One operational picture

Any operation that matters is already instrumented. Feeds arrive from environmental services, from transport and weather, from markets and from people on the ground. What is scarce is not the data but the ability to hold it together long enough to act on it.

When each feed lives in its own screen, the integration work falls on the person watching. They correlate by memory, they reconcile names and dates by hand, and every conclusion they reach is difficult to hand over because it exists in their head rather than in the system.

Ten dashboards do not make a picture; they make ten places to look.

A picture is not one screen, and it is not a summary. It is three things held consistent at once: one map, one timeline and one identity per thing.

One map means every layer shares the same geography — the same provinces, the same coastline, the same coordinate space — so two sources can be read against each other without one of them being rounded to a coarser shape. One timeline means every dated record answers to the same clock and the same range filters, with playback that turns a sequence of readings into a development you can watch.

• One map — every layer over the same geography, comparable by default.
• One timeline — one clock, one set of range filters, playback over any window.
• One identity — a place or an organisation is one entity across every source.
• One queue — signals raised from any of them land in the same place to be worked.

Maps and timelines are engineering; identity is the part that quietly decides whether a picture is real. If the same port is three records, the platform cannot count anything reliably, and every question about that port is answered three times in three ways.

That is why the entity layer sits underneath everything else. Mentions are resolved as data is ingested, relations are stored against canonical identifiers, and a query returns what every source knows about one thing — which is the difference between a picture and a collage.

• Resolution at ingestion — names matched before a record is created.
• Relations — the meaning between entities, readable in either direction.
• Proximity — what is near a place or an event, returned as entities rather than points.
• Provenance — every entity keeps the sources that describe it.

A picture that only shows is a briefing. The second half of the idea is that the reading and the work that follows it share a frame: a rule that fires on a metric raises an alert, the alert opens as an event with an owner and a history, and the response work stays beside the analysis that prompted it.

That ordering is what keeps the picture current. Because the same metric can be charted, alerted on and summarised without a second definition drifting away from the first, there is one version of events to argue about — which is the only kind worth arguing about.

• Rules — conditions evaluated against the same metrics the map draws.
• Events — a signal raised, owned, worked and closed with its history attached.
• Reports — the period summarised in writing for whoever was not watching.
• Assistant — questions asked over platform data and over your own documents.

The measurable change is time to awareness: how long passes between something happening and somebody being in a position to respond. The less measurable one is the quality of the handover. A team with a shared picture starts a shift, a meeting or an escalation from the same reference point instead of from a handful of links and a recollection.

It also changes what can be delegated. A question with a defined answer — what is within fifty kilometres of this event, what has changed at this passage since yesterday, what did the period look like — can be answered by the platform, which leaves the interpretation to the people paid to interpret.

The test of a picture is whether somebody else can start from it without you in the room.

It does not improve its sources. A feed that publishes late, revises its figures or covers one area more densely than another will put exactly those properties into the picture, and the platform's answer is to show timestamps, provenance and sparsity rather than to smooth them away.

It does not replace judgement, and it does not remove the need for thresholds to be set by people who will be wrong sometimes. What it does is keep the evidence of those choices — which value was seen, which comparison was made, which rule failed to fire — so a bad setting stays correctable instead of invisible.

The reasonable way to evaluate a claim like this is to test it against something you already know. Name a region and a period, bring the sources you trust, and read the picture alongside what you remember of it.

That is what our briefings are: not a tour of features but a walk through one area, one period and one question, with the layers, the entity graph, the rules and the reports in the order the work needs them.