How ratings work
Every rating on this site is computed from real, verifiable data — restaurant databases, police crime statistics, transit authority records, and OpenStreetMap. We collect data from 6+ sources, normalize using statistical percentiles across all 1,493 stations in Greater Tokyo, and combine multiple signals per category.
252 stations also have individually researched descriptions with human-verified ratings. The remaining stations are purely data-driven.
Rating pipeline
- 1Scrape — Automated scrapers collect POI counts, crime data, passenger volumes, and rent prices from official APIs and open data.
- 2Normalize — Raw counts are log-transformed (to handle extreme skew), then ranked by percentile across all 1,493 stations.
- 3Cap — Absolute caps gate the top tiers: a “10” for transport requires 5+ train lines, not just being in the top percentile.
- 4Merge — Pipeline ratings are merged with the 252 human-researched stations. Where they agree, pipeline confidence is inherited; where they differ, the human rating is marked “Curated.”
- 5Export — Ratings, confidence metadata, and source attribution are baked into the site at build time. No database at runtime.
Data sources by category
Food & Dining
100% coverage- •HotPepper Gourmet API (restaurant/izakaya/bar counts)
- •OpenStreetMap (restaurant/cafe/fast food POIs)
Two independent sources with r=0.855 correlation.
Nightlife
100% coverage- •HotPepper (late-night shops via midnight=1, izakaya, bars)
- •OpenStreetMap (bars, pubs, nightclubs, karaoke, hostels)
Weighted composite of 7 signals including late-night establishments.
Transport
100% coverage- •Station line count (ekidata)
- •MLIT S12 daily passenger counts
94% of stations have official MLIT passenger data.
Rent / Affordability
100% coverage- •Suumo station-level scrape
- •Ward-average fallback (Nominatim)
- •Log-linear distance regression
18% station-level, 48% ward average, 34% regression estimate. Inverted: cheaper = higher rating.
Safety
100% coverage- •Keishicho ArcGIS (Tokyo neighborhood polygons)
- •Prefectural police ward-level data
Tokyo stations use neighborhood-level crime polygons. Other prefectures use ward-level data.
Green & Parks
94% coverage- •OpenStreetMap (parks, gardens, nature reserves, forests)
Currently uses park count. Area-based scoring in progress.
Gym & Sports
94% coverage- •OpenStreetMap (fitness centres, sports centres, swimming pools)
Vibe & Culture
98% coverage- •OpenStreetMap (theatres, cinemas, arts centres, bookshops, record shops, vintage shops)
- •Pedestrian street density
Cultural venue density differentiates neighborhood character. 252 stations also have editorial ratings.
Quietness
100% coverage- •MLIT S12 daily passenger counts
- •HotPepper commercial density (fallback)
Inverted: fewer passengers = higher rating.
Daily Essentials
100% coverage- •OpenStreetMap (supermarkets, pharmacies, clinics, dentists, banks, laundry, post offices, schools, kindergartens)
9 subcategories weighted by daily-life importance. 1491 stations from direct OSM data, 2 from proxy.
Confidence levels (Data Depth)
Each category rating shows a shape icon indicating how much data backs it. Shape encodes the level — readable without color.
Color system
Map markers and the ranked list use a diverging palette based on traditional Japanese pigments. Color encodes how a station compares to the median, not the raw score.
The palette recalculates as you adjust weight sliders, so the full color range always spans the current distribution.
Weighted scoring
The composite score you see on the map is a weighted average of all 9 category ratings. Default weights emphasize rent (20%) and transport (20%), but you can drag the sliders to match your priorities. Dealbreaker filters (max rent, max commute, category minimums) are applied independently — they hide stations outright rather than lowering their score.
Known limitations
- ⚠Rent: Only 18% of stations have real station-level rent data (Suumo). The rest use ward averages or a distance-based regression. Tourist towns (e.g., Hakone) may show unrealistically low rents.
- ⚠Safety outside Tokyo: Kanagawa, Saitama, and Chiba stations use ward/city-level crime data, not neighborhood-level. Actual safety may vary within a ward.
- ⚠Green spaces: Currently based on park count, not area. A station near one large park (Yoyogi, 54ha) may score similarly to one near many small pocket parks.
- ⚠Transit times: Computed from a geographic model calibrated against 252 ground-truth values (MAE 5.5 min). Not timetable-based — actual times depend on transfers, express services, and time of day.
- ⚠Food data: HotPepper API is the primary source. Small independent restaurants without HotPepper listings may be undercounted, especially in rural areas.
- ⚠Last train: Sourced from mini-tokyo-3d (MIT licensed). We show the latest boardable departure in any direction. Weekday vs Sat/Holiday are split; Saturday and Sunday/Holiday timetables are combined in the source. Post-midnight times appear as 00:xx (no 24:00+ convention).
Data freshness
Ratings were last computed in April 2026. Crime data is from 2024 (Keishicho annual report). Passenger counts are from MLIT FY2021. Rent data is from Suumo snapshots taken in April 2026. OSM data reflects the state of OpenStreetMap at scrape time (April 2026).
Disagree with a rating?
Every station page has a feedback button. If you live near a station and think a rating is wrong, tell us — your local knowledge helps improve the data.