An interactive MapLibre GL map + timeline of every recorded bomb and shell impact on Paris and its suburbs during the First World War.

Open index.html in a browser (keep the images/ folder alongside it). The event data is embedded in the HTML; only the basemap tiles (OpenFreeMap) and the MapLibre library (unpkg CDN) need a network connection.

Published site — the only files GitHub Pages serves (everything the browser fetches):

Source you edit:

Generated data — committed so the site can be rebuilt without re-fetching everything; do not hand-edit (regenerate via the scripts):

Not committed (see .gitignore): __pycache__/, geo_in.csv/geo_out.csv (geocoder scratch), .claude/, .DS_Store.

The Python scripts are a build-time pipeline that fetches from external services (Wikipedia, the Base Adresse Nationale, geo.api.gouv.fr, Wikimedia Commons). They are not run by the website or by the deploy workflow — run them locally only when you want to refresh the data, then commit the regenerated index.html + images/. The published site is pure static HTML/JS + images.

.github/workflows/deploy.yml deploys on every push to main: it stages just index.html + images/ and publishes them via GitHub Pages (it does not run the Python pipeline). To update the live site, rebuild locally, commit, and push.

• 820 impacts plotted, coloured by weapon type (aircraft / Zeppelin / long-range gun).
• Vector basemap (OpenFreeMap, no API key) re-tinted to a period parchment palette.
• Impacts sized by casualties; a gold ring marks those with a photograph.
• Circle size scaled by casualties.
• Paris Gun firing positions, time-linked (with the long-range-gun layer): the timeline names the active firing position as it moves over 1918 — Forest of Saint-Gobain (~113 km) → Beaumont-en-Beine (~109 km) → north of Château-Thierry (~91 km) — with a red trajectory line to Paris. "View firing area" zooms out to show all three positions at true regional scale.
• Timeline with a play button (scrub or animate through the war), in cumulative or 7-day window mode.
• Filter by weapon type; live killed/wounded stats.
• Click any impact for date, address, casualties, the original French note, and per-event source links (the contemporary newspaper article, plus the Wikipedia table).
• Photographs: events with a damage photo (or a present-day commemorative plaque) are marked with a gold ring; the popup shows the captioned image(s) — historical damage photo plus, where one exists, the modern plaque at that address — click to enlarge in a lightbox. The Sources drawer also has a Weapons & context gallery (Taube, Gotha, Pariser Kanon, shell diagram, and the historical Paris Gun shell-impact map).
• A Sources & method drawer listing the authoritative collections behind the data.

Wikipedia compiles three layers of documentation, all surfaced in the app:

• Per-event: contemporary newspapers (L'Excelsior, Le Petit Parisien) digitised on Gallica (BnF) — linked in each impact's popup where present.
• Authoritative collections: Musée de l'Armée — “Paris bombardée”, the Commission du Vieux Paris ArcGIS dossier, and Ville de Paris archives.
• Period works: Mortane's La Guerre aérienne illustrée and Rousset's La victoire (Gallica).

1. page_wikitext.json — raw wikitext of the French Wikipedia article, fetched via the MediaWiki API.
2. python3 parse.py → events.json — parses the recapitulative table (date, weapon type, address, place, comment, casualties).
3. python3 fetch_insee.py → insee.json — resolves each commune to its INSEE code and official centre via geo.api.gouv.fr (Île-de-France only, with alias handling for quartiers).
4. python3 geocode.py → events.geo.json — geocodes addresses via the French Base Adresse Nationale bulk API. The commune is applied as a strict INSEE citycode filter (arrondissements by postcode), so a street name can never match a same-named street in another town. Each result is validated (right area + plausible street name); anything unconfirmed falls back to the commune / arrondissement centre.
5. python3 fetch_images.py → images/ + credits.json — downloads every image used (damage photos, plaques, context illustrations) from Wikimedia Commons at 1280 px, and records each one's licence + author for attribution.
6. python3 build_html.py → index.html — embeds the data, credits and image paths.

Rebuild everything: python3 parse.py && python3 fetch_insee.py && python3 geocode.py && python3 fetch_images.py && python3 build_html.py

Images are cached locally in images/ so the map works offline (only the basemap tiles need a network). Licences (from Commons): ~33 public domain, a few CC0 / "no restrictions", and several CC BY-SA (the modern plaque photos and a handful of illustrations). The CC images carry their author + licence in the lightbox and a short credit under the popup thumbnail; every image also links to its Commons file page. Redistribution is permitted under these licences provided that attribution is shown, which it is. images/ (~16 MB) ships alongside index.html.

• Casualties are parsed from free-text French notes. Totals: 285 killed, 755 wounded across the table. Many minor impacts list no casualties.
• Locations: 678 of 820 resolved to exact street addresses (geocoded within the correct commune via its INSEE code); 142 fell back to the area centre (shown with a ⚠ note in the popup) — renamed/demolished streets, intersections, and landmarks that can't be confirmed to a single point.
• The deadliest single event is correctly the Saint-Gervais church shelling (29 March 1918, 92 killed).
• Source data is as complete/accurate as the Wikipedia table; it is the most granular public list of these events.