Counts

There is no total on this page, and there will never be one. Summing the bands would double-count every structure that satisfies more than one — most of them — and would re-create exactly the single global integer this project exists to reject.

Two different questions get two different columns, and conflating them is the most common error in this field:

• Held — how many records this dataset holds. A fact about our progress.
• Population — how many exist in the world, per the bounding register. A fact about castles.

And a third distinction, which matters just as much:

• Assessed — somebody applied the band’s criterion to the structure.
• Register-derived — a register asserted its own typology and we transcribed it in bulk.

They are never summed. Copying 104 rows out of Cadw is not the same act as assessing 104 structures, and a headline that added them together would be a dataset reporting how many rows it copied. Register-derived verdicts all carry confidence: low and say so on every record.

Per definition

Fortified residence

Closable per jurisdiction, because national archaeological registers classify by monument type and a “castle” class can be filtered and counted. Ireland is the proof: the SMR returns 4,552 Castle-class records, and 129 if you accept only Anglo-Norman masonry castles. The band closes one jurisdiction at a time, never globally.

Enclosure fortification

Strongly closable where a national survey exists, and the numbers are large: 31,431 ringfort-class records in the Irish SMR, 4,147 in the Atlas of Hillforts across Britain and Ireland. Closure fails wherever no national survey has been done, which is most of the world.

State fortification

Closable in principle — states keep records of their own works — but much weaker in practice than the two bands above, because military estates are documented by defence ministries rather than heritage registers and the twentieth-century material is barely surveyed anywhere.

Palatial seat

Not closable. No register enumerates “buildings called castle that are not fortified” — the category exists in language, not in law. OpenStreetMap offers the best available evidence of scale rather than a bound: of the 28,544 objects carrying castle_type, 23.7% are tagged stately, 20.5% manor and 16.4% palace, against 19.5% defensive. This band publishes that distribution, not a total.

Revival and folly

Not closable, and honestly labelled: no register recognises “folly” as a designation class. Individual examples are protected as listed buildings — Leyburn Sham Castle is Grade II listed — but they are listed as buildings, so no filter extracts them. This is also the one band whose authority is this project itself, which is stated rather than dressed up.

Popularly a castle

Not closable, ever, and this is the band’s point. Popular usage has no register and no boundary, so any single global integer for it — 2,435, for instance — is a selection presented as a census. This band exists to hold that number and label it correctly.

Held counts are derived from 4668 records by scripts/counts.py. CI regenerates this file and fails on drift, so a number here cannot disagree with the records behind it.

Contested is a result, not a gap

A verdict of contested means competent sources genuinely disagree under that band’s criterion — not that we could not be bothered. It is reported in its own column and never folded into yes or no.

Crac des Chevaliers is contested on two bands at once because the Knights Hospitaller sit between “private lord” and “state”, and those are precisely the categories the criteria depend on. The ambiguity is in the twelfth century, not in our evidence.

Where the real populations are

Held counts are small; this dataset is young. The bounding registers are not, and these are the figures that matter:

The first three rows are the same country, the same register and the same day.

The number that started this

Of the 28,544 OpenStreetMap objects carrying a castle_type, only 19.5% are tagged defensive. The rest are stately (23.7%), manor (20.5%), palace (16.4%) and fortress (10.9%).

Read one way, four in five things tagged as castles are misfiled. Read the other way, the crowd’s working definition of “castle” is overwhelmingly stately residence, and the scholars are the ones out of step with usage.

Both readings are defensible. That is the finding.

See also: the claims ledger, where circulating castle numbers are graded by how well they are sourced.