Binding rules and current implementation

Cities of Europe methodology

The project separates binding policy, implemented behavior, candidate evidence, and planned work. This page describes the current public candidate, including the COR-2026-006 fail-closed withholding rule. The complete binding method remains in the repository methodology and machine contracts.

Prefer on Google

1. Current publication state

The runtime contains 695 reachable Eurostat City Statistics provider profiles. Six additional stable United Kingdom profile routes are withdrawn under COR-2026-003. ENTITY-007 accepts the exact 701 provider identities only within ESTAT:CITIES(7.0), preserving 695 reachable_noindex and six withdrawn states. It accepts no route, page, publication cohort, or indexability. The owner-directed interim cohort includes 197 provider-profile pages; its route admission is separate from source identity and the unfinished fixed preindex audit.

Provider records are not accepted permanent city identities. The default displayed heading is the current label attached to the provider code in the cities dimension. ENTITY-006 selects the KSH 2025 official endonym Győr for coe:entity:000413 while preserving Gyõr as Eurostat's separate source literal and export value. PAGE-004 projects the immutable ENTITY-008 and ENTITY-009 records into bounded linked-authority-record panels for Graz and six Hungarian city profiles. Provider, administrative-place, and named-place records remain distinct. The panels admit no page. The runtime has 522 profiles labelled city, 172 labelled greater city, and one record, MT001C, whose type is unresolved under COR-2026-007. Provider labels alone do not establish an authority-backed place name, legal place, administrative boundary, or complete national directory.

2. Facts currently rendered

OPS-008 retains the original eight accepted City Statistics snapshots. OPS-012 clears selected use of seven exact later responses, and OPS-013 makes those releases the current pointers while retaining the accepted 2023 perception release. DATA-016 applies 37 metric-specific recipes to select the newest eligible accepted numeric observation for each provider and metric. It supplies 9,847 non-comparative profile facts, withholds 323 observations under COR-2026-006, and preserves 51 status-only cells as typed missingness. It preserves every source value and infers no conversion; comparison and ranking selections remain unaccepted, and DATA-016 changes no route, page, or indexability state. Source: Eurostat, using the exact datacode links below. Accessed 1 September 2026 from the exact Eurostat datacodes. OPS-012 clears the selected use of seven later captured responses, and OPS-013 makes those releases the current pointers while retaining the accepted 2023 perception release. DATA-016 keeps exact value tokens, source-status codes, periods, and release lineage for every selected profile fact.

ENTITY-006 and OPS-014 use exact CITY-LAU 2021 and KSH 2025 rows to select Győr for one English display name. Gyõr remains Eurostat's separate source literal. PAGE-004 separately renders seven compact panels from the selected Graz and Hungarian public source slices. The 2025 KSH Jogállás 2005 fields remain off the panels as 2005-referenced source literals, not current or provider legal status. These panels create no rank and do not change the API, download, route, page admission, comparison, ranking, cohort, sitemap, or indexability state. See the source page for exact attributions, releases, access dates, and reuse links.

Every displayed numeric value keeps its exact retained source token with digit grouping only, indicator code, selected period, source-status code and meaning, project treatment, provider-grain status, release URL, access date, and publisher update date. Numeric b, d, e, and p flags remain visible. The 51 status-only cells retain field, year, source status, meaning, and typed missingness without exposing a numeric value.

CSV exports use 13 columns: provider code, current source label, provider grain, dataset ID, datacode URL, metric code, reference period, exact source value text, source-status flag, source-status meaning, project treatment, comparison status, and access date. The token column is value_text; exports contain no selected authority-name column, foundation field, rank, or country column. The exact-token path does not invoke a rounding decision: P-012 remains pending, the calculated age-65-plus share remains private, and P-009 remains open for comparisons, rankings, other source families, and future releases.

DATA-016 selects the newest eligible accepted numeric observation for each provider and metric. Selected periods vary by observation from 2022 through 2025 and appear beside every value. This mixed-vintage profile view is not an accepted comparison or ranking cohort.

3. Geography, location, and coverage

City and greater-city provider products remain separately labelled. COR-2026-007 marks MT001C's type unresolved without inferring a replacement. PAGE-004 projects seven bounded Graz and Hungarian associated-record panels. The current public projection has seven panels and does not project the remaining 688 profiles. All 695 profiles still lack a provider-geography point, map, or location pin, and zero pass PAGE-002.

Cities of Europe uses a reviewed internal reporting-country correspondence, but it establishes no provider country. The seven panel-only country-or-area attachments appear only in their bounded panels, never generic profile fields. The public bundle has no provider-level country field, and no page uses the correspondence for labels, filters, grouping, metadata, navigation, or exports.

