7. Gazetteer Workbench¶

Coming soon — preview feature

The Gazetteer Workbench is not yet available to WHG users. It is currently in staff testing behind a “Reconciliation beta” menu tab and is documented here in advance so that the design can be reviewed and refined before release. Screens, labels, and behaviour described below may still change. When the feature is released this notice will be removed.

7.1. What it is¶

The Gazetteer Workbench is a browser-based tool for taking a table of places — a spreadsheet, CSV, TSV, or JSON export — all the way from a messy first draft through to reconciled, standardised data, without uploading anything to WHG first. You import your file, tidy and interpret its columns, match its place names against WHG’s gazetteers, review the matches, and export an augmented copy — all in a single page.

It reaches WHG’s servers for one thing only: to look up candidate matches for your place names. Everything else — your file, your column choices, your review decisions — stays on your own computer.

7.2. Why a new tool¶

WHG already has a well-established reconciliation and accessioning workflow (see Reconciliation & Accessioning and the Workbench Pathways guide). That flow is built around publishing: you upload a dataset into your WHG workspace, run a server-side reconciliation task, and review the results there. It works well once your data is clean and you intend to contribute it.

The Gazetteer Workbench addresses the stage before that, and some cases that sit outside it:

Local-first and private. Your data never leaves your browser except as anonymous name-lookups. That suits work that is unpublished, sensitive, still in progress, or simply not (yet) destined for WHG at all.
Messy real-world data. Historical sources rarely arrive in tidy WGS84 decimal degrees and ISO dates. The Workbench recognises a wide range of coordinate formats and historical/calendrical date styles and converts them for you, up front.
Instant feedback. Because it runs in the page, you see the effect of every column choice, filter, and match immediately, with no task queue to wait on.
Small jobs as well as large. It is designed to be pleasant for a single place as much as for a bulk table.

Think of it as a workbench for preparing a gazetteer, complementary to — not a replacement for — the existing publication and accessioning pipeline.

7.3. Getting started¶

Open the tool from the Reconciliation beta tab in the main menu (staff only, for now). The page is organised as a set of numbered, collapsing panels; you work down them in order, and each one summarises its state in its header once done.

New to it? Take a tour (top of the page) loads the sample dataset and walks you through the whole flow — import, column roles, reconciliation, review, map, and export — highlighting each stage as it goes.

Your work is saved automatically in the browser as you go, so you can close the tab and come back to it later. Two controls make this explicit:

Save writes a .whgproj backup file — your whole project, including matches and review decisions — to your computer. Restore it later by dropping it back onto the same import area.
Clear my data removes the project from the browser entirely.

Note

“Save” produces a local backup file; nothing is sent to or stored on WHG’s servers. Because the data lives in this browser, on this computer, clearing your browser storage or switching machines will lose an un-saved project — take a .whgproj backup for anything you care about.

7.4. 1 · Import a dataset¶

Drag a file onto the import area, or click to choose one. CSV, TSV, and JSON are accepted (including JSON in WHG’s {id, fields} shape). A .whgproj backup dropped here is recognised and restored in full rather than treated as new data.

To try the tool without your own data, use Load a sample dataset — a small demo that exercises coordinate and date conversion and multi-column (County → Parish → Place) reconciliation.

Once a file loads, the panel collapses to a summary and the tool moves you on to confirming the columns.

7.5. 2 · Confirm column roles¶

The Workbench shows a preview of your table and its best guess at what each column is for. You assign each column a role from a dropdown:

Place name — the place name to reconcile (required);
Coordinates / grid ref, or Latitude and Longitude — location;
Date / year — a column of dates to interpret;
Country / ccode — a country hint used to narrow matches;
↳ Contains “…” — marks this column as a spatial container of another column, building the containment hierarchy (see below);
Feature type, Identifier — carried through and used as hints;
Other (ignore) — columns carried through untouched but not processed.

Ignored columns are hidden in the preview by default; a toggle shows or hides them.

7.5.1. The spatial hierarchy¶

If your table has administrative columns (a county, parish, region, province…), you can tell the Workbench how they nest, and it will use that nesting to disambiguate place names during reconciliation. You express the hierarchy directly in each column’s role dropdown: a container column’s role reads “↳ Contains ‹child column›”. So for a County, Parish, Place table you would have:

County → Contains “Parish”
Parish → Contains “Place”
Place → Place name

From these links the Workbench derives the containment chain County → Parish → Place — coarsest first, place name last. There is no separate ordering control: to re-order or re-nest the hierarchy, just change a column’s “Contains” choice (for example, point Parish → Contains “County” to swap the two levels), and the chain updates. The number of levels is unlimited — a Country → Region → District → Settlement table works the same way.

