The lake sits at about forty-seven hundred metres behind a moraine dam, and the monitoring installation on it is good. Water level, seismic, a camera, telemetry over satellite because there is no other signal up there, solar with battery sized for the winter. It has run through four monsoons. When the level moved sharply in 2022 the alert reached the district office in a little over ninety seconds.
The siren is mounted on a pole beside the district office. The nearest houses in the valley below are two kilometres away and about four hundred metres lower, around a bend, beside a river that is loud.
I have worked on mountain hydrology in Nepal for a long time and the engineering here is plainly not the weak part. Detection was the hard problem, it was solved by people who knew what they were doing, and it works. That deserves saying because the standard version of this essay implies the technical people failed, and they did not; the instrument does exactly what it was specified to do.
The specification ended at the district office.
What happens between an alert arriving on a screen and a family leaving a house is a chain of things that are not instruments: somebody must be at the screen at four in the morning; that person must have authority to act without calling anyone; the message must reach households by a means they actually monitor, which is now a phone rather than a siren and was never a siren for the houses around the bend; and the households must know where to go, which requires that a route was agreed and walked, in daylight, more than once, with the people who would be using it in the dark.
Every link in that chain is somebody's job in principle and nobody's budget line in practice. The project that funded the installation was a monitoring project. It bought sensors, telemetry and a siren, and it closed successfully, and the completion report is accurate.
The counter I take most seriously is that last mile has become a phrase people use to avoid the engineering — a way of saying the hard technical part is done when it usually is not, and of shifting a funder's attention to workshops and awareness activities that are cheap to run and impossible to evaluate. I have sat through those workshops. Some of them are theatre. If the choice were between another sensor and another awareness campaign I would take the sensor, and anyone quoting me should quote that sentence too.
But this is not that choice, because the two kilometres is not a workshop problem. It is a siting problem, and siting is engineering.
The pole is beside the district office because the district office has power, a compound, and a person to sign for it. Those are procurement facts. Nobody stood at the houses and asked whether the sound arrives, and the answer, which I have now measured on four occasions with the river at different levels, is that it does not reliably arrive at the two settlements that would need it soonest.
Three repeaters would fix it. They are cheap. They need power, which is the actual obstacle, and solar with a battery is affordable and is what the lake installation already uses at forty-seven hundred metres in February.
The 2022 alert turned out to be a false positive from a sensor mount that had shifted, and the district office made the correct decision, which was to wait for the camera. Nobody in the valley heard anything, which on that occasion was the right outcome, and is the reason nobody has yet learned that they would not have.
