Research dispatch 02
Published 26 September 2026
Okoye Sunday
There is a clock in the corner of the footage.
Most people who watched the security camera recording from Gyirong Port watched the water. In the upper right of the frame there is a running date and time overlay, advancing second by second, put there by the camera.
I went back and watched the clock.
What the camera shows
Read frame by frame, the overlay gives this:
| Time (China Standard) | Frame |
| 10:59:47 | the border area still visible |
| 10:59:48 | the first clearly identifiable dark debris plume |
| 10:59:49-52 | the plume expands rapidly |
| 10:59:53 | the surge already overwhelming the built-up border area |
Six seconds, from an ordinary morning to the thing the world has seen.
The overlay also shows the date, 26 August 2026, and characters indicating Wednesday. No camera identifier, coordinate or location string is visible in the portion examined.
Against the seismic record
The United States Geological Survey records the initiating event as US7000tbwb, “M 5.2 Landslide, 55 km NW of Kodari, Nepal,” with an origin time of 02:52:10 UTC. In Beijing, 10:52:10.
Seismic origin 10:52:10 CST
First visible debris 10:59:48 CST 7 min 38 sec
Overwhelming surge 10:59:53 CST 7 min 43 sec
Seven minutes and thirty-eight seconds.
What else agrees
Kristen Cook, a geomorphologist at the Universite Grenoble Alpes, reconstructed the flow independently from seismic and video records and gave roughly seven and a half minutes.
She was within eight seconds of the measured figure, and she had never read that clock.
Su Pengcheng of the Chinese Academy of Sciences told a press conference in Tibet on 30 August that the flow struck Gyirong Port at 10:59. The camera says 10:59:48. Correct to the minute.
Guo Zhaocheng of China’s Ministry of Natural Resources put it at about seven minutes, at roughly fifty metres per second. CCTV headlined its coverage: seven minutes, twenty kilometres.
An American seismic network, a French research university, a Chinese government remote sensing centre, a Chinese state broadcaster, a Chinese Academy of Sciences researcher at a government podium, and a security camera that was simply running.
They agree.
What it means
Seven minutes is shorter than the time it takes to read this dispatch.
If someone had been told at 10:52:10 that a mass of ice had detached twenty kilometres upstream, and had believed it immediately and started running immediately, they would have had seven and a half minutes to clear a customs complex, an immigration hall, a vehicle queue and whatever hotels and shops surround a border crossing at the end of the pilgrimage season.
Nobody was told. There was no mechanism by which anyone could have been. The collapse happened in a high basin inside another country’s national park, in terrain Reuters described as relatively unmonitored. The seismic signal was analysed in the United States over the following days.
At Gyirong Port, no warning system on earth would have helped.
I want to be plain about that, because the temptation to imply otherwise is considerable and implying it would be a lie told at the expense of the people who died there.
The accountability questions in this story lie downstream, where there was more time, and in the years before, where there were meetings.
Method, and its limits
Provenance. The sequence was first seen as published by Al Jazeera on 26 August, “CCTV captures terrifying mudslide engulfing Nepal-China border.” The circulating recording was then obtained from the Wikimedia Commons archive and examined directly rather than relying on the broadcast reproduction.
The file. 69.226 seconds, 854 x 480, approximately 20 frames per second. A circulated re-encode, not the original camera file. Wikimedia Commons holds a 1,280 x 720 version.
The reading. The recording was stepped through frame by frame. The displayed seconds were read directly from successive frames, not inferred from the file’s creation time and not taken from the clock of the device used to view it.
On transcoding. Commons warns that file metadata may not represent the original recording, because the file may have been modified or transcoded. That warning is correct and it applies to file metadata: creation date, encoding timestamps, container information.
It does not apply to the burned-in overlay. The date and running clock are part of the image, rendered by the camera into the video frames. Transcode or downscale a file and the overlay travels with the picture. That is why the reading was taken from the visible display rather than from file properties.
The limits, stated plainly.
- The camera clock’s synchronisation could not be independently established. Security clocks drift. This one has not been checked against a network time standard. The timestamp is properly described as an uncorroborated internal clock reading which shows strong temporal agreement with an independently recorded seismic event. It is not a synchronised reference.
- The timestamps record when the flow became visible to that camera, in that field of view. Commons describes the camera as looking south and downstream. A fixed camera sees one direction. The flow may have reached other parts of the complex marginally earlier or later.
- The camera’s mounting coordinates are unverified and are not stated as exact.
Why it matters
Not for the drama of the number.
The seven and a half minutes divides this disaster into two categories. Above a certain point on that river, the flood outran information and nothing indicts anybody. Below it, the water was slow enough to outrun.
At Betrawati, further down, people had around forty-five minutes.
Everything that happened in that forty-five minutes is a question about systems, and it is the subject of the rest of this work.
Okoye Sunday is writing a book on the 26 August Nepal-Tibet flood. [email protected]