Haunt Observatory — Methodology
Haunt Observatory is a processed directory desk for 734 community-listed places — pubs, castles, folklore sites, and related venues. It is not a sensor network, not a claim verdict engine, and not a UFO-style sighting event database. Pins mark places people report as haunted or lore-linked; they do not establish that any claim is true.
What we refuse to invent
- Haunt-report years without a note we can stand behind
- Source channel tags we did not actually record
- Quality scores, emotion classifiers, or sentiment polarity over thin narratives
- Playback of “sightings over time” as if each place were an event stream
- Field-report links inventively glued to haunts without a real foreign key
Fields you will see
- Place basics — name, type, region, short description, optional longer story, links, photos when provided.
- Country — structured field when curated; otherwise a region-tail fallback (for example York, England → England). Parsing is imperfect and stated plainly.
- Haunt-report year — earliest year a haunting or folklore claim is documented in our notes — not the year a castle was built. Many listings remain undated. Year filters hide undated places and say so.
- Source channels — affirmative tags only (historical record, folklore, eyewitness, media, community submission). Empty means untagged, not “we checked and found nothing.”
Pipeline architecture
Honest rebuild path for directory growth. LLM or scrape enrichments that invent provenance are rejected at review.
1. Seed / community submit place 2. Admin review (approve → public pin) 3. Story + short_desc + site_status audit 4. Images / links enrichment (Wikimedia, official sites) 5. Structured country curation (or region-tail fallback) 6. Provenance pass — earliest_haunt_report_year + note only when notes already support the year 7. Source channel tags — affirmative only 8. Facets RPC refresh (types, countries, eras, sources) 9. Observatory Map filters + shareable URLs 10. Timeline / Insights read live facets (no mock series) 11. Cross-dossier links when ScareBnB / events / venue FK exists 12. Batch growth (+N audit-ready places) → repeat from 3
Map behaviour
- Density mode shows geographic listing clusters — not credibility. It respects the year filter when active; source filters apply to pins and search, not the heat layer.
- Shareable URLs carry category, country, year range, and sources so a desk view can be sent.
- Cluster taps open a place list. Selecting a haunt loads linked ScareBnB, events, and venue pages when they exist.
- Community submissions approved from admin are tagged
community_submission.
Timeline & Insights
Timeline charts dated places and curated historical anchors. Insights shows coverage (types, countries, dating %, source tags). Both read the same live facets as the Map. Neither fabricates per-year sighting volumes, median quality, movement categories, or emotion distributions.
Live coverage snapshot
- Places
- 734
- Countries
- 129
- Dated
- 18
- Undated
- 716