lufthansa A380 196, and among the subjects N2100L shows 378 and N872RA 294.

Sort that table by "is this aircraft part of the investigation" and it makes no pattern at all. Sort it by "did this aircraft fly in the last seven days" and it sorts perfectly.

The paywall, which is the real answer​

There is a reason the free page only ever reflects the last week, and FlightRadar24 prints it under the table in plain language. On N102DZ's page it reads:

"More than 7 days of N102DZ history is available with an upgrade to a Silver (90 days), Gold (1 year), or Business (3 years) subscription."

That line appeared on all twenty pages — every subject and every control, the empty ones and the 699-row ones alike. A logged-out visitor is shown seven days. Everything older is a paid product and always has been.

This is the sentence that matters most on this page:

The multi-year flight history the overlap spreadsheets were built from was never on the free FlightRadar24 page. It cannot have been removed from it.

Somebody who checked N102DZ in 2023 and again in 2026 and found "the history gone" is describing the seven-day window rolling forward, which is what it does every day for every aircraft on the site. An empty table on a private jet is the normal state of that page, not an event.

The same test applied to the other two claims​

FlightRadar24 was the third of three "erasures" this site had published. The other two failed the same control test, in the same way, and both are set out in full on the overview:

  • The adsb.lol 12 October – 15 December 2025 band. The live API returns HTTP 403 for those dates. It returns 403 for those dates for every aircraft asked, including airframes with no connection to this case. And adsb.lol publishes those exact days itself, in full, as one Open-Database-Licence release per day on GitHub — v2025.10.12 and v2025.10.13 among them. An organisation that publishes the data on GitHub is not suppressing it on its API. It is a serving condition.
  • adsb.lol holding no 2026 data. An unrelated Ryanair airframe and an unrelated Lufthansa airframe return 404 identically for 2026 dates. It is archive-wide retention, and the hole is wider than first described rather than narrower — which is the correct direction for a finding to move when it is tested.

Three claimed removals. Three control tests. Three failures. In each case the control aircraft behaved exactly like the subject aircraft, which is the definition of a result that carries no information about the subject.

What would actually prove a removal​

None of this means data never gets removed from tracking sites. It does — owners request it, and the mechanisms are public. The point is that proving it takes more than a status code. Here is the checklist this site now uses:

  1. An archived copy showing the page populated. A Wayback Machine snapshot, or any dated third-party capture, in which the page carried the content that is now missing. Without a "before" there is no removal to demonstrate.
  2. The same URL served to a BROWSER returning 404, or an explicit block notice, while control aircraft load normally in the same session. Not a script. A browser, and controls alongside it in the same minute from the same address.
  3. The tracking site's own block list, or the FAA's LADD list, naming the tail. Both are documented programmes with published mechanics. A tail that has been blocked can be shown to have been blocked.
  4. The data absent from independent networks the site does not control. A removal at one company does not touch volunteer ADS-B networks. If the tracks are still retrievable from archives that company has no authority over, nothing has been destroyed — at most one publisher stopped displaying it.

Not one of these four conditions is met for any aircraft in this investigation. There is no "before" snapshot showing content since removed. Every page loads 200 in a browser. No block list or LADD entry names any of these tails. And the underlying ADS-B record is retrievable from independent archives — including throughout May 2026 itself, the month the removal was reported to have happened.

Why this page exists​

An investigation that calls a robot block a cover-up hands its opponents the entire argument. One demonstrable overreach is all anybody needs to dismiss every careful finding sitting next to it, and they will be right to, because the same standard produced both. This site drew a conclusion from a number it had not baselined, published it in bold, and had to take it back. The correction is the credibility. The rule that came out of it — ask an unrelated aircraft the same question before you conclude anything about this one — is now applied to everything on these pages, and it will keep removing claims as readily as it supports them.

Reproduce this​

Both halves of this page can be re-run by anyone. Neither needs an account, a key, or a payment.

# the control test: 5 unrelated aircraft + 5 subjects x 5 tracking sites
cd site/docs/Planes/following/apis/public_open_source/code/
node control_page_probe.js
# writes ../data/recovery/page_control_probe.json

The browser half was done by hand, by opening the twenty pages in an ordinary browser — and that is the entire point. A script cannot pass the bot check, which is why a script sees 403 and a person sees the page. Any reader with a browser and no account can load flightradar24.com/data/aircraft/n102dz right now and see the identity block, the empty table, and the subscription line under it. Then load flightradar24.com/data/aircraft/n1ke and see the same three things on an aircraft nobody has ever claimed was involved in anything.

The captured result of that pass is committed at site/docs/Planes/following/apis/browser_capture/captures/fr24_page_availability_2026-08-24.tsv, one row per aircraft, with the HTTP status, byte count, rendered row count, the no-data flag, the paywall flag, and the Mode S identity block for each.