Cream City Almanac

EACH WEEK

1. Read DYCU's Hub catalog and its inventory sheet.
2. Group the yearly and geographic versions of each measure into one family with a permanent code, like F02.
3. Read the 180 report PDFs so search can match the text inside them.
4. Open the Neighborhood Portrait spreadsheets, one per neighborhood and year, and store their tables, so the N03 sheet shows any neighborhood's numbers without a download.
5. Ask AI (Claude) to write each family's plain-English explainer, column guide, caveats and story angles. DYCU's own definitions always win.
6. Index everything for search by meaning and by exact words.

A failed week never replaces the live catalog. Data as of Oct 19.

WHAT THE AI DOES

See the diagram of how Ask answers a question: the AI picks a tested tool, code does the math, and every number on the card comes from that code.

Every fact carries a tag saying who wrote it:

• DYCU DYCU's own definition, from its inventory sheet.
• HUB From DYCU's Hub listing.
• SOURCE From the source agency's own website.
• AI Written by AI from the facts above, and always labeled.

Numbers on a sheet come from the Hub's data, never from the AI.

Ask's number cards can be exported as a CSV, as plain text or as an image. Each export copies the card: its figures, its fine print, the dataset's caveats and the source. The AI doesn't write any of it. A card can also be shared as a link anyone can open. It shows the card as it was the day it was shared; the server asks the question again to build it, so nothing on the page comes from the reporter's browser.

Ask can also say how long things take in a few City files, like complaints, 311 requests, permits and fire calls. The median is the middle case: half took less time. The “9 in 10” time is how long all but the slowest tenth took. Cases still open aren't counted, so the slowest ones are missing, and the card says how many. It can also say how long things still open have been open so far, like buildings on the City's vacant list or complaints not yet closed, measured to the day the data is as of. It also says how long City employees have worked there. In files that name people, Ask shows no group smaller than 10 and doesn't split them by year or date, so no one can be picked out.

Ask can describe one ZIP code in Milwaukee County: who lives there and how it compares with Milwaukee, from the U.S. Census Bureau's survey for that ZIP's Census area, marked CENSUS. When most of a ZIP lies outside the City, the card says so; the figures cover the whole ZIP.

For health, food, hospital visits and similar measures by ZIP code, Ask uses Health Compass Milwaukee, the Milwaukee Health Care Partnership's data site, marked HCM. Ask keeps a weekly copy, shows each figure's range, compares it with Milwaukee County, and credits Health Compass and the original source on every card, download and share.

Ask can also show the Fire Department's calls and response times year by year, 2015 to now. The City's yearly files are counted once a week into stored totals; from 2024 the City publishes from a new system, and the card says where the years don't line up exactly.

You can also ask about one neighborhood's census tracts, like “the poverty rate in Bay View's tracts.” Ask lists every tract with at least 10% of its area in the neighborhood, each with its own figure and range, and says how much of it is inside: “61% of this tract is in Bay View.” It never adds tracts together into one neighborhood number.

Each neighborhood Data You Can Use profiles has a fact sheet you can print: its headline figures, each beside the City of Milwaukee's from the U.S. Census Bureau's same survey and years. “Clearly higher” means the gap is larger than both margins of error together; otherwise the sheet says they're about the same. Code writes every figure and every verdict; the AI writes none of it.

You can type an address or intersection to find which census tract, City neighborhood and Data You Can Use area it's in. The address goes to the U.S. Census Bureau's geocoder to find its location; Cream City Almanac doesn't keep it, and never shows anything about the property itself.

The City's boundaries (aldermanic districts, voting wards and polling places, Targeted Investment Neighborhoods, NRSAs and NSPs, Tax Incremental Districts, Business Improvement Districts and Opportunity Zones) are copied from the City of Milwaukee's map server once a week. How much of a neighborhood or tract a boundary covers is estimated from a grid of points, so shares are about ±2%. Polling places can change before an election; confirm yours at myvote.wi.gov.

Crime counts reach back to 2005: the City publishes its crime records in two files, one from 2024 on and an older one, and a count that spans both reads them as one. Earlier years are left out because the City's older file holds only a few stray records before 2005.

Property questions can compare the City's yearly assessments from 2005, citywide and by aldermanic district, by reading one property file per year.

Stop-and-search questions count the people MPD stopped from 2019 to 2024, by race, sex and quarter. The Almanac reads the City's quarterly files once, stores only the counts, and shows frisk and search rates for field interviews.

Turnout questions compare 12 Milwaukee elections from 2016 to 2022, citywide or ward by ward, from the City's election files. The City's open data has no election results after 2022.

Ask can also add up a few City money columns, like back property tax owed, City employee pay and property sale prices. It shows the total and the median (the middle value), and the average only when you ask for it. In files that name people, groups of fewer than 5 records are hidden so no one's figure can be singled out.

Spreadsheet numbers appear as DYCU wrote them, Estimate ± margin of error, with any problems in DYCU's file noted above the table. No AI touches them.

HOW SEARCH WORKS

Type what you're reporting on, in your own words. Search matches meaning (text that means something similar) and exact words, then ranks what agrees. Try kids who can't afford food: it finds Food Insecurity Prevalence. Try Lincoln Park employment status by sex: it opens that neighborhood's table.

When nothing is close, it says so instead of listing unrelated data.

WHAT “NEIGHBORHOOD” MEANS HERE

Milwaukee has more than one map of its neighborhoods, and this site uses two.

The City's map. The City of Milwaukee draws 190 official neighborhoods. When Ask counts City records (crimes, fire calls, 311 requests), it keeps the records whose location falls inside the City's boundary, and the card says “City of Milwaukee boundary.”

DYCU's report areas. Data You Can Use publishes its numbers (poverty, rent, health) by census tract, a small area the Census Bureau counts, usually 1,200 to 8,000 people. For its Neighborhood Portraits, DYCU groups tracts into 27 report areas it chose with its neighborhood partners. Riverwest, for example, is tracts 71, 72, 79, 80 and 107. Some names are DYCU's own, like United Community, and aren't on the City's map.

Same name, different ground. Harambee on the City's map and Harambee in DYCU's reports (10 tracts) don't cover exactly the same streets. So a number never moves from one map to the other: a City count stays with the City's boundary, a DYCU number with DYCU's tracts, and every card names which one it used.

On the tract cards, each tract shows two names. Neighborhood is the City's: every official neighborhood covering at least 10% of the tract, biggest first (tap “+2” to see the rest with their shares). Each week the site lays a fine grid of points over every tract and counts which City neighborhood each point falls in. DYCU area lists every DYCU report area that includes the tract; some tracts are in two or three. A tract outside the City, or in no DYCU area, shows “—”; the site never guesses a name.

WHAT IT IS

An unofficial, phone-first way to find, understand and download Milwaukee's public data from Data You Can Use. Search in your own words; every fact shows where it came from.

WHERE THE DATA COMES FROM

DYCU's ArcGIS Hub, the public website where Data You Can Use publishes its data: 93 raw data · 282 reports · 7 visualizations, in the Hub's own words.

Refreshed weekly and never altered. Downloads come straight from the Hub.

HOW IT'S BUILT

Next.js on Vercel; Convex for the database, search and the weekly job; Claude through the Vercel AI Gateway; Firecrawl to read PDFs. A weekly automated check grades search on 30 reporter-style questions. Every trade-off is in the decision log.