Weaker evidence
Medium and High sensitivity can lightly reveal some report-bearing signals between the city baseline and trained elevated cutoff. Low keeps them uncoloured.
Uncoloured never means guaranteed safety. Prototype
Our methodology
The map compares patterns—not personal danger.
StreetLens estimates where severity-adjusted police reports are unusually concentrated for the selected time—not the chance that any one person will be harmed. It corrects for window length, street length, setting, and measured pedestrian activity where available, then tests its thresholds on a later year.
Explore the mapDeliberate display threshold
Most reported activity stays uncoloured.
Low sensitivity begins colour only when the normalized signal clears a threshold learned from 2024 data. Medium and High reveal one or two thirds of the gap from that cutoff toward the model's city baseline without changing the underlying score or strongest-pattern cutoff. The shooting-and-homicide view is the explicit exception: every report-bearing corridor is visible. A report by itself is not a warning.
Street-corridor model
Evidence before colour.
117,582 deduplicated reported-occurrence records were matched to 56,812 City street segments. The public map requests only the compact vector tiles visible on screen.
Separate volume from harm
Deduplicate the police feeds, retain raw category counts, and apply bounded severity weights. Every selected category combination is calibrated as one distribution, rather than displayed as a growing union of independent hotspots.
Limit the geography
Anchor each anonymized point to a City intersection and conserve its spatial mass along connected streets within 90 metres.
Normalize and stabilize
Divide by days, selected hours, street length, and available activity exposure; then shrink sparse estimates toward the city baseline.
Persistence check on 2025
Learn thresholds from 2024 and check whether the same daytime, evening, and late-night corridor groups remain elevated in 2025. This is not a forecast or proof of real-world safety accuracy.
What the data means
Street signal, not street verdict.
Counts alone cannot measure individual risk. Busy stations, nightlife areas, and retail streets expose more people and generate more records. Recent City pedestrian counts therefore provide a conservative one-sided correction: high measured activity can lower a signal, while a missing or low count is never allowed to raise one.
Explicit location groups: Public spaces means streets, roads, open areas, parks, and parking lots; Residential means homes, apartments, hotels, group homes, shelters, and seniors' residences; Businesses means shops, offices, bars, restaurants, construction sites, and other commercial places; Transit means ttc, go, passenger rail, cargo rail, stations, stops, tracks, yards, and transit facilities; Institutions means schools, universities, medical facilities, detention centres, police, courts, religious, government, and non-profit places; Unknown means records whose published location is missing or explicitly marked unknown.
Area display: Elevated corridor estimates are projected into a continuous, kernel-smoothed context layer. The smoothing changes only the display—not the underlying score or the claimed incident location—and is generalized at citywide zooms so broad patterns remain legible.
Three months, six months, YTD, and one year are annualized onto the same per-day, per-hour scale. Longer windows are emphasized because they are more stable.
9,232 corridors near recent pedestrian counts receive a capped activity denominator; the rest are explicitly not claimed as per-person rates.
Severity is bounded at 5×, and indoor or institutional occurrences receive less street relevance. Published location groups remain filterable and visible.
Community submissions are private leads and never alter the map automatically.
Privacy
Conditions without surveillance.
We do not publish names, faces, identifying descriptions, or exact locations of vulnerable individuals.
- No tracking or profiling of individuals
- No unsupported allegations about a person or facility
- No tools for coordinating confrontations or harassment
Limitations
An honest map admits uncertainty.
Police reports and pedestrian counts both have incomplete, uneven coverage. The model reduces known distortions but cannot remove reporting bias or create a complete exposure denominator.
- An incident does not establish a pattern
- No colour does not prove an area is safe
- Colour is relative concentration, not a probability or causal claim
- Fixed shortcuts use vector tiles; custom dates load only overlapping daily aggregate chunks
- Three-month and shorter custom windows carry substantially more uncertainty
Early public-interest prototype
Submit a correction StreetLens is not a real-time safety service. It modifies and combines Toronto Police Service reported-occurrence data with the City’s Toronto Centreline and multimodal intersection counts. TPS describes its records as preliminary, approximate, and potentially incomplete. Results may be delayed or wrong; no place is guaranteed safe or unsafe. StreetLens is not endorsed by TPS or the City. Read the source terms and attributions.