# The registers against which a band's count can be closed. # # A register is what makes an integer honest: a finite, citable list kept by a # named body, so "complete" becomes checkable rather than hopeful. Several # entries here are subjects of study as much as sources — Wikidata, OSM and # thecastlemap appear because other castle datasets treat them as ground truth # and this one does not. # # `entries_total` is never estimated. Where a body publishes no exact figure the # field is null and the note says so; an estimate belongs in a sourced claim # with its basis, not in a field that reads as fact. version: 1 registers: - id: ie-smr name: "Sites and Monuments Record — Archaeological Survey of Ireland" operator: "National Monuments Service, Department of Housing, Local Government and Heritage" url: "https://www.archaeology.ie/collections-and-publications/open-data/" jurisdiction: IE enumerable: true entries_total: 151308 entries_matching: 4552 query: > SMROpenData FeatureServer, MONUMENT_CLASS LIKE 'Castle%' — returns 4,552: unclassified 2,466; tower house 1,321; motte 258; motte-and-bailey 175; ringwork 138; Anglo-Norman masonry castle 129; hall-house 51; ringwork-and-bailey 14. The same service returns 31,431 for MONUMENT_CLASS LIKE 'Ringfort%'. licence: "CC-BY-4.0" access: api inclusion_criterion: > Archaeological monuments recorded by the statutory survey and classified against the NMS monument-class vocabulary. The register classifies by monument type, not by whether a thing "is a castle" — which is precisely why it can answer the question three different ways. accessed: "2026-07-21" note: > The single most valuable register for this project. Ireland is simultaneously a 129-castle country, a 4,552-castle country and a 31,431-fort country, and all three figures come from the same official database on the same day. Nothing argued elsewhere on this site demonstrates the thesis as cleanly. - id: hillforts-atlas name: "Atlas of Hillforts of Britain and Ireland" operator: "University of Edinburgh, University of Oxford and University College Cork (AHRC-funded)" url: "https://hca.ed.ac.uk/research/about/research-projects/an-atlas-of-hillforts-in-britain-and-ireland" jurisdiction: WORLD enumerable: true entries_total: 4147 entries_matching: 4147 query: "Full published atlas as released in 2017; every entry is in scope for enclosure_fortification." licence: "See project terms" access: bulk-download inclusion_criterion: > Sites meeting the project's own hillfort definition across Britain and Ireland — a definition the project concedes is strained, noting that many of its entries "are not on hills and are not really forts." accessed: "2026-07-21" note: > Count discrepancy left open deliberately: 4,147 is the figure given by the project announcement and by Wikipedia, while DigVentures reports 4,174. The transposition is unresolved and this dataset carries the lower, better-attested figure rather than picking the convenient one silently. - id: gatehouse name: "Gatehouse Gazetteer" operator: "Philip Davis (independent)" url: "https://www.gatehouse-gazetteer.info/home.html" jurisdiction: WORLD enumerable: false entries_total: null entries_matching: null licence: "See site terms" access: search-only inclusion_criterion: > Medieval fortified sites in England, Wales, the Channel Isles and the Isle of Man, built or in use 1000–1600. The standard is explicitly derivative — it counts as a fortification "that which the various authors from which the information is drawn have considered to be a fortification," with rejected sites flagged rather than deleted. Excludes purely passive defences (moated houses, walls without wall walks or parapets). Includes royal and episcopal houses and hunting lodges as a deliberate control group. accessed: "2026-07-21" note: > Described by its author as around 6,000 site pages plus a further 1,000 of historical references, but that total spans every class it maintains — timber castle, masonry castle, siege work, palace, fortified manor house, tower house, pele tower, bastle, artillery fort, fortified ecclesiastical site, urban defences and more. It is therefore NOT a castle count and entries_total is left null rather than misrepresenting it as one. The control-group design — cataloguing undefended palaces alongside castles so the boundary can be examined instead of assumed — is the most methodologically sophisticated thing in this whole registry. - id: gb-nhle