Coverage counts describe only the retained provider bundle. They do not mean every city in Europe, national completeness, or accepted project_europe_v1 entity coverage. Stable country routes remain reachable and noindex with no profile membership or counts.

4. Source-status boundaries

Accepted source-literal sliceDATA-016 selects profile facts from exact accepted City Statistics releases under OPS-008, OPS-012, and OPS-013.
Public projectionPAGE-004 renders only the seven records named by its public allowlist. Private evidence has no public assertion or count.
Retained missingness and blocked dataStatus-only missing cells reach profiles without numeric values. Derived age-65-plus values, GISCO geography, country maps, regional statistics, PPP, EEA air quality, and ERA5-Land climate remain outside public use.

DATA-016 accepts no comparison selection, direct cohort, ordinal rank, route, page core, or indexability. No blocked source supplies a profile fact, relationship, filter, rank, export column, structured-data entity, map, or coverage claim. PAGE-004 points remain nested associated-place records and never become generic city coordinates. A missing source status flag appears as "none: no source status flag supplied" rather than being inferred. The source page distinguishes artifact-and-use acceptance from page acceptance.

5. Comparisons, rankings, and missing data

Comparison status: context_only. DATA-016 is a mixed-vintage latest-available profile selection, not an accepted comparison or ranking cohort. Rows are alphabetical, every value shows its own period, and no numeric ordering or same-period comparison is implied.

Home, the population overview, and population-band pages show latest accepted profile values in alphabetical order with each row's year. They provide no numeric sort control or ordinal position. Country pages publish no profile membership. The compare page shows field availability but withholds every DATA-016 numeric side-by-side value until a separate comparison selection is accepted. None of these views creates a best-city, affordability, safety, livability, or accepted ordinal rank.

Climate routes remain HTTP 200 and noindex, with an explicit data-unavailable state. They publish zero climate values, ranks, rows, structured data, or exports. Missing data are not replaced with country averages or prior blocked values.

6. Publication and corrections

Registered, reachable, indexable, and withdrawn are different machine states. Hard gates cover identity, geography, provenance, licence, claims, structured data, parity, and unresolved blockers. A depth score cannot offset failure. The owner-directed interim indexable cohort contains 200 routes: 197 provider-profile pages and the home, methodology and sources hubs under PAGE-005, PAGE-006 and PAGE-007. Its sitemap follows that selected cohort. This admission does not certify completion of the separate fixed 150-profile and 50-non-profile preindex audit. The three new GHSL routes remain noindex and outside the sitemap.

Material project errors follow the correction policy: withdraw affected facts or pages, keep stable routes reversible, preserve superseded evidence, and publish a correction record. COR-2026-003 removes the six affected profiles from the provider bundle without redirecting or erasing their stable routes. COR-2026-006 is fixed: version 58 withheld all 323 affected observations and passed the deployment-pinned 760-route, 196-profile, and 206-page browser checks. COR-2026-007 is confirmed and implemented only in the current candidate build; it is not fixed until deployment and post-deployment validation. COR-2026-010 is fixed: version 61 deployed the DATA-016 bundle and passed the deployment-pinned 760-route and 109-profile checks. The profile-default recipe had chosen each series' newest number before checking whether its source status allowed display, so 234 German facts on 109 profiles showed a 2022 value while an eligible 2023 value existed in the accepted newer release. DATA-016 applies the eligibility filter before choosing the period, and the superseded selection evidence is preserved unchanged. The superseded MT001C classification remains in frozen evidence while the runtime overlay fails closed. No replacement type or broader identity decision is inferred.

7. Licence and source dates

LIC-USE-001 clears the original eight exact Eurostat snapshots for their scoped facts. OPS-012 clears LIC-USE-018 only for named selected fields and public profile surfaces from seven exact later responses; neither is a blanket licensed-dataset clearance. AGENT-LIC-003 preserves the exact historical PAGE-003 use and release records for the seven panels and their labelled point links. PAGE-004 does not amend immutable contracts or those historical records: it requires exact-surface review of its selected records before a new public projection may operate. None is a blanket identity, comparison, page, or indexability claim. The accountable licence role is Cities of Europe project owner, as project publication accountability rather than legal advice. Public reuse requires affirmative evidence for the exact artifact and intended use, including commercial reuse, attribution, redistribution, modifications, and third-party rights. Under OPS-016, one agent prepares a frozen data-decision package and a distinct nonauthoring subagent must independently pass it for methodology compliance and general quality before adoption. OPS-160 separately permits site-code, template, copy, renderer and route delivery; it does not accept data, clear rights or admit pages to an indexable cohort. A PASS covers the complete unchanged diff, durable non-symlink Git-tracked evidence, every finding, and no unresolved finding. Historical review records remain commit-pinned provenance; later revalidation uses its own record and does not rewrite original authorship. Large raw source files remain outside Git and are independently rehashed against tracked manifests. If a release has multiple exact-use dispositions, its accepted snapshot identifies the linked use; the pipeline cannot choose one implicitly. Affirmative evidence must meet the bounded exact-use threshold. Concrete negative terms control; residual or theoretical uncertainty alone does not block a use whose affirmative evidence meets that threshold. Insufficient exact-use evidence stays private, and clearance does not carry to another artifact or use.

