It is easy to write about early warning as a problem nobody has solved. In Odisha that is not quite true, and the exception is worth studying rather than skipping.
The October 1999 super cyclone crossed the coast and killed on a scale that reset how the state thinks about storms. The models had seen it coming. The satellites had seen it. What failed was the last step, the distance between a correct forecast and a household that acts on it.
What was built afterwards
The response was not a better model. It was institutions and concrete: a dedicated state disaster authority, a network of purpose-built cyclone shelters along the coast, trained volunteers in the villages that would need them, and an evacuation protocol with named people responsible for triggering it.
When Cyclone Phailin came ashore in October 2013 as a comparably severe storm, the state ran that protocol. Around a million people were moved inland before landfall, in what was widely reported at the time as one of the largest pre-emptive evacuations the country had carried out. The death toll was in the tens rather than the thousands.
The physics did not improve between 1999 and 2013. The distribution did.
Why we keep pointing at it
Because it settles an argument that otherwise runs forever. When the warning reaches the person who has to move, and that person has somewhere to go and a reason to believe the message, the outcome changes by orders of magnitude. The bottleneck was never the meteorology.
It also sets the bar for anything built now. A cyclone gives a coast a day or two of visible, unambiguous warning, and it arrives through a chain that has been rehearsed. A river flood in an ungauged reach gives an officer far less, through a chain that has not been. Closing that gap is a distribution problem before it is a modelling one, which is the whole reason this exists.
Field notes. The 1999 and 2013 events are drawn from the public record, and figures are given as the approximate orders of magnitude that were reported. Nothing here is a ModelEarth measurement.