72 festivals filed · 72 with full entries · 9 with a sourced attendance figure
Country pass 06 · 72 entries, Q-JP-001 to Q-JP-072
Five passes in, each country has contributed one thing the Codex did not have. France schedules; Italy inherits; Spain is governed by Easter; Switzerland made dates conditional on the weather; the Emirates made them conditional on the sight of the moon. Japan contributes something none of them did: the same festival, on two different dates, both correct.
On 1 January 1873 Japan replaced the lunisolar calendar with the Gregorian one overnight. The festivals did not all follow. Some kept their old day and took the new number — Tanabata, the seventh night of the seventh month, became 7 July. Others kept the season and moved a month later to find it again: this is the tsukiokure, the month-delayed reckoning. Sendai keeps Tanabata on 6–8 August, and Sendai's is the largest Tanabata in Japan, at 2,107,000 over three days. Both are the same festival. Neither is a variant of the other.
Obon splits the same way and more sharply, because it is a national observance rather than a city's: 13–16 July in central Tokyo and parts of Kantō, 13–16 August nearly everywhere else, and in some places the old lunar date, which drifts. The August dates are when the country actually moves — the trains, the hotel prices, the closed offices.
A month column cannot say this. A month column must choose, and whichever it chooses is wrong somewhere. We have added a calendar_note field for it and it is used twice in this file, on Sendai Tanabata and on Obon. It is the smallest new field we have added and the one that was hardest to avoid.
The six timing classes have held for five countries. Japan breaks two of them, and the honest thing is to say so rather than to round the entries into the nearest class.
Hatsu-uma falls on the first Horse day of the second month. Tori no Ichi falls on the Rooster days of November — of which there are two in most years and three in some, and in a three-Rooster year the markets are held three times and the old saying has it that fires follow. These dates come from the sexagenary cycle, a sixty-day rotation of twelve animals and five elements that runs underneath the Gregorian calendar and takes no notice of it. It is not fixed, not anchored to a weekday, not lunar. We have filed both as anchored because that is the least wrong of the six, and the date_text carries the real rule.
This is the second time a country has outgrown the classifier, and the pattern is now clear enough to act on. Switzerland asked for conditional and the Emirates asked for it again; as of this pass the field is rendered, on eleven Japanese entries — the dry grass that has to burn on the Nara hillside, the blossom forecast that moves four entries by up to two weeks, the river level that can cancel the Sumida fireworks, the rain that moves Jidai Matsuri to the 23rd. The recurrence object that would replace prose date_text is still owed.
One more thing about the lunar class. Japan has exactly one lunar entry out of seventy-two: the Nagasaki Lantern Festival, which the city's Chinese community keeps to the Chinese New Year. Japan is the only large country in the atlas where the lunar calendar was abolished and survives in a single place.
Japan counts its festivals, and counts them better than most of the countries in this atlas. What it publishes is 人出 — hitode, turnout: a crowd estimate for a period and an area, issued by the organising committee or, in Kyoto, by the prefectural police. It is not a gate count and it is not a unique-visitor count, and the entries say so in the basis field rather than presenting the number bare.
The range of practice inside one country is the finding.
Hakata Dontaku is the best-documented festival in the whole Codex. The Fukuoka Citizens' Festival Promotion Association publishes a turnout table by day and by edition running back to 1986, in its own implementation plan, with the record edition marked: 2,950,000 in 1989. The 2025 festival drew 2,350,000 over two days — 1,150,000 on 3 May and 1,200,000 on 4 May. For 2026 the association projects 2,000,000, which is a forecast and lower than any full edition since 2022, and we have recorded it as a forecast.
Gion Matsuri is at the other end. No body publishes a turnout total for Gion Matsuri. What exists is a point-in-time crowd estimate issued by the Kyoto Prefectural Police for 21:30 on 16 July, one night of a month-long festival: about 380,000 in 2026, against 240,000 in 2025, 257,000 in 2024, 340,000 in 2023 and 300,000 in 2022. The series is not monotonic and we have not smoothed it. The Ato Matsuri yoiyama a week later draws a fraction — the Kyoto Shimbun reported 11,000 for one of its nights.
Hakata Gion Yamakasa publishes nothing at all, in the same city, four hundred metres from Dontaku's route, with the same promotion machinery available. We have recorded that as refused rather than estimating it. The Oiyama runs at 04:59 through streets nobody cordons for counting.
And Chichibu Yomatsuri shows the arithmetic underneath the word. Chichibu City published 257,000 for the 2024 festival — explicitly 延べ, cumulative, 30,000 on 2 December and 227,000 on 3 December, summed. A press figure of about 290,000 also circulates. The two are not reconciled anywhere and we have not reconciled them. Where two numbers exist for one festival, the basis is almost always the difference.
Nine Japanese entries carry a figure, and three of them carry the same cause.
The Aomori Nebuta Festival Executive Committee reported 1,010,000 for 2025, down 40,000, and named the Osaka Expo among the reasons. Akita's Kantō committee saw its second-lowest turnout in ten years and named rain, the weekday arrangement and the Expo. And the Awa Odori committee in Tokushima reported 1,110,000, up 8.8% on 1,020,000, and named interest raised by the Expo as a reason it rose.
Three independent organising committees, three separate prefectures, one world's fair, cited twice as a drain and once as a lift. Yosakoi in Kōchi gives the cleanest three-year series in the file for the same kind of reason: 870,000 in 2024 after a Nankai Trough earthquake advisory was issued mid-festival, 970,000 in 2025 in rain, and press reports of 1,020,000 in 2026. Each year has a stated cause and the causes are not trends.
This is why the Codex records a year on every figure and refuses to carry a figure without one. A single number for a Japanese festival is close to meaningless; the series with its causes is the information.
Sōma Nomaoi moved from late July to late May because of the heat — a cavalry festival with horses and men in armour could not keep a July date safely, and the move is now permanent. Our own record had it in July and was wrong; this pass corrects it. Sendai's 2026 Tanabata, by press accounts, lost up to four tenths of its daytime turnout to the heat. Kobe Luminarie has moved out of December to late January for different reasons, but it has moved.
Across six countries this is the first time a climate reason has shown up in the date-change record rather than a civic or religious one. Switzerland's warm winters cancel events. Japan reschedules them.
Five Japanese entries are not annual events, and two of them are filed as annual nearly everywhere: the Setouchi Triennale and the Sapporo International Art Festival both run every three years. Onbashira runs every six, in the years of the Monkey and the Tiger, and in an off year there is nothing whatever to see. The Kanda and Sannō festivals alternate: Kanda's full procession in odd years, Sannō's in even ones, a division made in the Edo period because the city could not absorb both.
A cycle is still the property most often lost when a festival is written into a month column. It has now happened in four of six countries, and it is the single most common factual error we have found in other people's calendars — ours included.
Seventy-two entries, Q-JP-001 to Q-JP-072. Thirty-one religious, fifteen seasonal, fifteen cultural, five arts, three music, two food, one unique. By date behaviour: thirty-four fixed, twenty-two seasonal windows, ten anchored, five cyclic, one lunar. Nine carry a turnout figure, one carries a passport-level origin survey — the City of Sapporo's survey of foreign independent travellers taken at the Ōdori site, in which Hong Kong at 17.9% and Taiwan at 14.7% came ahead of mainland China at 11.9% and Australia at 11.5%. That survey is from 2013 and we have said so; the city repeated it at the same site in 2023 and we could not reach the country table.
Sixty-two entries still need a figure. For Japan, unusually, most of them exist and are published somewhere in Japanese by a city office or a committee. That is the next round's work, and it is a round worth doing here more than anywhere else in the atlas.