The Detention Ledger — Who Runs the Beds

Explore every facility

The map below shades each state by the number of people held on an average day. Filter by inspection status, Fail rating, paid-empty beds, or whether an operator is named; click any facility for its full record and operator chain.

Loading the facility map… The full dataset is at /detention/index.json.

Who runs the beds

The question ICE's file does not answer is who operates each facility — most carry only a county or agency label, even when a corporation runs them. This ledger fills that in only where the federal record itself does. Of 203 facilities, 7 are attributed to a named operator today; the rest default to the honest finding.

Coverage is bounded by design: operator subcontracts between a county and a corporation are not in federal award data, so most facilities resolve to “no private operator identified” and stay there until a federal record names one. The deeper dig lives in the writing series, never asserted above its evidence tier.

What the record shows about oversight

Every figure below is computed directly from ICE's own cells — no editorializing, no outside source.

• 61 facilities carry no inspection result on record in ICE's file — 38 of them held under US Marshals inter-governmental agreements.
• 5 facilities carry a Fail rating and were still holding detainees on this file.
• 45,621 guaranteed-minimum beds are contracted across the system; 20 facilities ran more than 10% below their guaranteed minimum — 8,813 beds guaranteed and paid for while sitting empty — even as 36 others ran more than 10% over.
• 73.1% of the people held carry no ICE threat-level classification in the file.

Fail-rated, still holding people

The largest facilities (top 15 by average population)

All 203 facilities, per-fiscal-year presence, and full flags ship in the keyless JSON.

Contraction and rebuild, not a straight line

The system did not simply grow. ICE's files show 213 facilities in FY19, 135 at the FY21 pandemic trough, and 203 in FY26 — a contraction followed by a rebuild. Any statement about expansion on this page is anchored to the FY21–FY26 rebuild, not to a false FY19 baseline.

Method & caveats

• Source: ICE Detention Statistics, fiscal years FY19–FY26 (ice.gov), a US government work in the public domain. All eight source files were archived with SHA-256 checksums before parsing; the raw spreadsheets are never republished (their file metadata carries contractor employee names).
• A blank rating means no inspection result appears on record in ICE's file of this date — phrased throughout as “no result on record,” never “never inspected.” Inspection PDFs are linked, never parsed.
• Operator attribution is never asserted above its evidence tier. Federal award (T1) and SEC-filing (T2) attributions are self-attested by the government or the company; deeper resolution of county-labeled facilities lives in the writing series with per-claim citations, not in this dataset.
• Average daily population is year-to-date and summed from ICE's four criminality columns; it is a population measure, not a capacity or a headcount on any single day.
• Regenerate: python3 scripts/build_detention.py against a fresh archive of ICE's files; per-fiscal-year column whitelists fail the build on any schema change.

Machine access

The full dataset — metadata, integrity figures, and all 203 facility records with per-year panels and evidence-tiered operators — is one keyless fetch:

import requests
d = requests.get("https://ai-analytics.org/detention/index.json").json()
named = [f for f in d["facilities"] if f["operator"]]
print(len(named), "of", d["meta"]["facilities"], "facilities attributed to an operator")

Also listed in the Voidly datasets manifest and /data. License: public domain source; this compilation CC0.