The Beds and the Badges: Two ICE Files, One Map

ICE publishes two spreadsheets that almost nobody reads together. One is a price list for custody: the detention file, 208 facilities and 62,517 people held on an average day. The other is a signature ledger: the 287(g) file, 2,123 agreements deputizing 1,804 local agencies into immigration enforcement, 1,994 of them signed since January 2025. Join them at the state level and they draw one map.

Where both boards lead

Texas holds the most people (16,450 on an average day, 27 facilities) and signed the most agreements (463, across 378 agencies). Florida is second on signatures and fourth on custody. In 86 percent of cases, a person detained on an average day is held in a state where at least one local agency has signed on to enforce immigration law:

Top ten states by average detained population; both files, joined on state. Full tables ship in each dataset's keyless JSON.

The corners where one file goes empty

The map's corners say as much as its center. California holds the third-largest detained population in the country — 6,412 people across 8 facilities — and appears exactly zero times in the 287(g) file: California law has barred its agencies from signing Section 1357(g) agreements since 2017 (SB 54, the California Values Act). Washington and New Jersey show the same shape — over a thousand people held on an average day each, zero signed agreements, both under state laws that restrict local participation in federal immigration enforcement. The beds are there; the badges are not.

West Virginia inverts it: 38 agreements signed by 38 agencies, and not a single ICE detention facility in the state's rows of the detention file. Badges, no beds. Pennsylvania is the sharpest recent case of the wave itself: 108 agreements — 108 distinct agencies, no agency holding two — every one signed since March 2025, 105 of them the street-level task-force model, in a state holding 2,172 people across 7 facilities.

What the joined map does not say

The discipline that governs both datasets governs their join. The detention file records contracts and custody; the 287(g) file records signatures and models. Neither records what any deputized officer did, whom they arrested, or whether a single 287(g) encounter put a single person in a single bed — and so neither does this article. The two files describe the same eighteen months from two directions — capacity bought, authority signed — and the state-level correlation is geography, not a causal pipeline. Where the record is silent, the map stays silent.

What the join does establish, in the government's own numbers: the buildout has a shape. It is heaviest where beds and badges stack in the same states, absent where state law forecloses one side of it, and — in the newest signatures — moving fastest in states like Pennsylvania where the badges arrived before any expansion of the beds.

Check the arithmetic

Both files ship as keyless JSON — every figure here recomputes from two fetches, no key, no rate limit: /detention/index.json and /287g/index.json.

Sources: the Detention Ledger (FY26 year-to-date per ICE file dated 2026-07-20) and the 287(g) Wave (ICE participating-agencies file, workbook last modified 2026-07-10), both CC0, zero personal data. One detention facility carries a blank State cell in ICE's own file (0 average daily population) and is excluded from the state join — the defect is preserved, not repaired. Naming rules: data standards.