Change Ringing Corpus · The week, the year, and the days the country rang
Four years of the national record, day by day. One layer is a genuine rhythm — Sunday service ringing, the Saturday peal day. The other is not periodic at all: a small number of days on which the whole country rang at once, and which turn out to dominate the calendar so completely that they had already produced a false finding in this repository.
One · The week
Two · The year
One square per day, shaded by how many performances were rung. The weekly stripe is the Sunday column. What is not subtle is the handful of near-black squares — days between four and fifty-five times their own weekday's normal volume. Click any square.
Three · The days the country rang
Found by rule, not by recollection. Each day is compared with the median of the same weekday within six weeks either side — same weekday because Sunday is two and a half times the quietest weekday, so a plain rolling mean would flag every Sunday; median because the thing being looked for would otherwise inflate its own baseline. A day qualifies at .
Nothing on this list was typed in. The reason in the right-hand column is the most repeated footnote the bands themselves wrote that day — a national occasion produces hundreds of ringers independently writing near-identical text, which is a signal no other kind of dataset offers.
Four · The three grammars
This is the part a calendar cannot show. Ringing has distinct registers for distinct occasions, and on the twenty-four days they separate almost perfectly — not as a matter of interpretation, but on two numbers per day.
Five · A number in the method field
Nothing in four corpora records a person's age. There is no age column in Dove's Guide, none in BellBoard's schema, none in the Methods Library. And yet the age of the person who has died is sitting in the data, in the field reserved for the name of the method, because a tolling bell is counted — one stroke per year of life.
Six · The most reliable number in the corpus
The second Sunday in November. Not a volume spike — it barely registers on the calendar — but the proportion of bands ringing half-muffled is the steadiest figure this project has found in any of its four corpora.
Seven · The correction
“September is the busiest ringing month (12,067 performances)
and nobody knows why.” — docs/IDEAS.md, and repeated in the README.
The count is right. The finding is not.
Remove those days and the seasonal shape changes character entirely. Grey is every performance; bronze is what is left once the national days are set aside.