name: "National Heritage List for England" operator: "Historic England" url: "https://historicengland.org.uk/listing/the-list/" jurisdiction: GB-ENG enumerable: false entries_total: null entries_matching: null licence: "OGL-UK-3.0" access: bulk-download inclusion_criterion: > Designation under the Planning (Listed Buildings and Conservation Areas) Act 1990 or scheduling under the Ancient Monuments and Archaeological Areas Act 1979. Designation is a protection decision, not a typology — "castle" is not a filterable designation class. accessed: "2026-07-21" note: > Marked NOT enumerable despite being openly licensed and bulk-downloadable, because the bulk schema HAS NO TYPE FIELD: the published columns are entry number, name, grade, dates, capture scale, link and grid reference. Grade is importance (I, II*, II), not typology. A controlled vocabulary exists — the FISH monument-type thesaurus — but it is not applied in the List, so it cannot be filtered on. Live figures on 2026-07-21: 379,680 listed buildings, 19,998 scheduled monuments, 1,721 parks and gardens, 47 battlefields, 57 protected wrecks. Keyword matching on Name returns 1,190 scheduled monuments and 3,237 listed buildings, a figure that simultaneously catches "Castle Farm" and misses every unnamed motte. A live count against the ArcGIS FeatureServer returned "499 Token Required", and the site returns HTTP 403 to automated access generally. The type-aware route is the Heritage Gateway, which browses FISH classes across the List and local HERs but is not bulk-extractable. - id: gb-wls-cadw name: "Cof Cymru — National Historic Assets of Wales" operator: "Cadw (Welsh Government)" url: "https://cadw.gov.wales/advice-support/cof-cymru" jurisdiction: GB-WLS enumerable: true entries_total: 4237 entries_matching: 104 query: > DataMapWales WFS, Cadw_SAM layer: SiteType = 'Castle' → 104, against 4,237 scheduled monuments in total. BroadClass = 'Defence' → 1,019. Adjacent classes counted separately: Motte 156, Motte and Bailey 72, Hillfort 287, Moated Site 69. Welsh listed buildings (30,116) carry BroadClass but no SiteType. licence: "OGL-UK-3.0" access: api inclusion_criterion: > Scheduling under s.3 and listing under s.76 of the Historic Environment (Wales) Act 2023 — "of national importance" and "special architectural or historic interest" respectively. Uniquely among British registers, Cadw then classifies scheduled monuments by an explicit SiteType typology, which is why "castle" is a real filter here and a keyword guess everywhere else. accessed: "2026-07-21" note: > The cleanest type-classified castle register in Britain, and the one that shows most starkly how much the filter does: Wales is a 104-castle country, a 332-castle country if mottes count, and a 1,019-fortification country, all in the same layer on the same day. GOTCHA for reproduction: SiteType values are title-case and a case-sensitive filter silently returns zero rather than erroring. Note also that the Historic Environment (Wales) Act 2016 is REPEALED — the operative statute is the 2023 Act, in force 4 November 2024, and secondary sources citing the 2016 Act are out of date. - id: gb-wls-coflein name: "Coflein — National Monuments Record of Wales" operator: "Royal Commission on the Ancient and Historical Monuments of Wales" url: "https://www.coflein.gov.uk/" jurisdiction: GB-WLS enumerable: true entries_total: 112894 entries_matching: 158 query: > site_type = 'CASTLE' → 158; a looser ILIKE '%castle%' match → 180. site_type is a semicolon-delimited multi-value field, so a record may read 'MOTTE; CASTLE' and belong to both classes. licence: "OGL-UK-3.0" access: api inclusion_criterion: > The national record of the historic environment of Wales — every known site, whether designated or not. A RECORD, not a designation register, which is why its total dwarfs Cadw's and its castle count differs. accessed: "2026-07-21" note: > Carried alongside Cadw deliberately, to make a distinction most castle datasets collapse: a record of what EXISTS and a register of what is PROTECTED are different questions with different answers. Coflein holds 112,894 sites against Cadw's 4,237 designations, and returns 158 castles against Cadw's 104. Neither is wrong. - id: gb-nir-nismr name: "Northern Ireland Sites and Monuments Record" operator: "Department for Communities" url: "https://apps.communities-ni.gov.uk/Buildings/buildMain.aspx" jurisdiction: GB-NIR enumerable: true entries_total: 