The neighbourhood a duty officer trusts
A duty officer works a shift against a risk map. The map is built from last year's incident density, staffing patterns tuned to it, and a mental model of where the borough gets difficult on a Friday night. Most of the time this works, because most incidents cluster where incidents have always clustered: the same three postcodes generate the same knife calls, the same road generates the same collisions after the pub shuts. That clustering is real, and it is why local knowledge earns its keep. A duty officer who has worked a patch for ten years reads a thin call log the way a sailor reads a change in the wind — correctly, usually, before the data confirms it.
The trouble starts with the word "usually." Public safety is not one neighbourhood. It is incident feeds, dispatch telemetry, sensor networks and weather, all running as separate streams that occasionally reach across the map and detonate somewhere the risk model never priced. A gas main fails eleven miles from the nearest crew because a contractor's excavator, working on an unrelated job, sheared it. A heatwave three days old has quietly pushed ambulance call volume up 40 per cent while nobody re-staffed for it. A protest scheduled for a stadium car park pulls resources from a housing estate that then has its worst night in a year, purely by the coincidence of shift arithmetic. None of these are anomalies in the sense of being rare events from nowhere. They are ordinary transmissions along long edges that the risk map does not draw, because the risk map is built from local history and local history is clustered by definition.
Two ways to read the same map
Set the two positions against each other plainly, because both are defensible and neither is stupid.
Position one: the local model is doing its job. Emergency response is triaged locally because it must act locally — a crew can only go to one place. A risk map built on last year's pattern is not naive; it is calibrated on the base rate that actually governs most nights. Overfitting resources to rare cross-network shocks means under-resourcing the Friday-night knife calls that happen with dull, awful regularity. The duty officer who chases every distant signal will starve the neighbourhood that needs the crew now. Damping matters here as much as anywhere: most disturbances upstream — a substation trip, a motorway pile-up two boroughs over — never reach the shift's variables at all. They are absorbed by other agencies, other capacity, other people's problems. A model tuned to realised local demand, not to everything that is theoretically reachable, is the responsible design, not the lazy one.
Position two: short paths make the local model wrong exactly when it matters most. Public safety networks are small-world in the strict Watts–Strogatz sense: dense local clustering — call types, streets, shift patterns that all correlate with each other — bridged by a small number of long edges that carry weather systems, mutual-aid corridors, transport disruption and epidemic curves across the whole map in a handful of hops. Watts and Strogatz's original result held for a power grid of nearly 5,000 nodes averaging 19 hops end to end; small numbers of rewired edges collapsed distances that geography alone said should be enormous. Public safety grids are smaller and better connected than that. A regional ambulance trust, a police force and a fire service typically sit two or three hops from any weather station, any hospital's bed-occupancy feed, any highways-agency incident log. The risk map's local clustering is genuine — and it is also the reason the officer does not see the shock coming until it is inside the neighbourhood.
The map has worked for six years. Every serious incident this shift has fallen inside the pattern it predicts. Why rebuild trust in something that isn't broken?
That objection deserves a straight answer, not a dismissal. The map has worked because damping has, until now, absorbed the long-edge shocks before they reached this particular set of variables. That is not the same as the shocks not existing. Ambulance reserve capacity, mutual-aid slack between forces, spare beds — all draw down under sustained pressure elsewhere, quietly, off the local risk map entirely. The regime in which the local model looks vindicated is the regime in which slack is being spent by events the model was never built to see. When the slack runs out, the failure looks sudden. It was not sudden. It was three hops away for a fortnight.
What the streams are actually for
This is the design consequence, and it is narrower than "watch everything." A duty officer cannot and should not monitor every sensor network on the continent. The claim is not that global surveillance improves local triage. It is that the specific streams already sitting outside the local risk map — regional weather telemetry, neighbouring forces' dispatch load, hospital and utility sensor feeds, transport network status — are the long edges through which the actual shocks travel, and that keeping them open, revisable and provenance-tagged is cheaper than rebuilding the risk map after the fact.
Provenance matters specifically because a duty officer's decisions get audited. If a heatwave advisory two boroughs over quietly reallocates a crew, and that reallocation turns out to matter, the record needs to show which stream carried the signal, when it arrived, and how much confidence it deserved at the time. That is not bureaucratic caution. It is what makes a revised belief defensible rather than a guess dressed up after the event.
The measurement objection, taken seriously
There is a real methodological problem lurking here, and it should not be waved away with the word "topology." Whether a public safety network is formally small-world depends on how nodes are drawn — is a "node" a street, a ward, a force boundary, a sensor? Clustering coefficients drift with that choice, and someone determined to find small-world structure in almost any moderately connected system usually can. The argument does not need the label to survive that critique, though. It needs one number, and public safety has it repeatedly: the lag between a shock at an unobserved node and its arrival at the duty officer's variables. In the 2003 North American blackout the originating fault was a single Ohio transmission line; the collapse reached 55 million people within hours, not because the grid was small in geography but because it was short in hops. Flood and heat events show similar lags of hours to a few days between an upstream trigger and the emergency-call surge it produces downstream. That lag is what defeats the static risk map, regardless of whether the underlying graph earns the small-world label in a journal's strict sense.
Robustness is not a substitute, but it is not a rival either
The strongest version of the local-model defence is not "trust the map" but "build margin instead of watching harder": staff to a buffer, keep mutual-aid agreements standing, let redundancy absorb what the model misses. This is good design, and where it is available it beats observation on cost grounds every time. But margins are set from beliefs about how much slack the system needs, and those beliefs go stale exactly as fast as the network around them changes. A mutual-aid buffer sized for last year's typical cross-border call volume is silently wrong the year an unrelated regional restructuring shifts that volume by a third. Robustness without live intake is a threshold set once and trusted forever — which is superstition wearing an engineering diagram.
Where this narrows, not resolves
The two positions do not collapse into one. Local clustering is real and deserves the weight the duty officer gives it: most nights are exactly what the pattern says they will be, and a model that chased every distant signal would fail the neighbourhood it exists to serve. What narrows is the claim about when static intake is adequate. It is adequate for the modal night and dangerous for the tail, and the tail is precisely where public safety failures become inquiries. The honest position is not "replace the risk map" but "keep it, and keep the long edges open alongside it" — not because everything three hops away is relevant, but because the officer cannot know in advance which three hops will be, and the cost of finding out after the fact is measured in people, not budget lines.