What you can do

Which of three permitted spots should I park my pupusa truck at?

Rosa has sold pupusas at a weekend farmers' market for years. Now she has a food truck and three spots she is allowed to use: near the station, beside a public square and on a quieter neighborhood corner. Where should she spend her lunch hours?

She brings her market sales and what she knows about her regular customers. She also spends time at the three spots, noting when people pass, where they are going and where they stop for food. The crowd rushing to a morning train is different from people looking for lunch.

Her AI assistant brings together that knowledge with walking routes, nearby workplaces, transit schedules and other places to eat. Rosa checks the proposed hours and assumptions about who might stop. They compare the spots on an ordinary office day and a Saturday.

The busiest spot is near the station in the morning, when Rosa would still be preparing lunch. The square brings fewer people overall, but more are nearby when she wants to sell. On Saturday, neighbors take different paths and the quieter corner becomes more interesting.

Then Rosa asks what happens on a day when more people work from home. Her AI assistant explores fewer people coming out of the offices at lunch. She compares that possibility with taking the truck to the neighborhood corner. One location need not be the right choice every day.

Rosa asks where the estimate of potential customers came from and how much the comparison depends on people choosing to stop. The Epistemic Protocol Machine keeps the reasoning connected to the observations and assumptions behind it. She can bring another observation into the work and ask again.

She plans a trial at two spots, using the comparison to choose days and hours worth trying. If she shares her approach, another food seller can adapt it to their own customers and neighborhood.

What Rosa knew. Her regular customers buy lunch at different hours on workdays and weekends.

Will the landlord's shade canopy keep our tenants safe in the next heat wave?

Grace Okafor chairs the tenants' association in a six-story apartment building. Most of the tenants are older, and none of the apartments has air conditioning. Last summer, during a heat wave, two neighbors fell ill at night in their own apartments and went to the hospital. The landlord has offered to put a shade canopy over the courtyard.

Grace knows who in the building lives alone, who uses a walker, and who takes medicines that make heat dangerous. The tenants trust Grace, and a dozen of them kept small recording thermometers, about $110 each, in their apartments for three weeks. Most went in upper-floor apartments facing west, where the heat is worst. One went in a ground-floor apartment facing east, for comparison.

The AI assistant added the shape of the building and the trees around it from aerial photographs, along with local weather records and the forecast.

The simulation worked out how hot the courtyard would feel on a heat-wave afternoon with each of three changes: the landlord's canopy, two street trees, or awnings over the west windows. It also estimated that the walk to the nearest cooling center would take 25 minutes for someone using a walker. Grace walked it with a neighbor who uses one. It took 38 minutes, because the neighbor had to stop and rest twice on the avenue, which has no shade. With the rests added, the simulation agreed.

The canopy makes the courtyard benches bearable at 3 p.m. The tenants' readings reveal the problem the proposal still needs to address: west-facing apartments stay above 86°F (30°C) until nearly midnight, while the east-facing apartment cools off by early evening. A more comfortable afternoon outside would help, but the neighbors who fell ill were inside at night.

The association took the thermometer readings, the timed walk and the simulation's results to the landlord and to the city's heat-wave program. The tenants got awnings for the west windows and a shuttle to the cooling center on the hottest days.

What Grace knew. Who in the building lives alone, who uses a walker, and who takes medicines that make heat dangerous.

How can our block keep the rain out of our basements?

Amara Nwosu leads the block association on a street at the bottom of a hill. In heavy rain, water pours into the basements of three houses. Last summer the Ortiz family lost their furnace and a freezer full of food. The city plans a bigger storm drain at the corner, in about five years.

Amara had watched where the water comes from. It does not back up from the street. It runs down the alley from a church parking lot uphill and pours in through basement windows on the way. During two storms, neighbors took phone videos of the water in the alley and marked on their basement walls how high it rose.

The AI assistant added the rise and fall of the ground from the city's maps, the storm drains and how much water each can take, and rainfall records for the heaviest storms in thirty years.