The Workbench guesses a sensible default chain on import (recognising common administrative column names and linking them coarse-to-fine down to the place-name column); adjust any that are wrong.

7.5.2. Coordinates¶

If you assign a coordinate role, the Workbench detects the format automatically and converts it to standard WGS84 decimal degrees. It recognises decimal latitude/ longitude (in either order, with a swap toggle when two columns are used), degrees-minutes-seconds, well-known text (WKT), OS National Grid references (e.g. SK690965), Irish Grid, and UTM. Where the format is ambiguous you can override the detected choice. A Validate all rows check reports how many values convert cleanly and lists any that do not.

7.5.3. Dates¶

If you assign a date role, the Workbench parses messy historical dates into ISO start/end values. It handles, among others:

day/month/year in UK order (dd/mm/yyyy) by default, with automatic detection when a value could only be month-first;
month names, ordinals, centuries, and bare years;
BCE/CE (including a leading minus), approximate dates (c., circa, ?), and open-ended dates (before, after, from…);
ranges written in many styles;
English/British regnal years (e.g. 8 Henry VI, and Latin roll clauses);
feast days (fixed and movable), and Julian ↔ Gregorian conversion, including Old-Style/New-Style dual dates such as 1641/2;
a range of non-Western calendars — Islamic (Hijri), Hebrew, Thai, Śaka, Persian, Coptic, Ethiopian, Japanese nengō, French Republican, and others — converted to Gregorian intervals.

The panel names the calendar or format it detected, and offers the same validate-all check as for coordinates.

Note

Date interpretation is a genuinely hard problem and the parser is deliberately cautious. Always skim the validation report, and treat the converted ISO dates as a strong first pass to be checked, not an infallible authority — especially for regnal years, movable feasts, and calendar conversions near a year boundary.

7.6. 3 · Reconcile against WHG¶

Reconciliation matches your name column against WHG’s gazetteers using the standard WHG reconciliation service. Only the place name (with a country hint where available) is sent — never your full rows. Progress is shown, and you can stop and resume.

Note

Every row is reconciled individually — identical names are not merged. Earlier designs de-duplicated names and fanned one result out to all rows sharing it. The Workbench deliberately does not do this: you have already disambiguated your places, so two rows both called “Newton” may be different Newtons and each deserves its own candidate list, its own match, and its own geometry. This makes the row count and the query count the same.

Iterative, containment-chained reconciliation. If you defined a spatial hierarchy, the Workbench reconciles it one column at a time, parent → child, and gates each step on your review:

You reconcile the outermost column first (e.g. County).
You review and confirm its matches (step 4).
Only then does the next column unlock — and it is matched within the places you just confirmed (a Parish is searched inside its County, a Place inside its Parish). This containment sharpens the disambiguation of same-name places far more than a country hint alone.

A column switcher — parent → child pills, shown in both the reconcile and review panes — tracks each column’s state (ready, in review, confirmed, or locked) and lets you focus a column to review or re-run it. Because a child inherits containment from its confirmed parents, changing a parent decision later resets the columns below it, so the chain always stays consistent.

Controls that shape the results:

Auto-confirm threshold. A match at or above the score you set (or an exact name match) is accepted automatically — unless it is ambiguous: if two or more different places tie for the top score (say a “Devon” in several countries), the row is held back for review rather than guessed. This leaves only the genuinely uncertain cases for you to decide.
Per-column Sources. Each column can draw on its own source gazetteers, which is often what you want: reconcile a County column against a historic-counties gazetteer, a Parish column against a parish gazetteer, and the place names against everything. Open Sources (its label shows which column it targets) and choose all sources, prioritise a chosen few (they sort to the top), or only those few (others are excluded from the query). The list of gazetteers, with record counts, comes from WHG’s registry. A column with no explicit choice uses all sources; a per-column choice never changes another column’s default.
Re-reconcile a column. Realised a column needs a different gazetteer after confirming it? Select it in the switcher, change its Sources, and press Re-reconcile ‹column› to run it again; the columns below it reset so the new containment flows through.
Phonetic matching (on by default).* As well as text matching, the Workbench computes a phonetic embedding for each name — using WHG’s Symphonym model running entirely in your browser — and sends it with the query, so candidates are ranked by how the name sounds, not just how it is spelled. This helps across spelling variants, transliterations, and scripts. A Language selector (auto-detected from your data, overridable) conditions the embedding; set it if the automatic guess is wrong. The first run downloads the model (~20 MB, then cached); untick the box to fall back to plain text matching.

