What live incident data taught us about uncertainty
A live map can look precise while its sources are not. One day of refining P2000 Lens showed us that uncertainty has to be designed into the data, the interface and every route we calculate.
The question behind the work
The problem
P2000 Lens had to turn short, inconsistent public alert records into something people could actually use: an incident, a location, linked units, a possible departure point and later context. Live use quickly exposed where a confident-looking interface could become misleading. An original message could lose detail during normalisation, separate unit alerts could look like separate incidents, approximate coordinates could place a marker in the sea, and a known vehicle or station could still appear as unknown elsewhere on the same page.
01
What we tried
- Kept the exact source message, parsed facts and public summary as three separate layers instead of letting one overwrite another.
- Merged matching source records and linked multiple units to one incident, while separating physical vehicles from crews and officer functions.
- Classified locations, stations and routes by confidence, then allowed a clearly labelled regional estimate when an exact destination was unavailable.
- Moved feed processing into one scheduled background ingest so every visitor reads the same normalised incident store instead of rebuilding external feeds in the browser.
- Used real public incidents and repeated desktop and mobile checks to correct the interface, rather than treating fixture success as product proof.
02
What happened
- The original alert text now remains available unchanged, while richer incident descriptions and unit relationships are built alongside it.
- Implausible coordinates are rejected before they become public markers. Incidents can remain useful in the feed even when their exact map position is unresolved.
- Known vehicle and station evidence now produces estimated routes without presenting them as live vehicle tracking. Verified addresses can support a street-level destination; place-only alerts remain visibly approximate.
- The route assessment became standard for every current public incident. In the live verification sample, all 13 incidents were assessed: two had enough evidence to calculate a route and eleven remained safely unknown at the missing origin or destination step.
- The public interface became faster and calmer: background caching removed external feed work from page views, while lifecycle rules and compact markers kept old or uncertain records from dominating the live map.
03
What we learned
Unknown is not an implementation failure. It is a real product state that should be visible, understandable and allowed to improve later.
Useful does not always mean exact. A clearly labelled regional estimate can help orientation, while invented precision damages trust.
An incident is not one source line. It is a changing record that can contain several alerts, units, sources, confidence levels and later updates.
A route is only responsible when both ends have evidence. The right default is to attempt the same assessment for every incident, then stop exactly where the evidence stops.
Live data tests more than the parser. It tests the map, wording, hierarchy, caching, expiry rules and whether the product tells the truth under incomplete information.
04 · Outcome
The decision
ADOPT. Treat uncertainty as first-class product data: preserve the source, separate facts from presentation, attach confidence to every enrichment and prefer an explicit unknown over a convincing guess.
Where the evidence leads
What this means next
Keep P2000 Lens in personal long-term testing. Measure missing information, false matches, source latency and route usefulness; expand the vehicle and station registry only when repeated, current evidence supports it.
Open P2000 Lens — Live Beta →