The neighbors had two ideas: a low curb across the top of the alley to turn the water into the street, and a sunken garden in an unused corner of the church parking lot to hold the water before it reaches the alley. The simulation worked out where the water from a heavy storm would run and settle with each idea, and with both. First it recreated the two storms the neighbors had filmed. It sent the water into the same basements, about as deep as the marks on the walls.

The curb keeps the alley dry but sends the water across the street, into the basement of a neighbor who had never flooded. The garden holds most of the water from an ordinary heavy storm. The AI assistant pointed out that the simulation cannot tell how much water the garden's soil would soak up, so it counted none; a real garden would hold at least as much. In the heaviest storm in the records, the garden overflows. With the alley's drains clear of leaves, the water stays below the lowest basement window. With the drains blocked, it pours in.

The church agreed to give up the corner of its lot. The association took the videos, the marks on the walls and the simulation's results to the city's stormwater program. The city paid for the garden's soil and plants and now clears the alley's drains every fall. The neighbors dropped the curb.

What Amara saw. In heavy rain, the water runs down the alley from a parking lot uphill. It does not back up from the street.

How can I bring the noise down enough to keep my market's permit?

Lin runs a Friday night market in a parking lot, with forty food stalls and a small stage. After neighbors complained about the noise last year, the city council told her it will not renew her permit until she brings a plan to reduce the noise.

Last summer, Lin walked around the block on market nights with a sound meter app on her phone. One reading puzzled her. Outside a senior residence across the lot, the market was louder than it was at a bus stop closer to the stage. She did the same walk on a Friday when rain canceled the market, to measure how loud the street is on its own.

She gave her AI assistant both sets of readings, the layout of the stalls, and the details of her speakers. The AI assistant added the heights of the buildings around the lot, the slope of the ground, and the usual wind on a summer night, which carries sound farther in one direction.

The simulation worked out how loud the market would be at each bedroom window facing the lot, with buildings blocking some of the sound and bouncing some of it back. It compared that with the street's usual noise on a Friday evening. It also explained her puzzling reading: the senior residence faces a tall building that bounces the stage's sound back onto its windows.

Turning the stage around and moving it behind the food stalls made the market quieter than the street's usual evening noise at most windows. At the senior residence, the bounced sound kept it loud, so on that side the music now ends at 9:30.

The council renewed the permit. Lin shared the simulation with the neighbors' association. They opened it, changed the closing time they disagreed with, and ran it again. The two sides settled on 10 p.m. for the rest of the lot.

What Lin noticed. Her phone read louder outside a senior residence than at a bus stop closer to the stage.

Will the new drop-off lane put car exhaust on the playground?

The school district plans to move the drop-off lane at Dev's children's school to the side of the playground. On cold mornings, Dev has seen parents wait in their cars with the engines running for twenty minutes.

For a week, Dev stood at the gate with a clipboard. For each car, he wrote down when it arrived, how long it waited, and whether its engine was running. The parents' association bought two small air-pollution monitors, a few hundred dollars each. Dev fixed one to the playground fence nearest the street where cars wait now, and the other at the far end of the playground, so he could check the difference between them against the simulation.

The AI assistant added readings from the nearest official air-pollution monitor, three years of hourly wind records, the recess times, and traffic counts for the street.

Using Dev's clipboard notes, the simulation recreated the line of cars on a school morning. Then it tried two changes: the district's new lane beside the playground, and the new lane with the younger grades starting 15 minutes later. In each case it followed the exhaust on the wind, hour by hour, and estimated how much of it reaches the children wherever they are that morning.

With the cars where they wait today, the monitor nearest the street read higher than the far one, by about as much as the simulation said it would. With the district's new lane, on most school mornings the usual wind blows the exhaust across the playground during the first recess. Starting the younger grades 15 minutes later cuts the line of waiting cars in half, and the exhaust over the playground with it.

The parents' association took the results to the school board. The board put the drop-off lane on the other side of the building and started the younger grades 15 minutes later.

What Dev saw. On cold mornings, parents wait in their cars with the engines running for twenty minutes.

If a cheaper studio takes some of my members, who can I least afford to lose?

Ana's salsa studio barely survived the pandemic. Now a chain studio has opened four blocks away, charging half her price.