7.7. 4 · Review & confirm matches¶

This step walks you through the rows that need a human decision. For each one you see the ranked candidate matches on the left and a map on the right:

candidates are numbered and colour-coded, and the same numbers and colours mark their locations on the map; a ★ shows your own coordinate for the place if you supplied one;
hovering a candidate in the list highlights its marker on the map (and vice versa); accepted candidates get a coloured ring;
hovering a map marker shows the candidate’s source gazetteer (by name) and its alternate names;
the map uses WHG’s portal basemap, with a layer switcher and terrain toggle (your basemap choice is remembered across page loads).

It is built for the keyboard: press 1–9 (or click) to toggle a candidate, x to reject, s to skip, n for no-match, u to undo, and the arrow keys to move between rows. If the first few candidates aren’t enough, load more candidates fetches a larger batch. A “review all” toggle lets you revisit even the auto-confirmed matches. Decisions are saved as you go.

Accepting more than one match. Toggling lets you accept several candidates for a row — each becomes a closeMatch (a place legitimately linked to more than one record, e.g. the same place in both GeoNames and a WHG dataset).

Setting the location. A Location toolbar lets you fix the geometry for the row: Use match location clones the selected match’s geometry (point, line, or polygon) into your data, or you can draw your own — Point, Line, or Polygon, clicking on the map to add vertices and Finish to complete. Press a shape’s button again to add another part (→ Multi-point / -line / -polygon). Clear removes the override. Whatever you set here wins on export.

7.8. 5 · Map¶

Every located row appears on a single map, built from the coordinates the Workbench converted (or a geometry you drew or cloned in review). Hover any point to see its details — name, administrative context, date, confirmed match, and coordinates.

The map is designed to stay fast at scale: points cluster as you zoom out (click a cluster to zoom in and split it), and a heatmap takes over at low zoom, so a table of thousands of places renders smoothly in the browser without a server round-trip.

7.9. 6 · Enrich & export¶

The final step produces an augmented copy of your table for use elsewhere — generated entirely in your browser, with nothing uploaded. Your original columns are always kept; you choose which augmented columns to add:

WGS84 latitude & longitude — the coordinates the Workbench converted from whatever format your source used;
ISO start & end dates — the standardised dates from step 2;
Confirmed WHG match — the identifier, title, score, and source gazetteer of the candidate you accepted for each place;
Enrich from WHG — richer detail for your confirmed matches (the matched place’s coordinates, variant names, description, type, and — for Wikidata-backed matches — a Wikipedia article link), fetched from WHG.

Choose a format and export:

CSV and JSON — your table plus the augmented columns, for spreadsheets or other software;
Linked Places (LPF GeoJSON) — WHG’s own upload format, and the best starting point if you go on to contribute the data.

Note

The LPF export maps a sensible subset of your data onto WHG’s upload schema as a starting point — you will usually want to review and complete it (titles, sources, licences, and so on) before uploading.

Contribute to WHG — one click. When your data is ready, the Contribute to WHG button builds the Linked Places file and submits it straight to WHG’s upload and publication workflow for you — no separate export/upload step. You land on WHG’s validation page to review and publish; your local copy stays in the browser. Publishing links your places with records for the same places from other datasets — the step that generates the rich Place Portal pages. The Workbench is the preparation bench; publication remains the way your work becomes part of WHG.

7.10. Caveats¶

Warning

Preview / staff only. The Workbench is not yet released to users; behaviour may change.
Your data is local. It lives in this browser only. Take a .whgproj backup before clearing browser data or switching computers, and note that this local copy is not itself a contribution to WHG — publishing still goes through the normal upload and reconciliation/accessioning workflow.
Automated conversions need checking. Coordinate and, especially, historical date conversions are best treated as a well-informed first pass. Use the validation reports and spot-check the results.
Reconciliation suggests, you decide. A high match score is a prompt for a human judgement, not a guarantee; the meaning of a confirmed match is a closeMatch assertion, explained in Reconciliation & Accessioning.