Lab · Open-data check · updated 25 September 2026

Biondi and Malanga's SAR tested on the known chambers

I rebuilt Filippo Biondi and Corrado Malanga's SAR Doppler tomography method from the public documents and tested it on the chambers of the Great Pyramid of Khufu that we already know, using Capella Space open radar data. Here are the rules fixed before looking, the results, the figures, the code, and what this test does not say.

The story of the paper

Filippo Biondi and Corrado Malanga published the method in Remote Sensing on 19 October 2022. The journal retracted it on 10 August 2026 for serious methodological flaws and statistical errors, without saying which. The authors disagree with the retraction.

The review report published by the journal shows how it got there. On the first version a reviewer asked for axes in metres and for the known structures as a control experiment, and wrote that the overlays with the pyramid sketch gave the impression of having been adjusted by eye. On the second version, read by three other reviewers, one writes that validation is missing, Lawrence Conyers calls it well done but says he cannot assess the equations, and Gaetano Ranieri calls it rigorous and adds that its validity can only be checked by excavation or geophysical methods. On 31 August 2026, speaking to Retraction Watch, Conyers said the method does not work.

A test of that kind, looking with the method for the structures we already know, is the one you find below.

The method, as they explain it

At a conference in Spoleto, Biondi describes it this way. The Earth's surface always vibrates a little. A single radar image can be unpacked into a sequence, like a film, and those vibrations are extracted from it. Where there is a void underneath the vibration is missing, and in their sections that void is blue.

To turn vibrations into metres of depth you need two numbers, the speed of sound in the stone and a frequency. The satellite measures neither. In Spoleto Malanga says "let's assume, we don't know, that this stuff is granite, 4,000 metres per second". In their book Khnum-Khufu the authors write that with frequencies close to 4.32 Hz and its multiples "things went better". And in their documents the resulting scale changes. In the 2022 preprint it is 6,000 metres per second divided by 12,500 Hz, that is 0.48 metres. In the published version it becomes 0.24 metres, because the calculation is written with 25,000 Hz while stating 12,500. In the patent it is 6,600 divided by 22,000, that is 0.30 metres.

On the Symposium Podcast, on 15 July 2026, Biondi also says that they threw many tests away because they made no sense, "very often we were inverting noise", and kept the ones that worked.

The test on the known chambers

The chain

Biondi's program is not public. I rebuilt it from the 2022 paper, the patent and the public documents of the BiondiProtocol project, which describes the recipe with the directions attributed to Biondi. In short, 50 pairs of sub-images taken a moment apart are extracted from the radar image, the shift of every point between the two is measured to one hundredth of a pixel, the disturbance common to the whole scene is removed, and a depth profile is derived from the sequence of shifts. Part of the parameters comes from BiondiProtocol, which is an anonymous project. I say so because it matters.

The data

Capella Space and Umbra are the two open data sources named by the Khafre Project in its statement of 16 March 2025. In this test I used 2 Capella passes over Khufu, of 16 September and 4 October 2024, downloaded from the Capella open data registry.

The rules, fixed beforehand

Protocol and code were sealed on 23 September 2026 at 13.17.29 UTC, with a digital fingerprint of every file, after the checks on synthetic data and before opening the radar images. The log shows that image processing starts a few seconds later. The fingerprint of the seal file is d74fb2f593ca74bf627fe785848230ec7cf2e1b81914eeea8a328e266803da71, and anyone downloading the files below can recompute it.

• The scale. I tried 41 of them, from 10 centimetres to 10 metres, a range that includes all the values I found in their documents. I kept the one that made the King's Chamber, whose position we know, stand out most from the stone around it. It is an advantage given to the method.
• The chambers. With that scale locked, I looked for the Queen's Chamber, the Grand Gallery and the Subterranean Chamber on the 4 October pass, and for all 4 chambers on the 16 September pass. Each chamber is read at its own height, taken from Petrie's survey. The scale is the ruler, not the height.
• The control. For each chamber I measure how much the image inside its volume stands out from the shell of stone around it. Then I do the same at 24 positions shifted by 6 and 12 metres east and north. A chamber counts as found only if it stands out more than every valid position, with the same sign as the calibration and with at least 16 valid positions.