She downloaded eighteen months of bookings from her scheduling app: 240 members, 30 classes a week, and who took which class with whom. She added who brought whom to the studio and who comes to the Friday social dances. The AI assistant added how long each member's trip to the studio takes by bus or on foot, based on their ZIP code, and the same trip to the new chain.

The simulations treated her community as a network: people connected by the classes they take together. They found its natural groups and the members who connect those groups. They removed the members most likely to leave for the cheaper chain and measured how much of the community would still hold together. They also traced how news of a new class would spread by word of mouth.

Ana recognized the names on the list of connectors, but several of them were not strong dancers and rarely took the advanced classes. At the next Friday social she asked five of them why they come. Three drive beginners home after the late class, and two run the studio's group chat. She added the rides to the network as links between people, and the AI assistant ran the simulations again. With the rides included, the connectors mattered even more than the bookings alone had shown.

The members who hold the studio together are a dozen social dancers who take classes at several levels, come on Fridays and drive beginners home. They link the beginners to everyone else, and few of them are among the members who come most often. The chain is closer than Ana's studio for 40% of the beginners.

Ana kept her prices. She made the Friday socials free for anyone taking more than one class, paired each new beginners' group with one of the social dancers, and started paying the drivers' parking.

What Ana noticed. Several of the members who connect her community are not strong dancers. They drive beginners home and run the group chat.

Where should I park my taco truck on parade night?

Marco can make a month's sales from his taco truck in one night: the city's Halloween parade. A week before the parade, the city publishes the route and the streets it will close. Every food truck wants to park on the main avenue.

Last year Marco parked on the avenue behind three other trucks. When the parade ended, he watched much of the crowd head down a side street toward the subway. No truck had parked there, because on the city's map of closed streets it looks like a dead end.

This year he wanted to know whether the new street closures would send the crowd the same way. He bought anonymous phone location data showing how last year's crowd arrived and left, block by block, every fifteen minutes. He walked the route as the barriers went up and found the side street closed to cars but open to people on foot, all the way to the subway. From counts he had made at three earlier events, he knew that about one person in twenty who walks past his truck buys something.

The AI assistant added this year's street closures and how many people use each subway station near the route.

The simulation sent the crowd home from the parade at 9 p.m., on foot through the closed streets, toward the subway and their parked cars. It estimated how many people would pass each spot where a truck can legally park, and how many other trucks they would pass first. Multiplying by Marco's one in twenty gave the sales he could expect at each spot.

The main avenue gets the most people, and every other truck will be there. The side street gets fewer, and all of them would pass Marco's truck first.

He parked on the side street and sold out by 10:30.

What Marco saw. Last year much of the crowd left down a side street that looks like a dead end on the city's map.

Will the new rail stop help the people who live here now, or the people who move in after them?

A study by the city says a new light rail stop will "benefit the neighborhood." Jamal chairs a tenants' association of 300 households near the stop. Most of them are renters who have lived there for decades.

The association's AI assistant gathered public data: rents and household incomes block by block, where residents work, bus and train timetables before and after the stop opens, eviction cases at the county court, and home sales near three stops that opened elsewhere in the region over the past fifteen years. The members added something no public source has. Since the spring, they had brought the association 60 lease renewal letters, which show the rents they are being asked to pay now.

The simulation counted how many jobs people on each block could get to by bus or train within 45 minutes, before and after the stop opens, and whether lower- or higher-income households gain more. Then it took what happened to rents and evictions near the earlier stops and applied the same pattern here.

People within a ten-minute walk of the stop can reach about a third more jobs. But near the earlier stops, rents rose so fast that most of the renters who lived there when the stop opened had moved away within five years. If the same happens here, most of the people who benefit will be people who move in later.

At the association's meeting, a member who has lived on the block for thirty years objected. The earlier stops opened before the city's law limiting rent increases, which passed last year, so their pattern overstates what would happen here. The association asked the AI assistant to run the simulation again with rent increases limited as the new law limits them. The number of renters priced out fell by about a third. In the two blocks nearest the stop, where the law does not cover newer buildings, most renters would still have to leave. The renewal letters from those two blocks already ask for increases of 15 to 25 percent. The public rent figures, which come from surveys several years old, do not show this.

