What is counted
The source is the project file of HUD's 2025 Housing Inventory Count. Each Continuum of Care lists every project in its area that holds beds for people experiencing homelessness, on one night in late January. The file lists 30,012 projects across every state and territory.
Year-round emergency shelter beds summed from the project file equal HUD's own published table for all 386 Continuums of Care.
The count of people comes from HUD's 2025 Point-in-Time Count, taken on the same night. County populations are Census Bureau estimates for 2024.
The parts of a project record
| Part | Values | What it decides |
|---|---|---|
| Project type | Emergency shelter, safe haven, transitional housing, permanent supportive housing, rapid re-housing, other permanent housing | Whether the bed is a place for tonight, a stay of months, or an exit from homelessness |
| Housing type | Single site, clustered sites, tenant-based | Whether the address in the record is the building or an office |
| Bed type | Facility, voucher, other | Whether the bed is in a shelter or a hotel room paid for by voucher |
| Household | Families with children, adults, children only | Who can use the bed |
| Dedicated beds | Veterans, youth, people experiencing chronic homelessness | Beds held for one group |
| Availability | Year-round, seasonal with dates, overflow | Which nights the bed exists |
| HMIS | Participating, comparable database, not participating | Whether the project's occupancy is visible to its Continuum of Care |
| Funding | ESG, CoC, HUD-VASH, SSVF, VA GPD, RHY, HOPWA, HOME and others | Who pays, and whose rules set eligibility |
Emergency shelter and safe haven beds are the shelter side. Transitional housing sits between. Permanent supportive housing, rapid re-housing and other permanent housing are the exits. A county panel lists all three, because a match needs to know both where a person can sleep tonight and where they can go next.
Where a project is placed
| Class | Projects | On the map |
|---|---|---|
| Single building | 13,504, of which 12,378 are placed | A pin at the address, located with the Census Bureau geocoder. An address the geocoder matched loosely is marked approximate. |
| One of several buildings | 2,544, of which 2,289 are placed | A pin at the listed building, labeled as one of several. |
| Scattered-site or voucher | 9,342 | No pin. The address in the record is the operator's office. The beds are counted in the county HUD assigns them to. |
| Location withheld | 3,478 | No pin and no address. Counted in county totals. |
| No address in the file | 1,144 | No pin. Counted in county totals. |
A project's county comes from its located building first, then from HUD's own geographic code, then from its ZIP code. County lines are the Census Bureau's 2023 boundaries, so Connecticut appears as its nine planning regions and Puerto Rico as its municipios. Streets and place names come from OpenStreetMap.
What is never shown
The location of a victim service project is never published here. Three rules apply. A project flagged as a victim service provider, or dedicated to survivors of domestic violence, has no address, city, ZIP code or pin. An operator whose projects are mostly victim service projects has no location published for any of its projects. A building that holds one withheld project is not published for the other projects inside it. A check runs before every publish and stops it if any served file carries a location for a withheld project.
Disaster response
The January 2025 count fell four months after Hurricane Helene. FEMA shelter and hotel assistance open that night was filed as emergency shelter, 4,036 beds in North Carolina. Those beds are listed and labeled as disaster response. They are left out of standing capacity.
Reading a county
Shading is year-round emergency shelter beds per 10,000 residents. A hatched county is one where HUD's inventory lists no year-round emergency bed. In North Carolina that is 30 of 100 counties.
The third figure on a county is the share of year-round beds held by the largest operator. In Orange County one operator holds 89 percent.
The occupancy bar on each project is the number of people in it on the count night. In Orange County that night, the family shelter at HomeStart had 12 of 32 beds in use, and 48 people in the county were unsheltered. A bed is open only to the household it is set up for, which is why supply has to be read by type.
How to ask for a bed
In North Carolina, 68 counties carry the way a person asks for a bed there: the coordinated entry line, the walk-in site and its hours, or the shelter door. Each line was read on the operator's, the county's or the Continuum of Care's own page and carries the date it was read. A county is shown only where those pages agree with each other.
Open beds tonight
The public record of occupancy is one night a year. A person looking for a bed, and the outreach worker beside them, need tonight. The map carries a live state for every emergency shelter, and the state has three values.
| State | Meaning |
|---|---|
| Open, with a number | The shelter sent a count in the last 18 hours and it has that many beds free. |
| Full | The shelter sent a count in the last 18 hours and it has no beds free. |
| Not reported | No count in the last 18 hours. |
A count holds for 18 hours. One sent at evening intake still stands the next morning and is gone before the next evening. An old count is never carried forward, and silence is never read as open or as full. Every count is kept with its time and its source, so a correction is a new count and the history can be audited.
Three ways to produce the count
| Instrument | What it measures | What leaves the building | What it needs |
|---|---|---|---|
| Staff count | Beds free at intake, as the shift lead sees them | Two numbers per project: open and offline | A phone and thirty seconds. The report page is built and running. |
| Check-in scan | Beds assigned and released through the night | The same two numbers, computed inside the shelter from scans of an anonymous bed card | A card per bed night, a scanner at the desk, guest consent and review by an institutional review board |
| Bed or door sensor | Presence in a bed or passage through a door | A count | Hardware in a sleeping space, upkeep, resident consent |
The staff count comes first. It costs nothing, it works in a shelter that does not enter data in HMIS, and it sends no information about a guest. In North Carolina 2,054 of 6,872 emergency beds sit in projects outside HMIS, so a tracker built inside HMIS cannot see them. Los Angeles runs live bed inventory inside its HMIS and reaches the projects that participate.
The check-in scan is a research instrument. With a card that changes every night it counts beds and nothing else. With a card that stays the same it measures length of stay and movement between shelters, which is the data a study of pathways out of homelessness needs. That version is client-level data and belongs to a university team under its own review board, held on its own systems.
Sensors serve as an audit. A few beds at a few sites show how far the staff count drifts from the truth. The network itself runs on the staff count.
Two errors, measured separately
A bed reported open that is full sends a person across town to a closed door. A bed reported full that is empty leaves a bed unused and nobody learns of it. The pilot measures each rate on a sample of nights against the shelter's own log, by household type, and publishes both.
Sequence
- Orange County. Two buildings run by the Inter-Faith Council hold 118 of the county's shelter and transitional beds. A staff count from those two desks covers most of the county's supply. Four weeks, with response rate and both error rates reported at the end.
- The neighboring counties, Durham and Wake.
- The Balance of State counties, where 2,225 people were unsheltered on the count night and the nearest listed bed can be a county away.
- A check-in scan at one consenting shelter, designed and run by the university team.
Sources: HUD 2025 Housing Inventory Count and Point-in-Time Count, Census Bureau county population estimates for 2024, Census Bureau geocoder. Built by Tom Ciszek.