Harmonized urban centres: the bounded GHSL delivery

GHS-WUP-MTUC R2025A v1.1 supplies a separate harmonized urban-centre universe. Its eligible 2025 set contains 1,351 source identifiers, with 664 unrounded population estimates below 100,000 and 687 at least 100,000. All 1,351 have a population estimate. The 47 low-plausibility records among 1,398 in-scope records are excluded. UCname is displayed as a GHSL source label; it is not an official name, legal city or permanent project identity. The statistical label is GHSL modelled population estimate for reference year 2025, in persons. Every digit of the source POP token is retained.

Scope follows whole source countries or areas in the frozen project scope; it does not draw a Europe–Asia line. Azerbaijan, Georgia, Russia and Türkiye use whole-member treatment. Kosovo remains a separate status-neutral source component. Cyprus requires caution about territorial coverage; Ukraine estimates require caution about their war-affected source vintage; Armenia, Azerbaijan and Georgia carry regional-dispute cautions. Source-defined national components of cross-border urban centres remain separate. Small areas can have no qualifying centre under the source threshold; that is a methodological outcome.

The three HTML routes show exactly the two fixed complete lists and one random result: below 100,000, at least 100,000, and one random harmonized urban centre. The random draw uses one cryptographically secure unsigned 16-bit sample, accepts only values below 64,848, maps the sample modulo 1,351, and retries every rejected sample. Each identifier has exactly 48 accepted inputs; source labels are never deduplicated. Random responses are not cached. These routes remain noindex and outside the sitemap.

The bounded rights clearances cover the exact source labels and population tokens on these HTML surfaces, with GHSL/JRC and OpenStreetMap attribution, the method citation, change notices and no-endorsement text beside the data. There is no public machine-readable cohort, export, API, search, pagination, map, coordinate, general population ranking or additional population band. GHSL records have no inferred association with Eurostat profiles or national-place directories. The seven PAGE-004 panels keep their separate unchanged contract.

8. Planned work is not current behavior

Broader entity crosswalks, authoritative national directories, provider centroids and boundaries, broader GHSL products, FUA products, climate transfers, regional context, PPP, air quality, localization, and indexable cohorts remain planned or candidate work. PAGE-004's seven-profile public projection does not admit a page or make a planned product current behavior. Private national-authority evidence remains outside the public projection and creates no public assertion or count.

9. Binding policy

Scope

project_europe_v1 is a frozen 57-entry country-and-area list: UN M49 Europe plus adopted Western Asia extensions and separately tracked Kosovo. Council of Europe membership is context, not the definition. A complete national directory requires an accepted authoritative locality register; a lowest-level municipality is labelled municipality_proxy.

Entities and crosswalks

ENTITY-007 accepts only release-scoped provider identity within ESTAT:CITIES(7.0). Structure citation: Eurostat codelist ESTAT:CITIES(7.0) (source update 19 March 2026, 23:00 CET) and the eight accepted data-structure definitions, accessed 2 September 2026, with exact SDMX URLs on the sources page; Cities of Europe selected and normalized that structure metadata, and Eurostat is not responsible for the project treatment. PAGE-004 may present only its seven public-allowlisted foundation panels and cannot copy a foundation field into metadata, JSON-LD, generic city coordinates, APIs, exports, directories, country grouping, comparisons, or rankings. Those relationships never become same_as, provider boundaries, permanent identity, legal status, lifecycle evidence, or country-directory eligibility. We never force a match to increase coverage. A pure rename redirects the old slug; mergers and splits retain dated predecessor/successor records. A future projection requires a separate contract, exact-surface rights review, new Worker/page baseline, and applicable page review.