18506 entries_matching: 4 query: > General_Type = 'CASTLE' → 4. Castles are classified elsewhere: FORTIFICATION 353, MOTTE 48. Of 61 records mentioning "tower-house", 55 are classified FORTIFICATION and exactly one record in the whole database carries General_Type = 'TOWER HOUSE'. licence: "OGL" access: api inclusion_criterion: "Archaeological sites and monuments recorded by the statutory survey." accessed: "2026-07-21" note: > entries_matching is recorded as 4 because that is genuinely what the register returns, and it is the single most useful cautionary number in this file. Northern Ireland plainly has more than four castles. Any dataset built by filtering national registers on a castle-type field would report 4 here and never notice — which is a good description of how castle datasets actually go wrong. The official page still describes the record as "around 17,000" entries, itself out of date against the live 18,506. - id: es-bic name: "Bienes de Interés Cultural" operator: "Ministerio de Cultura and the autonomous communities" url: "https://www.cultura.gob.es/" jurisdiction: ES enumerable: false entries_total: 18537 entries_matching: null licence: "No national open licence" access: search-only inclusion_criterion: > Declaration as a Bien de Interés Cultural under Ley 16/1985 — EXCEPT for castles, which hold the status automatically. The Decreto de 22 de abril de 1949 protected all Spanish castles "whatever their state of ruin" with no individual proceeding, and Ley 16/1985's second additional provision converted that blanket protection into BIC status wholesale. accessed: "2026-07-21" note: > The most interesting NOT-enumerable entry here: Spain is uncountable BY LAW, not by neglect. Because castles are protected by operation of law rather than by inscription, the protected class is open-ended and no enumerated list is derivable even in principle — anything later determined to be a castle was already protected in 1949. entries_total is the count of BIC RECORDS, which by construction cannot include goods protected without one. A name search for "castillo" returns 2,017; that figure is contaminated (it catches churches named for a castle) and incomplete (it misses every alcazaba, torre, fortaleza, castell, gaztelu and castelo), so it is neither an upper nor a lower bound and is not used anywhere on this site. - id: gb-sct-trove name: "Trove (formerly Canmore)" operator: "Historic Environment Scotland" url: "https://www.trove.scot/" jurisdiction: GB-SCT enumerable: false entries_total: null entries_matching: null licence: "OGL-UK-3.0" access: search-only inclusion_criterion: > The national record of the historic environment of Scotland — archaeology, buildings, industry and maritime heritage — plus the statutory designation layers (listing, scheduling, gardens and designed landscapes, battlefields). accessed: "2026-07-21" note: > Castle is not a category. The scheduled-monument layer's CLASS field has exactly one distinct value across every record — "Cultural", the INSPIRE protected-site code — so only keyword matching is possible, returning 624 scheduled monuments and 2,214 listed buildings. Designation layers are countable (67,480 listed-building rows, 8,072 scheduled monuments, 371 gardens and designed landscapes, 40 battlefields) but the listed-building figure is entity-level: a listed terrace counts many times, and HES's own prose says "around 47,000", so the distinct-designation count is unverified. Also recorded for its own sake: Canmore, Scran and ScotlandsPlaces were retired on 24 June 2025 and canmore.org.uk now issues a 301 to trove.scot, which is WAF-blocked to automated access. A national register changing platform is a standing hazard for any dataset citing its URLs — every dataset built on Canmore permalinks inherited a migration it did not choose, which is an argument for recording register identifiers alongside URLs rather than instead of them. - id: fr-pop name: "Plateforme Ouverte du Patrimoine (Base Mérimée)" operator: "Ministère de la Culture" url: "https://www.pop.culture.gouv.fr/" jurisdiction: FR enumerable: true entries_total: 45223 entries_matching: 6286 query: > Bulk extract of protected immeubles, denomination field. Prefix "château" excluding "château d'eau" (a WATER TOWER) → 6,286. Neighbouring rules: token exactly "château" → 6,159; token in {château, château fort} → 6,285; "château" as sole denomination → 5,703. The field is semicolon-delimited