These are real screens from PatrolDesk running on Pittsburgh’s published data — 67,631 incidents over 31 months. Every number below is what the tool computed on 6 September 2026. Nothing is a mockup.
The opening screen makes the case for everything that follows, in the department’s own numbers.
Violent crime, last 90 days, every record as a dot. The concentration card on the left: the busiest 45 of 895 occupied 150-meter cells hold 23% of everything in view.
This is the law of crime concentration measured on your own city. Every commander already believes it; seeing it in their own numbers is what makes the rest credible.
Each cell is compared against its own previous 84 days, seasonally corrected. A corner that is always busy is gray. One that just turned is red.
Three violent-crime cells running hotter than their own baseline. The Central Business District does not appear — nothing about it changed. It shows up under CHRONIC instead. The note in amber says these weeks normally run +4% against the baseline and that has been divided out.
Switch to Property and the list is completely different — vandalism, vehicle theft, different neighborhoods. Different problem, different places, different response. That is the case for never looking at “crime” as one thing.
Click any place — on the map or in the list — and it opens.
The reading is a sentence computed from the records: how many are one offense type, what hours they fall in, how tightly grouped, when the run began. Below it, a weekly count across the whole window with the 28-day boundary marked, the offense and day-of-week breakdowns, and the records themselves. Click one and the map goes to it — and stays open so you can work down the list.
Three intimidation reports on Penn Avenue, all between 07:00 and 13:00, all within 43 m — that reads like one address and one dispute, not a neighborhood problem. The tool says so before anyone sends a car. The same sentence on a chronic cell might read “28 of 47 are vehicle code” — the tool catching its own noise, the way a chief who knows the city would say that’s our tow zone.
Day of week by hour of day for whatever is filtered. Violent crime peaks Sunday 02:00–03:00 — bar close on the South Side. Click a square and the records behind it open, including how many of that weekday actually carried incidents.
“Recorded on 13 of 13 Mondays” is a standing problem worth a standing response. “34 of 39 records on one night” is an event that is already over. That is the resource decision a watch commander makes every day, and this is the screen that makes it for them.
The places that never stop: Bluff / Liberty Bridge, Wood Street, 7th Avenue. These are problem-oriented-policing targets — a different list from EMERGING, and it rarely overlaps.
The two panels answer different questions. Emerging says where to send a car this week. Chronic says where the problem is structural — the bar, the corner, the complex — and a car will not fix it.
The part nobody else is selling, and the one that pays for the whole thing.
Blue bars are incident demand by hour. The amber line is officers on duty from the watch model below — edit it to match your roster. Pink hours are under-covered. The finding is generated from the two curves: “40% of the work falls between 08:00 and 14:00, when 20% of the roster is on.” Choose a zone and the table breaks down by sector; choose a sector and it shows that sector’s own busiest hours.
Every figure is a share, so the model does not need real headcounts to say a third of the roster is covering an eighth of the work. Fixing a shift boundary is the only capacity in the building that costs nothing. The card also says what demand it is measuring — reported crime, a subset of calls — and that a CAD feed would make it the full curve.
Three deployments built from what changed. They differ in what they cost and how soon you could prove they worked — not in how clever they are.
Scoped to one zone, targeting its sectors. Each option: the places, the surge window and nights (computed from those places’ own records), a Koper-curve dose of three 12-minute visits, officer-hours per week, where those hours come from, the effect range interventions of this kind have produced, and — the key line — how long until a reduction that size would be measurable when pooled across the set.
HOLD is cheapest and takes a year to prove. SATURATE costs 26 hours a week and is measurable in 16. That trade is arithmetic, not opinion, and it is why a course of action assigns a set of places rather than a corner. No probability of success is ever printed — there is no basis for one.
One page a sergeant reads at roll call.
When, dose, cost. The numbered places with their sector, the pattern, the evidence and the sigma. What to expect, attributed to the published trials. And the review terms: compare the next 28 days against each place’s own baseline, check the 400 m ring for displacement, and record this order before deployment.
That last line is the difference between a measurement and a story told afterwards. It is the pre-registration — and it is on the paper the department keeps.
Incidents joined to arrests by report number. 19% of reported crime cleared by arrest; 84% of enforcement arrests — shown separately, because a drug or DUI record exists because an officer made an arrest and clears by definition. Traffic and parking excluded entirely, and the card says so. By offense: aggravated assault 61%, robbery 32%, theft from a vehicle 2%. By zone: 24% down to 11%.
Everything else in the product is prevention. This is the other lever — where patrol will not help because the problem is investigative. And it is clearance by arrest, stated plainly: comparable across places and offenses, not comparable to a published UCR rate.
East Carson Street as a corridor — a band along the street, because a circle either misses most of a strip or swallows two neighborhoods. Weekend nights against every other night: 3.72×, with weapons offenses at 11.3×. The size ladder shows where the effect lives; the offense table shows only what clears an exact binomial test, and lists the rest as “not separable from noise.”
For a ballgame it is a different profile: 2.15× within 400 m, nothing at 800 m, only on crowds over about 25,000, and the one crime that rises is theft from vehicles. Surge the parking, not the gates. Draw any shape on the map and every figure recomputes inside it.
One zone, one watch — or all three watches on a page. Headline, the previous 24 hours against a typical day, what changed, the near-repeat watch list with the windows that are still open, and the typical load for this watch on this weekday. Paged like a document, printed like one: six pages or eighteen.
Dashboards do not get used; briefs get read. No sentence here is written by a language model — every one is computed from records — which is what makes it safe for someone to act on and auditable afterwards.
Transcribed dispatch audio, logged on the left; select a call and the map goes to it. The historical cards step aside so the column is a monitoring surface; SHOW ANALYSIS brings them back. Pins only where an intersection resolves; where only the street is known, the street is drawn instead of a dot.
Medical and mental-health calls are dropped entirely, names stripped, addresses rounded to the block. It publishes less than the radio does. And it is the floor, not the ceiling — with a read-only CAD feed the same screen goes live with exact addresses.
The map fills the viewport. One button opens a full-screen picker for every selection — live layers, city, data, boundaries, the three sections and their views, window, zone, category — and returns to the map. Findings live in a sheet you drag between three heights.
A commander in a car or at a scene gets the same analysis as the desk, with nothing hidden behind a layout that assumed a monitor.
If your department publishes incident data, we can show you PatrolDesk running on it before the first call ends.
PATROLDESK