The tenants created a Perspective around a concern the city's definition of benefit had missed: people living near the stop should be able to stay and use it. They compared that Perspective with one focused on the total number of jobs within reach. The same project looked different through those priorities, and the association could show exactly what it wanted the proposal to address.

The association took both sets of results and the letters to the council hearing. It was the first time the association had answered a city study with evidence of its own. The council added a requirement for affordable homes in the development at the stop.

What the members knew. The earlier stops opened before the city's law limiting rent increases, which does not cover the newer buildings nearest the stop.

Where should our second market day go, so it reaches the families who need it?

Teresa manages a farmers' market that doubles food stamps: a family that spends $10 in food stamps takes home $20 of fruit and vegetables. A foundation has offered to pay for one more market day at a second site, and she has three sites to choose from.

She brought her own records of food-stamp purchases, by the shopper's ZIP code. For a month, the volunteers at the market's food-stamp booth also asked shoppers how they got there. Several older shoppers said they used to come every week but now come once a month, because the weekend bus schedule changed in May. So Teresa asked the AI assistant to count, separately, how many residents over 65 could reach each site by bus. Her volunteers also priced the same basket of fruit and vegetables at ten stores.

The AI assistant added the number of households on food stamps in each neighborhood, every store that sells fresh fruit and vegetables, bus timetables, and anonymous phone location data showing where people already shop.

The simulation worked out how far each household lives from fresh fruit and vegetables, and what they cost there. Then it found which site would put the most households within reach: on foot, and by bus for residents over 65.

On foot, the church parking lot is within reach of the most households. By bus, twice as many older residents can get to the library, because a bus runs straight there from two senior apartment buildings.

Teresa proposed alternating: the church lot on Tuesdays and the library on Thursdays. On the trial days, volunteers asked shoppers how they had arrived. Older shoppers at the library described taking the direct bus from the senior buildings; some had stopped coming to the original market. The comparison had helped Teresa find a way to reach them again. The foundation paid for both days.

What Teresa learned. Older shoppers who used to come every week now come once a month, since the weekend bus schedule changed.

Will the construction next door crack my house?

Helen Kowalski has lived in her brick row house, built in the 1920s, for forty years. On the lot next door, a developer will spend three weeks hammering steel posts deep into the ground to hold up a new building. The plaster in Helen's front room is already cracked.

The machines and the schedule are listed in the building permit, which is public. The AI assistant found the type of soil under her street on the national soil maps. Before the work started, Helen photographed every crack with a ruler held against it. She fixed small plastic gauges across the worst cracks to show whether they widen. On the first test day, she and two neighbors farther from the site each put a phone on their basement floor, running a free app that measures vibration.

The simulation worked out how hard the ground would shake under her house each day, as the hammering moves along the site, given the soil under her street. On the test day, the shaking was weaker in the basements farther away, by about as much as the simulation said it would be.

The AI assistant first compared the shaking with the safe limit for ordinary homes. Then Helen told it something no soil map records. Like every house on her street built in the 1920s, hers stands on a foundation of rough stone set in soft lime mortar, and it shares its side walls with the houses on either side. The AI assistant switched to the lower limit for fragile and historic buildings and ran the simulation again.

The shaking under her house goes over that limit in the week the hammering is closest to her wall, and comes close to it on two days of the second week.

She sent the results and her photographs to the developer and the city's building department. For the closest week, the developer agreed to press the posts in slowly instead of hammering them, and to pay a surveyor to record the condition of her walls before and after.

What Helen knew. Her foundation is rough stone set in soft lime mortar, and her house shares its side walls. No soil map records that.

What work can the plant's 600 workers do next, and can they get to it?

The plant that employed 600 people in Sofia's town of 9,000 closed in March. Sofia ran its human resources department for twelve years. Now she runs the small group the town set up to bring in new employers, and everyone asks her what comes next.

The government publishes free figures on jobs in each industry, but it leaves most small-town numbers blank so that no single employer can be identified. Sofia's group paid about $5,000 a year for a service that estimates the missing numbers. The AI assistant added public data on the skills each kind of job uses, where people in the region live and work, and where fast internet is available.

