Research dispatch 01
Published 26 September 2026
Okoye Sunday
On the morning of 26 August 2026, a debris flow travelling at fifty metres per second came down the Bhote Koshi and the Trishuli. More than fourteen hundred people are dead and more than six thousand are listed as missing.
Nepal operates a national disaster alert portal called BIPAD, maintained by the National Disaster Risk Reduction and Management Authority. It is public. Anyone can query it.
Between 1 and 3 September I did.
What the system normally does
Filtered to a six-month window, the three districts along the affected stretch of river return sixty-five alerts.
| District | Alerts, six months |
| Rasuwa | 32 |
| Nuwakot | 17 |
| Dhading | 16 |
Most are flood warnings, and they cluster in a particular way.
In Nuwakot: Bidur-7 on the nights of 5 and 6 August. Bidur-7 again on 14 August at 00:30, 01:10, 01:40 and 04:20. Bidur-7 and Tadi-3 on 20 and 21 August at 00:20, 00:40, 00:50, 01:20 and 01:30.
Nine flood warnings in the twelve nights before the disaster. Several after midnight. Twenty minutes apart.
Nothing issued by a person on duty behaves like that. That is an automated threshold trigger: a river gauge crossing a set level, firing, the level falling back, crossing again.
A machine watching water. Working all summer.
What it did on 26 August
Nuwakot returns nothing.
The portal’s response to a query for Nuwakot district on 26 August is: There are no alerts. The last preceding entry is 22 August at 02:30.
That is the district containing Bidur and Tadi. It is where, at Tribhuvan Trishuli Secondary School, a principal evacuated more than nine hundred children roughly ten minutes before the water arrived, on the strength of a telephone call from a parent who ran to the gate on foot.
Rasuwa returns one entry. A road closure at Gosaikunda-5, timestamped 11:30.
The flow crossed the border at 08:44. Rasuwagadhi was destroyed within minutes. Timure went. Syapru Besi went. The district has since returned 865 people to the national missing list.
The national alert portal logged a closed road, two hours and forty-six minutes after the water arrived.
Dhading returns five entries.
Gajuri-5 at 00:00. Then nothing until Galchi-6 at 11:20, Siddhalek-6 at 11:30, Gajuri-2 at 12:30 and Benighat Rorang-10 at 13:30. The flood passed through Nuwakot at around nine. The first substantive Dhading entry comes more than two hours later. Three of the five are road closures.
A month later
The obvious objection is that the portal was simply not working well that week.
Between 21 and 24 September I queried it across all seven of Nepal’s provinces.
| Province | Alerts |
| Lumbini | 11 |
| Madhesh | 4 |
| Gandaki | 4 |
| Bagmati | 1 |
| Koshi | 1 |
| Karnali | 0 |
| Sudurpashchim | 0 |
Lumbini received eleven alerts in four days, timestamped 00:00, 00:30, 02:54, 03:15, 08:15, 08:54, 10:00 and 14:00. The system fires, day and night.
Bagmati Province contains Rasuwa, Nuwakot and Dhading. It received one road closure, at Galachi-4 in Dhading. Rasuwa and Nuwakot generated nothing at all.
The silence is a property of the corridor, not of the portal.
What I think this means
I do not believe anyone failed to issue a warning they could have issued, and I am not alleging negligence.
The explanation available in the public record is mechanical. The system is gauge-based. Gauges sit in rivers. NBC News reported that the flow destroyed flood gauges along the river as it passed, hampering officials trying to warn people downstream.
And a threshold-crossing alert could not have helped in any case. Nepali officials recorded the Trishuli rising nine metres in thirty minutes. A gauge that fires when water reaches a set level is not an early warning for an event like that. It is a notification that the flood has arrived, sent to a district the flood has already reached.
The instrument was answering the wrong question. It was asking is the river high? when the question that morning was is something coming?
Within five days, Nepali officials had installed cameras and seismometers at Timure in Rasuwa, upstream and near the border. Instruments watching for movement rather than measuring depth. That is the correct response, and they reached it without waiting for anyone to tell them.
What this does not establish
- That nothing was sent. BIPAD is the national digital layer. Warnings also move by radio, by district administration, by police, by telephone. The principal at Tribhuvan Trishuli got phone calls. Something was moving. This work does not show what.
- That the portal timestamps by issue rather than entry. If alerts can be logged retrospectively with the time of entry, my reading of the Dhading sequence is wrong.
- That the new Timure instruments feed this portal. They may report to a separate system. I do not know.
