← Back to the dataMethod

How this was built

The source, the translation, the year system and what the data leaves out.

The source

All figures come from one page: the Tokyo Metropolitan Police Department's annual lost-property statistics, 遺失物取扱状況, last updated March 2, 2026. The tables there are HTML, not a data file, so the build script parses them directly and fails loudly if the expected tables are not present.

The prefecture directory comes from the National Police Agency's index of prefectural lost-property searches. All 47 are listed; 43 of them point at one shared national database and four run their own site.

The year

Japanese government statistics use imperial era years. This data covers 令和7年, Reiwa 7, which is the calendar year 2025. The comparison column throughout is 令和6年, Reiwa 6, or 2024. Reiwa 1 was 2019.

Translation

Category and disposition names were translated by hand rather than machine, and the Japanese original is shown on the site next to every category so the mapping can be checked. The full table:

JapaneseUsed here
証明書類Certificates and documents
有価証券類Securities and vouchers
衣類履物類Clothing and footwear
電気製品類Electrical goods
財布類Wallets and purses
かさ類Umbrellas
かばん類Bags
携帯電話類Mobile phones
貴金属類Jewellery and precious metals
カメラ眼鏡類Cameras and eyeglasses
遺失者返還Returned to the owner
拾得者引渡Given to the finder
都帰属Became property of Tokyo
廃棄・任意提出Discarded or surrendered

Two different counts that must not be added together

The police publish two separate series and this site keeps them apart:

These are counts of different events and there is no key joining them. The "found versus reported lost" view puts them side by side to show an asymmetry; it is not a matching rate, and you cannot subtract one from the other to get a number of unreturned items. More cash is reported lost than is handed in, which is what you would expect and is not evidence of anything.

Return rates are consistent; the totals are trickier

The disposition table is internally consistent: for each category, the four outcomes sum to that category's total. Return rates are computed straight from it. But that table's grand total, 4,750,389 items, differs slightly from the headline item count of 4,850,775, because the disposition table covers items processed in the year and the headline covers items received. Items received in late December are processed the following year. Do not mix the two.

What is missing

Rebuilding it

build_japan.py in this repository fetches both pages and writes data/data.json. It exits non-zero if the Tokyo page's table structure changes. The police update the page once a year, in early March.

Built with AI assistance. Figures are parsed from the police tables rather than retyped, but the source page is the authority.

← Back to the data