The result

No chamber shows up. The protocol counts the 3 tests without a value as not found, but I list them separately because they are not measured failures. At those points the calculation produces nothing to compare.

Why nothing shows up

The pairs of sub-images compared by the recipe share 95 percent of the spectrum and are almost identical. On the 4 October pass, with the main bank of pairs, 97.3 percent of the shifts measured along the flight direction are exactly zero, and along the other direction 226,792 readings out of 226,800 are. With the second bank of pairs zeros reach 99.8 to 100 percent. The non-zero values are 1 to 4 hundredths of a pixel, the smallest step the estimator can read. The depth profile is built on this rounding, not on measured vibrations.

And with a finer instrument

To rule out that the limit was only precision, I added a second test, sealed on 23 September at 13.49 UTC, with an estimator ten times finer, one thousandth of a pixel. Zeros drop to 13 percent, but the values have no coherence between nearby non-overlapping points, a correlation between 0.003 and 0.023 in one direction and a checkerboard pattern in the other. The best calibration gets the King's Chamber to beat 22 positions out of 24. The Queen's Chamber on 4 October reaches 23 of 24, the case closest to the threshold, but on the 16 September pass it is weak and of opposite sign, and the King's Chamber itself, with the calibration locked, comes out with the opposite sign. The result does not repeat. Again, no chamber.

All the tests, 3 to 23 September

The test on the chambers of 23 September is the last of a series. From 3 to 23 September I ran 101 tests on the method, and you will find them all here, including the ones I withdrew and the ones that gave results in favour. 44 are on real radar data, 22 on synthetic data, 15 are code and calculation checks, 20 are readings of the sources. In 45 of them the rule for success was written down before seeing the results.

The variants of their method I tried

• The tomogram obtained from the phase of the series of sub-images, as in the first reading of the paper.
• The same obtained from the magnitude, as they describe it in their conferences. Later an independent review made me realise it was a surrogate, not their method, and I said so.
• Their chain rebuilt piece by piece, with the sub-images, the tracking of the shifts and the inversion with the pseudoinverse.
• The same chain with the Fourier transform on the complex series and the magnitude only at the end.
• Tracking on Umbra raw data, pulse by pulse, with artificial signals added to see whether the method finds them.
• A full physical model, from the vibration of the chambers to the phase the radar should measure.
• The depth scale in metres from the paper's formula, with their numbers and with others.
• The BiondiProtocol recipe, at one hundredth and one thousandth of a pixel.

The results in favour of the method

I list them because they exist. The first step of the method, measuring small movements from the satellite, gives a hint near the power stations at the Mosul dam. On the pyramid the order of the sub-images carries information, with a chance probability of 0.002, but the same test also responds on a hill with no cavity and not on the body of the dam where the tunnel is, so that information comes from the shape of the surface. In the computer models the positive controls succeed. In mid-September some favourable results came out on the Grand Gallery and the subterranean chamber, but a later check showed they were artefacts of the calculation, and I withdrew them. In the test at one thousandth of a pixel the Queen's Chamber reaches 23 out of 24, below the threshold and not repeated. No test recognised the chambers in a repeatable way on the two dates, and no test shows that the cavities they talk about do not exist.

The tests, phase by phase

Open a phase to see the tests, with the variant used and the outcome. The full entries, with the numbers, the criteria and the corrections made afterwards, are in the downloadable catalogue below (in Italian).

Others who tried

I am not the only one. Other independent researchers, meaning outside Biondi and Malanga's group, tested the method with their own reconstructions, different from one another, and published their code and results. None of them finds the structures.