Sofia added the plant's own summary of its workers' skills and certifications. She also surveyed the laid-off workers about how far they could travel to work and whether they have a car. They answered because they know her.

The simulations looked at four things: which industries tend to grow in places with the town's mix of skills; which jobs the plant's workers could move into with the least retraining; which nearby towns draw on the same workers; and how many jobs each group of workers could reach with a car and without one.

Reading the list of jobs the workers matched, Sofia saw a problem in the plant's summary. It listed 120 certified welders, but she knew most of those certifications had lapsed after the plant cut its training budget three years ago. She called eight former supervisors, who confirmed it, and she corrected the figures. The AI assistant ran the simulations again. Making precision metal parts still came first, and the retraining it would take was now clear: about 90 welders would need to renew their certifications.

Making precision parts for medical devices uses most of the skills the plant's workers already have, and two companies that do it are hiring within a 40-minute drive. Sofia's survey found that a quarter of the laid-off workers have no car, and the only bus takes 90 minutes. A shared van to work would put 400 jobs within their reach.

The group used the results to win a federal grant for the van and a welding course, and to pitch the town to a medical-device supplier looking for a second location.

What Sofia knew. Most of the 120 welding certifications on the plant's list had lapsed after it cut its training budget.

Where should the flu vaccine van go on each of six Saturdays?

Nadia is a children's nurse who runs a small community clinic. Before winter, the county has offered her flu vaccine and a van for six Saturday pop-up clinics. She has to decide where the van goes each week.

The health department publishes the share of people vaccinated in each ZIP code, and the school district publishes how many children attend each school. Nadia added her clinic's count, week by week last winter, of children who came in with flu symptoms, by ZIP code. The AI assistant added anonymous phone location data showing how people travel between neighborhoods in a typical week, and bus timetables.

The simulation spread the flu from neighborhood to neighborhood as people travel between them, allowing for how many people in each are vaccinated. The AI assistant started with last winter, and the simulated flu matched her clinic's counts: it reached the same neighborhoods first and peaked in the same week. Then the simulation compared six schedules for the van by how high the winter's peak climbs across the whole area. It also worked out how long each site takes to reach by bus for families without a car.

The obvious choice was the neighborhood with the lowest vaccination rate, where the most children are unprotected. In the comparisons, another place stands out: the market district, where parents from every other neighborhood work and shop, and where the flu passes from one part of the area to the rest. Of the six schedules, the one that starts with two Saturdays there produces the smallest winter peak. Families from the least-vaccinated neighborhood take over an hour to reach the market district by bus. Their school nurse told Nadia they would come to the school on a Saturday, so the van goes there on the next two Saturdays.

The county used her schedule and offered her the van again the following winter.

What Nadia's clinic showed. When last winter's flu reached her waiting room, and which neighborhoods it reached first.

Which distributor will get my beer to the people who already drink it?

Owen runs a small brewery with a taproom, in a town on the outer edge of a big city's suburbs. He wants to sell his beer in shops and bars, and to do that he must sign with a distributor, a company that delivers beer within a set area. Each distributor's area follows county lines.

On a map on the taproom wall, visitors put a pin where they live. Many of the pins are across the river, in the next county, and the taproom fills up when the evening commuter trains arrive. Only some visitors add a pin, so Owen paid about $200 for a month of anonymous phone location data showing where his visitors live. He added his own sales records and the ZIP codes of his mailing list and online orders. The AI assistant added how many people commute between each pair of towns in the region, and where people live.

The simulation drew the map of the region by where people travel every day, instead of by county lines. Then it placed Owen's visitors and customers on that map and laid each distributor's area over it.

His own county line cuts his market in two. Nearly half of his visitors live across the river, a short train ride away, and people from the far half of his own county rarely come. The distributor that covers his county would reach about half the people who already know his beer. The one that covers the counties along the train line would reach nearly all of them.

Before signing, Owen delivered kegs himself for a month to four bars. The two along the train line across the river sold out. The two in the far half of his own county did not.

He signed with the distributor along the train line and set his first sales targets by where his visitors live.

What Owen noticed. Many of the pins on his taproom's wall map are across the river, and the taproom fills when the evening trains arrive.