Methodology

How a place enters the Codex

Every page here was assembled from open records: by software, and by language models reading those records, and then held against a gate before it could be published. This page sets out how that works, what the gate genuinely checks, and the harder half: what it does not.

The figures below are counted from the corpus each time the site is built. Last built 20 September 2026.

The tall gneiss pillars of Calanais standing against a wide Hebridean sky
Calanais Standing Stones, Outer Hebrides · Photo: Unknown / Wikimedia Commons / CC BY-SA 2.5
4,661
records held in the corpus
4,619
of them published as pages
21
shown as an area, never a pin
0
verified by standing at them

I · Inclusion

What earns a place

A place is included when it helps answer a real question: whether this ancient place is worth the journey, what can be seen standing there now, how it should be approached, and what is honestly unknown about it. A record that only proves a site exists is not a page. It stays in the working database.

Scope is a written judgement, not a filter on a map. Each record carries a date band: prehistoric, ancient, Roman, early medieval, medieval, early modern, or a deliberate exception for something later that sits over a much older layer, and a sentence saying why the place belongs here. Records judged outside that scope keep their data and never build a page: 37 are held back that way today.

A note on ranking

No public scores

Every record carries internal weights that decide listing order, card size and featured placement. None of them is shown. Importance is meant to be felt through where a place appears and how much has been written about it, not read off a badge: a number invites argument about the number rather than attention to the place.

II · Assembly

How a record is assembled

The same six stages run for every place, from a flagship monument to a single stone in a field. What changes between them is depth, not process.

  1. Anchor

    A candidate is matched to a Wikidata entity before anything is written about it, so that identity is fixed first and two nearby stones with the same name cannot merge into one page. 4,627 of 4,661 records carry that anchor.

  2. Gather

    Open sources are pulled around that anchor and stored as labelled references, each tagged with a source family and with what it supports: identity, date, access, location, story or media. Wikipedia stands behind 4,353 records, official heritage bodies behind 1,155, statutory registers behind 2,774, and published books behind 1,020.

  3. Write

    Sections are drafted from that source pack: the first impression, the story, a timeline, field notes, and how to plan a visit. Folklore is held apart from history and framed as what is said rather than what is known, and each claim carries its own confidence and a contested flag.

  4. Illustrate

    Freely licensed photography is matched to the place, with the creator, the licence and the source page carried alongside the image rather than bolted on afterwards.

  5. Gate

    The finished record is evaluated against the publication gate described below. The gate returns a list of reasons, not a score.

  6. Publish

    Publication state and search state come out of that evaluation. What an editor asked for is an input to the decision, not the decision itself.

The Brandsbutt Stone, a reassembled Pictish symbol stone incised with a serpent and Z-rod
Brandsbutt Stone, Aberdeenshire · Photo: Otter / Wikimedia Commons / CC BY-SA 3.0

Say it plainly

The backbone is Wikipedia and Wikidata

It would read better to lead with Canmore and the national registers. The corpus does not. Wikipedia and Wikidata stand behind the overwhelming majority of these records; direct citations of Historic Environment Scotland, Canmore, Trove, Historic England, Cadw and Coflein number in the dozens, not the thousands.

Those registers are the better source, and moving records onto them is the ongoing work. Until that has happened for a given place, its page rests on an encyclopedia, which is a reasonable starting point and a poor finishing one.

III · Verification

What “checked as far as we can” means

The notice at the top of every page is a precise statement, not a modest one. This is what sits on each side of it.

Checked

What the gate enforces

  • The record is in scope, and is a place: not a route or a collection.
  • Two or more source families are the standard a full or flagship page is written to. It is an editorial standard rather than a blocking rule, and the panel opposite says how far the corpus currently falls short of it.
  • Every declared source family must actually appear in the provenance list. A claimed official source with nothing behind it fails the record.
  • Every image carries a creator, a licence, a source URL and an attribution line. A single missing one blocks the whole page, not merely the picture.
  • Nothing in the copy encourages trespass, cave entry, climbing fragile ruins, damaging veteran trees or metal detecting.
  • Folklore, ritual meaning, astronomical alignment and Druidic claims are not stated as fact, and a contested claim is not presented as settled.
  • The page does not imply ordinary open access where access is limited, private or unconfirmed.
  • No sensitive record publishes an exact coordinate, an exact map pin, exact coordinates in its structured data, or independent route guidance.

Not checked

What nothing here verifies

  • No one has been. Field verification is recorded per record and the count is zero. Nothing on this site has been confirmed by standing at it.
  • Access is unconfirmed at scale: 1,820 records have an unknown access status, and only 251 carry high confidence about how a place is reached.
  • The gate reads structure. It can see that two source families are present; it cannot see whether either is right: nor whether they are genuinely independent. Of 3,799 full and flagship pages, 35 reach two families only as Wikipedia plus Wikidata, which are sister projects rather than separate witnesses.
  • Where the underlying open record is wrong, the page inherits the error faithfully and confidently.
  • The two-family standard is not enforced. 79 records rest on a single source family, and 0 of those are published at full or flagship depth rather than held below it. We would rather say so here than let the standard read as a guarantee.
  • Ownership, opening arrangements, paths and parking change. Nothing here is live, and official guidance always outranks it.
  • Photographs are matched to places automatically. Most are right; a wrong one is a real and known failure mode.
The rocky summit of Dunadd rising out of the flat green expanse of the Moine Mhor
Dunadd Fort, Argyll and Bute · Photo: Otter / Wikimedia Commons / CC BY-SA 3.0

The second gate

Why almost nothing is in search

Beyond the safety gate there is a harsher editorial one that reads the prose itself: length against the page’s tier, banned phrasing, voice, how many concrete facts a page actually contains, whether claims are hedged properly, and whether passages repeat across pages. It is deliberately set above where the corpus currently stands.