free text with 1,998 distinct values, unnormalised for case (château 5,703 vs Château 1) and hyphenation (château fort 104 vs château-fort 1). licence: "Etalab-2.0" access: bulk-download inclusion_criterion: > Classement or inscription as a monument historique. A protection decision across all heritage types — "château" is not the unit of designation. accessed: "2026-07-21" note: > entries_total is the Ministry's own verified figure for protected immeubles as of 31 December 2024 (14,333 classés + 30,890 inscrits). It counts monuments of every kind, overwhelmingly dwellings and churches, with military architecture roughly 6%. It is recorded here mainly because of its uncomfortable proximity to the folkloric "France has 45,000 châteaux" — see the count claim on the France coverage cell. - id: wikidata name: "Wikidata" operator: "Wikimedia Foundation" url: "https://query.wikidata.org/" jurisdiction: WORLD enumerable: true entries_total: 36716 entries_matching: 36716 query: > SELECT (COUNT(DISTINCT ?i) AS ?c) WHERE { ?i wdt:P31/wdt:P279* wd:Q23413 } — 36,716 including subclasses; 24,886 for direct instances of Q23413 only; 33,265 carry coordinates; 23,012 carry a heritage designation (P1435); 6,594 carry both coordinates and an English Wikipedia article. For contrast, the same walk against Q57821 (fortification) returns 92,798. By country: Germany 9,327, Italy 4,262, France 3,822, UK 3,816, Spain 3,160, Japan 2,266, Iran 950, Ireland 669. licence: "CC0-1.0" access: api inclusion_criterion: > Whatever editors asserted via P31. No authority and no consistency guarantee — but fully transparent and re-runnable, which most national registers are not. accessed: "2026-07-21" note: > The subclass walk more than doubles the count, from 24,886 to 36,716. The depth of P279* traversal is therefore a definitional choice, and any dataset built on Wikidata has made it whether or not it says so. - id: osm name: "OpenStreetMap" operator: "OpenStreetMap Foundation" url: "https://taginfo.openstreetmap.org/" jurisdiction: WORLD enumerable: true entries_total: 54182 entries_matching: 28544 query: > taginfo: historic=castle → 54,182 objects (14,638 nodes / 35,203 ways). castle_type is present on 28,544 of them across 223 distinct values: stately 6,775 (23.7%), manor 5,839 (20.5%), defensive 5,560 (19.5%), palace 4,692 (16.4%), fortress 3,105 (10.9%), castle 517, tower 465, towerhouse 387, shiro 159. licence: "ODbL-1.0" access: bulk-download inclusion_criterion: "None. Any mapper's judgement." accessed: "2026-07-21" note: > The best single piece of evidence on this site that the bands are real. Only 19.5% of typed OSM castles are defensive — so under the strict definition roughly four in five things tagged historic=castle are misfiled, or, read the other way, the crowd's working definition of "castle" is overwhelmingly stately residence rather than fortress. Note also that 223 distinct castle_type values are in use with no enforced enumeration, that castle_type=castle is a tautology typed 517 times, and that about 47% of historic=castle objects carry no castle_type at all. - id: thecastlemap name: "The Castle Map" operator: "thecastlemap.com" url: "https://thecastlemap.com/" jurisdiction: WORLD enumerable: true entries_total: 2435 entries_matching: 2435 query: > The site's own published total across 129 countries, categorised as castles 937, fortresses 389, palaces and châteaux 692, ruins 417. licence: "CC0-1.0" access: bulk-download inclusion_criterion: > In the site's own words, "the 2,435 most significant castles, fortresses and palaces — each verified to have a real photo, an English Wikipedia article and exact coordinates — balanced across 129 countries so the map stays explorable." Derived from Wikidata, with a fame score from Wikidata sitelink counts and English-Wikipedia pageviews. accessed: "2026-07-21" note: > Present as a subject of study: it is the dataset that prompted this project. Recorded honestly, it is a careful, openly-licensed, well-built curation of the popular_castle band — and its stated selection rule ("balanced across 129 countries so the map stays explorable") is a curatorial one, openly declared on its own homepage. The criticism this project makes is narrow and specific: a curated selection published under the title "The World's 2,400 Castles" reads as a census. See the count claim recording that its own stated filter defines 6,594 Wikidata items while it publishes 2,435 of them.