Hotel rooms

The inventory contains 24,116 published rooms at 79 visitor-hotel locations: 53 SF Travel meeting hotels and 26 individually sourced sites. That is roughly 68% of a separate 2024 benchmark of 35,527 rooms, although the sources cover different dates. Counts come from operators, owners, official reports and accommodation guides; the inventory links each property to its source.

Rooms determine the area of each blue dot, so a large hotel counts more than a small one. Suites are counted once. The Inn at the Presidio’s main building and four-room Funston annex have separate dots with non-overlapping room counts. Publication dates vary, and 2024 operating status has not been certified for every property.

SROs and known shelter or residential capacity are excluded. Cova is a shelter; BEI’s residential/hotel split remains uncertified; Donatello is a timeshare. Smaller and outlying hotels are still underrepresented, which may exaggerate downtown concentration. An area without blue dots may still contain hotels.

Most locations are Census address estimates; the Presidio properties and Seal Rock Inn use operator map markers. Points near boundaries can change areas if coordinates are corrected. These are hotel capacities, not occupancy figures or visitor itineraries: guests may walk into neighboring areas or spend much of their stay elsewhere. The overlap can reveal geographic differences, but cannot establish visitors’ exposure or explain the city’s reputation by itself.

Unsheltered people counted

People counted unsheltered on 30 January 2024, including vehicles, tents and streets. The January 2026 HSH Shelter Equity Analysis supplies neighborhood counts. Its 40 rows sum to 4,338, versus 4,354 citywide; 16 people are unallocated in this table. Lincoln Park is absent and shown as missing, not zero. The map uses 4,338 for mapped-count shares. This one-night estimate is not current street conditions.

The Point-in-Time count estimates people sleeping in unsheltered locations on one night. It is not a count of everyone who experiences homelessness during a year, everyone seen on the street, or people in shelters. A snapshot can miss people; it should not be read as a complete, current census.

The map retains the published Analysis Neighborhood totals because finer person counts have not been obtained. It does not divide them into invented census-tract estimates or distribute the 16 unallocated people across neighborhoods. The 2026 publication date does not make these 2026 observations.

Colors show people counted per square kilometre, using each neighborhood’s original polygon area. They compare that density with the average across areas with known counts; they do not show homelessness as a share of local residents. Use the separate February tents and vehicles layers to examine observed objects at finer geographic detail, while remembering that objects and people are different measures from different dates.

Residents’ views

The 2023 City Survey interviewed 2,530 adult residents in October–December 2022. This map uses the City’s published weighted neighborhood averages directly. City services ratings combine assessments of services, including safety; they do not measure overall happiness with living in San Francisco. Separate safety questions ask about walking alone in one’s own neighborhood by day and at night.

Ratings are mapped to respondents’ home neighborhoods, not the locations of services visited. Results appear for 34 neighborhoods. The seven identified by the City as having fewer than ten responses are hatched, including Seacliff, whose safety value remains in the public model. No tract-level survey values are inferred.

Neighborhood sample sizes and confidence intervals are unavailable in the aggregates, so small differences are not evidence of statistical significance. The City notes that 21% of responses citywide were collected in person, compared with 82% in Chinatown; this likely raised Chinatown’s service ratings. These historical opinions do not measure incident rates, personal risk or current conditions.

Scores run from 1 to 5, with higher ratings more positive. Safety runs from very unsafe to very safe; 3 means neither safe nor unsafe. These are average ratings, not percentages of residents who feel safe. Safety and service maps use red for lower ratings and green for higher ratings. Service ratings run from failing (1) to excellent (5); 3 means average. The six score bands are defined under Geography & comparison. Official citywide safety averages are 3.68 by day and 2.96 at night. No verified citywide average is available here for the service composite.

Tents, structures & vehicles

The February 2024 visual count records 259 tents and 135 structures, a combined 394. A separate layer shows 621 apparently lived-in vehicles: 134 passenger vehicles and 487 other vehicles. These are objects observed on the street, not counts of people. February is the recorded observation month, not a verified single count day.

The city coordinates these counts quarterly. February 2024 is the closest available snapshot to the January 30 Point-in-Time count, but the two sources measure different things on different dates. Tents and structures exclude vehicles; the vehicle layer keeps them separate. Neither can be converted into a person count by assuming how many people occupy each object.

Published locations are approximate. We assign the original points directly to census tracts, then divide object counts by Census land area. All 368 source rows have coordinates and count fields. The rows describe observed sites, not a complete register of every surveyed and unsurveyed street. A tract with zero recorded objects is not proof that nobody was sleeping outside.

Counts from different quarters are not added together: the same object may appear in several snapshots. This extract was retrieved in September 2026, and its February totals differ from some contemporary press summaries. The map retains the reproducible source totals and labels the historical period; it does not present them as current conditions.

Police reports

The police layers cover January–December 2024. Drug enforcement combines “Drug Offense” and “Drug Violation”; assault and robbery are combined; optional theft uses “Larceny Theft,” a broad property-theft category. These are recorded reports, not convictions or a measure of personal risk. Drug reports reflect enforcement effort as well as underlying activity, so more reports can mean more policing.

