Waypoint Corpus

This project builds a waypoint corpus from BTT Pirineus GPX files and assigns each waypoint a semantic emoji category. The corpus powers:

• Emoji markers on the map (Leaflet)
• Ordered waypoint markers on the elevation profile

Refreshing the corpus

1. Build/update the GPX index (from shortcode scan data):

npx tsx scripts/build-gpx-index.ts

2. Build the waypoint corpus with clustering:

npx tsx scripts/build-waypoint-corpus.ts

Outputs are saved to scripts/output/ (gitignored).

Output files

• gpx-index.csv: gpx_url, page_url, page_title, page_id, last_seen
• waypoints.csv: gpx_url, wpt_name, wpt_lat, wpt_lng, wpt_ele, order_index, track_dist_km, cluster_id, emoji, category, norm_name
• waypoint-clusters.json: Cluster summaries with emoji and example labels

Clustering + emoji assignment

Waypoint names are normalized (lowercase, diacritics removed) and embedded using Xenova/all-MiniLM-L6-v2. Clustering uses k-means with k ~ sqrt(N) (clamped). Each cluster gets an emoji and category via Catalan-first keyword heuristics:

Tuning

Refine or expand the keyword list in scripts/build-waypoint-corpus.ts (KEYWORD_META).

To adjust clustering:

• Change chooseK (cluster count formula)
• Swap the embedding model
• Add stopword handling for noisy names