On 6 August this outlet published “Japan's worst heatstroke day on record: 4,105 people taken to hospital on 22 July”. On 10 August it published “On Japan's worst heatstroke day, 744 of the 4,105 ambulance calls came from a workplace”. Both figures are now superseded. Japan's Fire and Disaster Management Agency posted its finalised report for July 2026 at 14:11 JST on Tuesday, and its day-by-day table puts 22 July at 4,134 ambulance transports with twelve deaths, not the nine we printed; the workplace count is 755 rather than 744. The record itself survives, and widens.
Our number was low because FDMA counts the same days twice, by design. Each Tuesday it publishes a provisional table, printed in red, tallied from fire headquarters the previous day; roughly eight weeks later the finalised table for the whole month follows, and the survey notice of 22 April 2026 sets out that cycle explicitly. Summing the five weekly tables covering 1 to 31 July gives 40,679 transports; the finalised table prints 41,174, a difference of 495 people, or 1.22 per cent. Twenty-seven days were revised up and four down, with none left unchanged. Running the same subtraction back through the archive shows the direction is not a fluke: July 2022 went 26,597 to 27,209, 2023 35,724 to 36,549, 2024 41,929 to 43,195, 2025 38,608 to 39,375. Five Julys, five upward revisions, never once down, and deaths revised up in all five as well. What it does not show is an accelerating problem — 2026's plus 495 is the smallest gap of the five, in both absolute and percentage terms.
One further correction is ours alone. Our 6 August piece said 22 July beat the previous single-day high by 296 people, which compared a provisional figure against a finalised one. Summing the national daily totals across all eighteen of FDMA's annual open datasets — 113,459 prefecture-day rows over 2,446 survey days from 2008 to 2025 — the previous high is 3,809 on 18 July 2018, a day that also killed twelve. Like for like, the correct margin is 325 people, and 22 July 2026 is the first day above 4,000 in the eighteen seasons the survey has run. The revision does not change July's standing, which remains the third-highest since 2008 on either count. None of this is concealed; it is simply that the corrected number arrives eight weeks after anyone is looking, and the provisional is the one that travels.