Comparability and transformation

GHS-WUP-MTUC R2025A v1.1 is accepted for the bounded harmonized urban-centre uses described below. The separate Eurostat profile defaults do not supply this cohort. Broader GHSL products and public uses remain gated. We never mix observed and estimated FUAs in one direct cohort. Different boundaries, dates, or methods do not measure population change. Source statistical status and project treatment remain separate. There is no global decimal rule; calculations use unrounded values. No publisher wins globally, and date alone is not acceptance. Do not infer one overall affordability value, and never reconstruct or rank a suppressed value.

Content and audit

Registered, reachable and indexable are separate states. Content depth cannot offset a hard-gate failure. The fixed preindex audit policy covers its named 150 profiles and 50 non-profile pages without sampling. The permanent publication threshold remains pending. Its automated reconciliation, deliberate content, desktop, physical-mobile, keyboard, screen-reader-runtime and touch requirements retain that exact scope. They do not automatically extend to every new noindex page. The three GHSL routes use exact source-to-render and runtime checks, an actual desktop browser, explicitly labelled viewport emulation, keyboard interaction and contrast/accessibility structure checks. The interim 197-profile and three-hub indexable cohort follows the separate owner directive; it does not claim the fixed audit is complete.

10. Current implementation, candidate evidence, and planned delivery

Current implementation. ENTITY-007 accepts 701 source-scoped provider identities; DATA-016 accepts profile defaults; ENTITY-006 accepts the Győr name chain; PAGE-004 projects seven bounded panels. Permanent city identity, country, lifecycle, coordinates, boundaries, routes, and indexability remain separate. The current provider bundle does not yet carry every required observation axis. A publisher reuse notice is not project commercial-reuse clearance. Planned work. No localized page or translation artifact is currently accepted.

The retained, blocked climate recipe records Copernicus ERA5-Land monthly-means reanalysis, a project-calculated 1991-2020 normal, and the nearest ERA5-Land grid cell to a GISCO provider label point. That assignment is a point approximation, not a centroid or boundary average, and not an accepted spatial transfer. It supplies no current public climate fact.

Language, names and LLM translation

Multilingual implementation is paused under LANG-010 until a new dated decision from the project owner. English is the only active implementation language. The pause does not accept an English page or weaken safeguards.

The adopted cohort is English (en), German (de), French (fr), Spanish (es), Italian (it) and Polish (pl). An LLM result is stored as a versioned project-generated translation with its model, prompt, inputs, glossary, hashes, generation time, separate review, and QA. An LLM may select an accepted name but may not create, translate or romanize a proper name. Back-translation or model confidence alone cannot accept a page. Accepted localized pages require complete data parity, target-language detection, self-canonicals, reciprocal hreflang, x-default to English, and a visible switcher.

An LLM may draft an original summary only from a versioned accepted fact packet. A specific recipe still needs benchmark acceptance.

11. Method safeguards

For each rendered metric, the project keeps the source token, period, source-status meaning, project treatment, provider grain, release URL, access date, and publisher update date together. A status-only cell remains missing, not zero or an estimated replacement. The project does not create a direct comparison cohort, rank, or affordability index from the profile-default selection. It calculates with unrounded values and applies no global decimal rule.

Different geographic grains, boundaries, observation dates, and methods remain separate. Observed and estimated functional urban areas cannot share a direct cohort. A later release is still a candidate until its artifact, scope, identity, geography, method, comparison, and exact-use rights checks pass. Community and commercial sources need their own coverage, freshness, stability, geography, and rights review.

For corrections, the project withdraws affected facts or pages, preserves superseded evidence, and keeps a stable route where the correction policy permits it. Within the fixed preindex audit, automated reconciliation does not replace its selected desktop, physical-mobile, keyboard, screen-reader-runtime or touch evidence. The bounded noindex GHSL checks preserve their separate scope and truthful capture provenance. The retained climate recipe is not a public climate product. Multilingual implementation remains paused under LANG-010; English is the only active implementation language.

12. Policy detail

Neither disposition is a blanket licensed-dataset clearance. LIC-USE-013 remains pending. The public bundle therefore contains no country field at provider level. The source correspondence establishes no provider country, and stable country routes remain reachable and noindex.

Accepted source-scoped identity is not permanent city identity. Cross-border groups must be labelled morphological_continuity, functional labour market, or institutional cooperation area. We never force a match to increase coverage.

We never reconstruct or rank a suppressed value. The fixed 150-profile cohort remains historical selection and audit evidence. The fixed preindex audit covers those 150 profiles and 50 non-profile pages without sampling. The interim owner-directed indexable cohort instead contains 197 provider-profile pages and three hubs. That route admission does not certify fixed-audit completion or a post-audit acceptance package. The three GHSL routes remain noindex. We do not describe all fields as official indicators.