How Pauserby reviews a walking route

A route review is not a safety assessment. No matching evidence means only that neither available source returned a match under the documented rules.

How route reviews retrieve nearby evidence

Pauserby looks for mapped pedestrian archive cases within a 100-metre route corridor and active LTA incidents within a 150-metre corridor. These distances decide what to retrieve for inspection; they are not safety thresholds or claims that an incident happened on the route.

Known-reference cases are compared using the shortest distance from their reviewed reference point to the route. For an approximate area, Pauserby subtracts the documented uncertainty radius from that reference distance, never going below zero. This lets an evidence area overlap a route without pretending its reference point is an exact collision location.

Approximate areas with uncertainty radii over 500 metres are not matched automatically. Road-only, under-review, and not-mapped cases also remain in the complete archive instead of receiving invented route positions. Results are ordered by overlap, then effective distance, then recency—not by severity.

A review does not compare alternate routes, produce a safety score, or establish that a route is safe. Evidence near a route also does not establish that the incident happened on the route.

The pedestrian case archive is a separate source

The pedestrian case archive is assembled from public Singapore Police Force releases and news reporting. It does not come from the live LTA feed and is not a complete accident history. A case appearing in the archive means a public report was found, not that comparable cases elsewhere did not occur.

Each case retains the wording that supports its reported location, the source link, the map reference selected during review, and a statement of what remains unknown. A marker represents a named junction, address, or approximate area. It is never presented as a surveyed collision point.

Cases without a defensible point remain available in the list. This includes road-only locations, locations still under review, and reports that do not publicly identify enough detail to place on a map.

How current LTA incidents are collected

Pauserby checks LTA’s Traffic Incidents feed every two minutes. It keeps the source response, records each observation, and groups observations that appear to describe one continuing incident. The standalone Live Incidents view also exposes this current and observed history.

Collection and traceability

The collector requests every page of the Traffic Incidents endpoint. A poll is authoritative only when every required page is received and decoded. The exact response is preserved for diagnostics, while each valid record is stored as a separate observation linked to its poll.

Observations and incidents are different

An observation means one source record seen during one poll. An incident is our deterministic grouping of observations believed to describe one continuing event. The feed does not provide a documented stable incident identifier, so this grouping is necessarily an interpretation.

Time in source messages

LTA messages can begin with a day, month, and time but no year. The collector interprets this in Singapore time and chooses the nearest plausible year around the observation date. When a time cannot be parsed, the observation is retained and the map simply omits that field.

Matching and lifecycle

Matching algorithm deterministic-v1 compares incident type, exact source coordinates, normalised message text, parsed source time, and distance. It first looks for an exact continuing observation, then uses strict nearby time-and-distance rules. No LLM, embedding, or invented confidence score is used.

An incident is marked ended only after it is absent from three consecutive successful, authoritative polls. Failed or partial polls never advance that count. A strongly matching observation may reopen an incident that ended within the previous 15 minutes.

What the live incident source cannot establish

• The source may not contain every traffic accident.
• Coordinates are approximate source locations.
• The source does not provide severity, casualty, or road-user information.
• Historical records begin only when this service starts collecting.
• Time in the feed is not necessarily the duration of an accident.
• More observed incidents do not automatically make a road inherently more dangerous.

← Return to route review