- Whether the alert feed is complete or paginated. If the displayed counts are not totals, my figures are minima.
Right of reply
I put these findings in writing to the NDRRMA on 2 September 2026, with six questions including the three above that could overturn my reading.
As of today, twenty-four days later, I have had no reply.
I also requested Situation Reports 1 to 10 on the same date. No reply to that either.
If NDRRMA responds, I will publish the response in full and correct this dispatch where it is wrong.
Method
All queries run at bipadportal.gov.np between 1 and 3 September 2026 for the August data, and 21 to 24 September for the provincial comparison. Filtered by province, district and municipality, with the date range set explicitly to 4 March to 3 September 2026 for the six-month figures.
Figures carry dates and are not retrospectively corrected. Where a figure is later revised, the new figure is added and the original left in place.
Screenshots of every query, with the filter state visible, are available on request.
Okoye Sunday is writing a book on the 26 August Nepal-Tibet flood. [email protected]
CORRECTION, 12 OCTOBER 2026
On 11 October 2026 I obtained the underlying alert data from the BIPAD portal as a CSV export, for the period 11 April to 11 October 2026, Realtime Module, for Rasuwa, Nuwakot, Dhading and Gorkha. The export carries full timestamps with an explicit +05:45 offset, which the portal’s on-screen views do not.
It corrects six timestamps in this dispatch. It does not change which municipalities appear, and it does not change the central finding.
WHAT I PUBLISHED ON 26 SEPTEMBER
That Rasuwa returned one entry on 26 August, a road closure at Gosaikunda-5 timestamped 11:30, two hours and forty-six minutes after the water arrived.
That Dhading returned five entries: Gajuri-5 at 00:00, then Galchi-6 at 11:20, Siddhalek-6 at 11:30, Gajuri-2 at 12:30 and Benighat Rorang-10 at 13:30.
WHAT THE EXPORT SHOWS
The Rasuwa road closure at Gosaikunda-5 is timestamped 2026-08-26T11:45:04+05:45, not 11:30. Three hours and eight minutes after the collapse at 08:37, not two hours and forty-six minutes. Its description reads: “Road is closed due to Huge flood from Tibet Region has swept away entire river route road.” It carries no expiry time.
In Dhading, the export returns five entries for 26 August, the same number I published and for the same five municipalities, but at different times:
11:35:15 Flood warning at Siddhalek-6. Water level 7.78 m against a warning level of 7.0 m. No expiry recorded.
16:45:03 Road closed in Gajuri-2
16:45:03 Road closed in Gajuri-5
16:45:03 Road closed in Galchi-6
17:45:04 Road closed in Benighat Rorang-10
Every municipality I named appears in the export. None has been added and none has been removed. What has moved is the clock. The entry I recorded for Gajuri-5 at 00:00 appears in the export as a road closure at 16:45, and the three closures I recorded at 11:20, 12:30 and 13:30 appear as three closures at 16:45 and one at 17:45.
WHAT THIS CHANGES, AND WHAT IT DOES NOT
It changes the times. The first flood warning in the corridor on 26 August is timestamped 11:35:15, two hours and fifty-eight minutes after the slope failed, and the road closures came in the late afternoon rather than at midday. The gap between the disaster and the record is larger than I reported, not smaller.
It does not change the finding. The portal returned no flood alert at all for Nuwakot or for Rasuwa on 26 August. That is unchanged and is confirmed by the export.
It adds one thing I did not have in September. The Siddhalek-6 flood warning of 26 August carries no expiry, and it is the only alert that station produced in six months. The station has not reported since.
WHY THE TIMES DIFFERED
I cannot fully establish this and I am not going to guess. The portal’s card view, its table view and its CSV export display different times for the same alerts. The export is the only one that states its own timezone. I have put the question to the National Disaster Risk Reduction and Management Authority and will publish their answer.
It is also possible that records were amended between my September queries and the October export, or that I transcribed times wrongly in September. All three remain open, with one observation that bears on them. In Dhading the export returns exactly the five municipalities I published, no more and no fewer. A database that had been amended would be expected to differ in content as well as in time. A display that renders the same rows differently would be expected to differ only in time. That is what I am looking at.
WHAT I AM CHANGING IN PRACTICE
From now on this project cites the CSV export and nothing else from this portal, and states the data source and date range with every figure. The export for all four districts is available to anyone who asks.
The original text of this dispatch of 26 September is unchanged above.
Okoye Sunday