The gap between a system knowing and a system telling.
A public, versioned dataset. One row per incident: when the hazard began, when instrumentation registered it, when a human confirmed it, when a message left the building, and the signed lead time at each affected location — with the provenance of every field.
Nothing like it exists. It is the reason the answer to "who are you to tell me this" stops being a defence and becomes a fact.
| field | what it holds |
|---|---|
incident |
name, date, country, system class |
t_onset |
when the hazard condition began |
t_detect |
when instrumentation registered it |
t_confirm |
when a human confirmed it |
t_disseminate |
when a message left the building |
lead_time[] |
per affected location, signed |
instrument_state |
what the sensors were doing at failure |
failure_mode |
missed detection · false positive · delivery · comprehension |
provenance |
source and retrieval date for every field above |
The incidents so far — rendered from archive/warning-latency.csv,
| incident | date | t_onset → t_disseminate | lead times | failure |
|---|---|---|---|---|
| Bhote Koshi flash flood, Rasuwa | 2026-08-26 | 08:37 → 09:16 | Syabrubesi: -26 to -16; Betrawati: 4 to 14 | missed |
| False ballistic-missile alert, statewide | 2018-01-13 | 08:07 → 08:45 | n/a: n/a; — | false-positive |
2 incident(s) published. Download the dataset (CC BY 4.0): CSV · JSON · changelog. Corrections are versioned and dated.