Each case is counted once per layer. Where a case has several records, we use its latest available geocoded record by data-as-of timestamp, breaking ties with the numeric row ID. A case can still appear in different layers if it contains different offense categories; adding the layers would not produce a count of unique cases. A later record may also place an incident differently from an earlier one.

DataSF shifts locations to nearby intersections for privacy and notes an intersection-mapping change around April 24, 2024. Points are assigned directly to the original census-tract polygons, using the same boundaries as the housing maps. No cases are spread across tracts or assigned through grid-cell centers. Boundary proximity and changing location estimates make very local comparisons uncertain.

Colors show reports per square kilometre of tract land, not reports per resident. They indicate where reports concentrate, without accounting for how many workers or visitors spend time there. Compared with the neighborhood map, one additional assault/robbery case and one theft case fall outside the displayed tract boundary.

Observed street & sidewalk litter

Controller’s Office and Public Works evaluators inspect sampled street segments. These layers use the official dashboard’s weighted averages for July 2023–June 2024, keeping the FY24 comparison period despite newer FY25 results. They measure historical observed litter, not 311 complaints, resident satisfaction or current conditions.

The 24 reporting groups combine 36 Analysis Neighborhoods. Each group is drawn as one shape; its score cannot establish conditions in individual neighborhoods or blocks. Golden Gate Park, Lincoln Park, McLaren Park, Presidio and Treasure Island lack published estimates and are hatched. Sampling accounts for residential and commercial streets; raw inspection rows are not simply averaged. The FY24 report describes 2,662 representative evaluations. Group sample sizes and uncertainty are unavailable in the aggregate extract, so small differences should not be treated as meaningful.

Both scores run from 1 to 5: 1 means no litter; 2, a few traces; 3, more than a few traces without accumulation; 4, distributed litter with some accumulation; and 5, widespread litter with significant accumulation. Green means less litter, red more. These are absolute scores, not ratios to the SF average; Geography & comparison defines the six display bands. Separately recorded dumping, graffiti, feces, syringes and other hazards are excluded. There is no combined cleanliness score.

The dashboard model was refreshed in October 2025. Some figures differ from the December 2024 annual report for reasons the source does not explain. This map consistently uses dashboard values, including citywide scores of 2.36 for street litter and 2.55 for sidewalk litter, rather than averaging neighborhood averages.

Geography & comparison

Police reports, tents, vehicles, population and housing use 241 census tracts. Unsheltered people and resident ratings use 41 Analysis Neighborhoods; litter inspections use 24 larger reporting groups. Each layer keeps the detail its source supports. Tracts are statistical areas, not necessarily named neighborhoods. Familiar place labels and subdued OpenStreetMap tiles provide orientation.

Tract shapes come from DataSF’s 2020 mapping, matched to the 2020–2024 ACS and 2024 Census land areas. Water-only tracts and the unpopulated Farallones tract are omitted. Original polygons determine point assignments; simplified shapes are used only for display. Tract and neighborhood boundaries differ slightly, so totals can differ.

Density is count divided by area, not residential population. The city benchmark is the total mapped count divided by the corresponding area with known data—not an average of area rates. Missing values are excluded from both totals. Parks and sparsely populated areas still affect the benchmark; small counts and different area sizes can produce sharp color changes that do not represent sudden changes on the street.

The six relative bands are much below (<0.5×), below (0.5–0.8×), around average (0.8–1.25×), above (1.25–2×), much above (2–4×) and far above (≥4×). The first three are green, then yellow, orange and red. Colors describe each measure, not a neighborhood’s overall quality. Green does not mean safe.

Safety, services and litter instead use absolute 1–5 scores, grouped as 1–2, 2–2.5, 2.5–3, 3–3.5, 3.5–4 and 4–5. All bands include their lower bound; only the final band includes its maximum. Hatching means missing data. Different dates, measures and geography prevent combining these layers into one score.

Residents & housing

The 2020–2024 American Community Survey provides five-year estimates of population (B01003), home value (B25077), gross rent (B25064) and tenure (B25003). The 241 mapped tracts contain an estimated 830,235 residents, including people in group quarters. These are multiyear residential estimates, not a September 2026 count or the number of people present during the day.

Home value is reported by owners, not measured from recent sales. Of 241 tracts, 217 have estimates and 29 are capped at $2m+. Missing estimates are hatched. The city median comes directly from the county estimate, not an average of tract medians. Homeownership is the share of occupied homes owned by their occupants: 38.2% citywide.

Gross rent includes estimated utilities and describes renter-occupied homes paying cash rent. It is not current asking rent. Values at or above $3,500 are capped as $3,500+. Six color bands run from green for lower rents to red for higher rents: below $1,500, $1,500–2,000, $2,000–2,500, $2,500–3,000, $3,000–3,500 and $3,500+.

Population density, home value and ownership share use purple: higher is not inherently better or worse. Rent colors describe cost, not affordability relative to income. None of these colors measures safety, and tract medians do not describe every resident’s circumstances. ACS estimates have 90% margins of error, often large for individual tracts. Small differences between neighboring areas may reflect sampling uncertainty rather than a meaningful difference in living conditions.