Scores are global-scale research estimates. R marks work in progress; none are travel advice.
Found a hotel? Check the area around it before booking.
Search sends your query and selected city to OpenStreetMap Nominatim. Don't enter private addresses or sensitive information. Search policy
Know before you roam
A global-scale research estimate anchored by official 2025 Tokyo crime evidence, with Japan-level fallbacks and explicit uncertainty for dimensions without city-level coverage.
Safrom's citywide research estimate is 86/100. Higher scores mean safer on the model's scale. Evidence as of . Confidence: B.
3056 areas with estimates; 88 additional boundaries without estimates. Area evidence as of .
Global-scale research estimate anchored to Tokyo's city score. Official 2025 recorded-crime counts are divided by 2020 census resident population and shrunk toward the Tokyo-wide rate before mapping to the shared safety scale. Resident population does not represent commuters, nightlife, or traveler exposure; zero or masked population groups and boundary-only areas remain uncolored. Statistical small-area boundaries are not parcel boundaries. Not travel advice.
Boundaries: Statistics Bureau of Japan / e-Stat · 2020 Population Census small-area boundaries — Tokyo (e-Stat Terms of Use / CC BY 4.0 compatible).
Research-only estimates. Scores combine city-local evidence with documented national or regional fallback inputs; confidence communicates uncertainty. They are not travel advice.
UK-traveler perspective; named entertainment districts are qualitative alerts rather than comparable incidence rates; current advice must be refreshed independently from the long-term baseline
alerts are not long-term baseline risk; event footprint may not equal traveler exposure; alert severity cannot be copied directly into Safrom score
country-level only; humanitarian risk framing differs from traveler exposure; composite inputs can overlap Safrom dimensions; current event conditions require separate signals
country-level only; reporting and justice-system differences; latest year differs by country; not traveler-specific
country-level only; availability does not measure emergency accessibility; latest year differs by country; distribution within a country is hidden
country-level only; collection is not annual for every country; some values are modeled estimates; resident exposure differs from traveler exposure
U.S.-traveler perspective; Level 1 is a broad country advisory and cannot represent Tokyo neighborhood variation; advisory level is not an absolute safety score
country-level only; modeled perceptions rather than traveler incidents; cannot represent neighborhood variation or current events; confidence intervals must not be discarded
police-recorded offences exclude unreported crime; 2020 resident population is five years older than the 2025 crime snapshot and does not represent commuter visitor hotel-night or traveler-day exposure; 58 evidence groups contain a zero or statistically masked population component and remain uncolored; 30 boundary-only features remain uncolored; neighborhood labels do not carry stable codes and require the separately versioned e-Stat identity join; the reviewed 2025-to-2020 join maps 3,113 of 3,120 neighborhood labels; 6 unknown-geography rows and the excluded Koto Central Breakwater row remain intentionally uncolored; the non-zero former Saneicho and current Yotsuya Saneicho rows are retained separately in provenance and summed once on their unchanged official boundary; each special ward appears as both a ward total and a ward subtotal and those rows must not be added to neighborhoods; the 37 columns have exactly one versioned product role so category totals and components are never added together; area values are a global-scale resident-rate research estimate with shrinkage rather than traveler victimization rates; the latest annual resource is fixed by URL and checksum because the Tokyo CKAN resource list had not yet registered R7 at validation time
e-Stat states that statistical small-area boundaries were created for statistical work and do not necessarily match general regional or administrative boundaries; the source is 2020 while the crime snapshot is 2025, so 13 changed-name or split-area labels use explicit versioned reviewed mappings rather than fuzzy matching; five shape records are duplicate parts of stable KEY_CODE areas and must be merged rather than scored separately; two selected special-ward shapes have no S_NAME and are excluded from address identity matching; six unknown-geography rows and Koto Central Breakwater remain intentionally uncolored; the 8.8 MB research GeoJSON is appropriate for the local demo but should become a vector-tile layer before production deployment
registered residents do not represent commuters visitors hotel nights or traveler-days; the publisher states that neighborhood names and ordering follow municipality reports and the table does not expose stable e-Stat small-area keys; the real 2025 CSV returned HTTP 403 from the current execution environment so checksum and row counts are intentionally not invented; the connector requires an explicit reviewed SHA-256 and validates 23 ward identities sex totals ward totals and normalized-name uniqueness; the older CKAN datastore preview stops at 1250 rows in Ota and must not substitute for the complete 2025 file; no rate score or Web color is generated before snapshot identity exposure and calibration review