At the last full run, on 1 August 2026, it passed 13 pages outright, sent 2,275 to review and rejected 2,322. That is the reason 4,659 of 4,661 records are marked noindex and the corpus sits outside search while it is worked through region by region: not a wait for traffic, a wait for the writing to be good enough.

IV · The gate

The gate, and what it decides

Four fields travel with every record. The gate reads them together and returns a publication state and a search state.

Scope status
In scope, or out of scope. Out of scope ends the matter: the record keeps its data and builds no page.
Requested public state
Public page, database only, draft review, or do not publish. This is what was asked for. A request for a public page that trips any hard reason falls to draft review instead.
Sensitivity
Public, approximate, restricted, or do not publish. Anything other than public forces human review and drops the page out of search automatically.
Search state
Indexable, noindex, or not public. A stub is never indexable. Nor is anything awaiting review. The same field drives the page’s robots tag and its membership of the sitemap, so the two cannot disagree.

The gate distinguishes two kinds of finding. A hard reason: low source confidence, a missing image licence, an exact pin on a sensitive place, copy that encourages trespass: blocks publication at the requested state outright. Source-family breadth is not among them: it is an editorial standard we write to, and the gate does not currently test it. A review flag: an unconfirmed access status, a non-public sensitivity, a flagship page, a main listing with no photograph: does not block the page but does hold it out of search until a person clears it.

Neither outcome is silent. The gate returns the reasons in a list, and that list is what a reviewer works through.

V · Sensitivity

Why some places have no pin

Some places are fragile, sacred, on private ground, or vulnerable to the sort of attention a precise coordinate invites. Four tiers govern what leaves the build.

  • 4,640

    Public

    Exact coordinate, exact pin, ordinary visit guidance.

  • 21

    Approximate

    The coordinate is rounded to one decimal place: roughly eleven kilometres: before it is written anywhere, and the map draws a five-kilometre area rather than a point. No parking note, no start point, no route.

  • 0

    Restricted

    The place has a page but never appears on the map at all, and is given no independent visit guidance. The route is the official custodian’s, not ours.

  • 0

    Do not publish

    Held in the database. No page, no pin, no sitemap entry.

A drystone Neolithic long cairn standing alone on open Caithness moorland
Grey Cairns of Camster, Highland · Photo: David Shand / Wikimedia Commons / CC BY 2.0

Where the rule lives

Coarsened in the build, not the browser

The rounding happens in the data layer while the site is being built. An exact coordinate for a non-public place is never written into the page, its structured data or the sitemap, so a later mistake in browser code cannot leak what was never sent. The content schema refuses outright to validate a non-public record configured to display an exact location.

Where an authoritative source has deliberately obscured a location, it is not reconstructed from other sources.

An honest caveat

Not all 21 of those flags are well judged

The approximate tier was applied cautiously to an early Scottish batch, and it caught things it should not have: ticketed abbeys and managed castles among them. A property with a car park and an admission desk does not need its location withheld.

The gate now flags that mismatch: a sensitive record that is also a ticketed visitor attraction, or one for which official access links are published, and raises it for review rather than quietly correcting it. The flags stay conservative until a person clears them. Over-caution costs a visitor a map pin; under-caution costs a place.

VI · Rights

Photographs and rights

4,617 of 4,661 records carry at least one photograph, 18,675 images in all. 18,673 of them come from Wikimedia Commons; the rest are a handful from other freely-licensed collections. The creator, the licence and the source page travel with each image, render beneath it, and are repeated in the page’s structured data.

Nothing here is a generated photograph. No image of a real place is invented, extended or retouched into something that was not photographed. Where no rights-clean picture exists, the page shows a drawn emblem for its type and says nothing further.

Reuse

The licence shown is the whole of it

Reuse is governed by the licence printed beneath each image, not by the Codex. The credit and licence on the page are the terms, and no broader right can be granted here: a photographer’s conditions are not ours to relax. A fuller account of every source and its licence is being prepared as its own page.

Text is a different matter. Where prose here draws on an openly-licensed source, that source is listed on the place’s page. It is not a copy of it: a page judged to be a rewrite of a single source is held back by the gate rather than published.

VII · Corrections

Reporting an error

Corrections are the fastest way this gets better, and the reason the notice at the top of every page exists.

Say which place, and what looks wrong. A link to a source that contradicts the page is worth more than anything else: it turns a disagreement into an edit. Errors of access, ownership and route matter as much as errors of date.

If a page shows something that puts a place at risk: a precise location that should not be public, a route across land that is not open, a photograph that should not be reused: that is treated as urgent, and the page can be pulled back to database-only while it is looked at.

Report a correction Raise a sensitive-site concern

There is no public edit button, and no plan for one that bypasses the gate: a change that has not been through the same checks is not an improvement. Shorter answers to most of this live in the questions; what the Codex is for is set out in about.