Two more names circulate. Jashua Luna published his own version of the program, but that it does not get the same results is reported only by press articles. Emmanuel Pio Pastore analysed the Vesuvius case and argues that some negative replications are not equivalent to the published method, because they lack displacement tracking and image pairs. The chain on this page has both.

The Gran Sasso and the demonstrations

In Spoleto Biondi says that when they tested the method on the Gran Sasso mountain, they already knew about the laboratory inside and wanted to see if it could be seen from space. The point is one of method. A test where you already know what to look for and where is not an independent check, because in this method the slice to look at and the depth scale are chosen by whoever runs the calculation. And those data are not public, so nobody can check them again. I knew where Khufu's chambers were too. The difference is that I fixed and sealed the rules before opening the images, and compared each chamber with the positions around it.

What this test does not say

• It does not say there is nothing inside the pyramid.
• It is not Biondi's private program and not his COSMO-SkyMed data. It is a declared reconstruction on public data, and the result holds for that.
• Part of the recipe comes from an anonymous project, BiondiProtocol. Whoever defends the method can say the real chain is different, and that is a fair objection until the real chain is public.
• 4 of the 7 tests can be assessed. The other 3 cannot be measured.
• I had already looked at the two Capella dates in earlier tests. The rules of the 23 September test were sealed before opening the images, but the data were not new to me, so the test is not blind in the strict sense.
• The 97 percent of zeros is a property of this estimator. The ten times finer estimator removes the zeros, but the chambers still do not show up.

The request to Biondi

I wrote to Filippo Biondi asking for his data and his program. If he sends them, I will run the test again with those and publish it here, whatever the result.

The first tests of 4 September, and why I withdrew them

A first set of tests, published here on 4 September 2026, used a simplified version of the chain, without the displacement tracking the paper describes. An independent review of my work found errors in some conclusions of that page. Comparing the 23 nanometres of seismic noise with the 0.41 millimetres of sensitivity set two different quantities side by side, and the theoretical margin, in the most favourable case, is about 4 times, not 17,826. The criterion "the true order must beat the shuffled one by 3 times" had no basis. And the shuffle test, redone with 999 permutations, did pass the check of the simplified version on one Khufu scene, a result that instrumental effects can also explain and that says nothing about the subsurface. For this reason on 12 September I took the page down and redid the test from scratch, with the rules fixed beforehand. The 4 September report and the independent replication can still be downloaded below for transparency, with this warning.

Data, code, files

The protocol and outcome files are in Italian; the numbers and the code are language-independent.

Sources

• Biondi F., Malanga C., "Synthetic Aperture Radar Doppler Tomography Reveals Details of Undiscovered High-Resolution Internal Structure of the Great Pyramid of Giza", Remote Sensing 14(20) 5231, 2022, DOI 10.3390/rs14205231, retracted. Review report, mdpi.com/2072-4292/14/20/5231/review_report. Retraction notice, Remote Sensing 18(16) 2679, 2026, DOI 10.3390/rs18162679. arXiv preprint 2208.00811.
• H. Santoro, "Journal retracts paper claiming network of corridors inside the Great Pyramid of Giza", Retraction Watch, 31 August 2026.
• Malanga C., Biondi F., Khnum-Khufu, 2023. Khafre Project conference in Spoleto. Symposium Podcast, interview with Filippo Biondi, 15 July 2026. Khafre Project statement of 16 March 2025.
• Capella Space Open Data, registry.opendata.aws/capella_opendata.
• Petrie W. M. F., The Pyramids and Temples of Gizeh, 1883, for the chamber heights.
• Hassanforeman, github.com/Hassanforeman/subsurface-sar-tomo. teoAmico, github.com/teoAmico/ResonarSat. Pomposi S., Zenodo 10.5281/zenodo.19574701, 2026. ejhong, github.com/ejhong/sar. BiondiProtocol, anonymous project on GitHub, September 2026. Luna J., github.com/not-JASH/sar-doppler-tomography. Pastore E. P., github.com/emmanuel6474/biondirepro.