Nason/ice-detention-management-database

ICE detention-management statistics

Frozen Datapond snapshot: 22 data tables, 58,810 stored source rows, 10,498,048 bytes. The catalog identifier requires separate registry integration. Row totals mix table units and are not unique people, events or places.

Table	Rows	Unit	Time semantics
dtm_adp_and_stay_length_by_agency_fy_ytd	1,551	FY-to-date average daily population (people) and mean stay length (days)	snapshot
dtm_adp_and_stay_length_by_agency_monthly	1,164	Monthly average daily population (people) and mean stay length (days)	latest_cell
dtm_atd_by_aor	14,657	Active alternative-to-detention participants and average days in program by AOR/technology; nullable technology is a subtotal	snapshot
dtm_atd_court_appearances	462	Appearance count and percentage by hearing type and outcome; includes totals	snapshot
dtm_atd_population	1,099	Active participants; technology rows report daily cost USD, status rows report ALIP days	snapshot
dtm_bond_stats	330	Counts, percent, USD average bond, or ALOS days, distinguished by unit	latest_cell
dtm_book_ins_by_arresting_agency_fy_ytd	387	FY-to-date count of book-ins by ICE/CBP/Total	snapshot
dtm_book_ins_by_arresting_agency_monthly	282	Monthly book-in count by ICE/CBP/Total	latest_cell
dtm_book_outs_by_reason_fy_ytd	4,521	FY-to-date final release/book-out count by reason and criminality	snapshot
dtm_book_outs_by_reason_monthly	2,606	Monthly final release/book-out count by reason and criminality	latest_cell
dtm_currently_detained_by_criminality	1,536	Detained population count and arresting-agency share within each criminality category (fraction; source rounding retained)	snapshot
dtm_currently_detained_by_disposition	1,738	Current detention population count by processing disposition and facility type	snapshot
dtm_facilities	20,430	Facility-by-release characteristics: beds, threat composition, inspection information; not the geocoded facility directory	snapshot
dtm_facility_alos	2,045	Annual mean stay length (days) for named facility	latest_cell
dtm_fear_decision_time	21	Time from USCIS fear decision to ICE release (days), fiscal-year/facility-type series	latest_cell
dtm_fear_decisions_by_facility_type	348	Current detained count in fear-decision source table by facility type	snapshot
dtm_iclos_and_detainees	1,665	Mid/end-month population count by ICLOS/detainees section and duration bucket	latest_cell
dtm_movements_by_facility_type	1,875	FY-to-date book-in/book-out counts by facility type; book-outs only have Total criminality	snapshot
dtm_removals	128	FY-to-date removal counts including family-unit removals; not individual-removal events	snapshot
dtm_segregation	1,383	Monthly segregation-placement count per facility; placements are not unique people	latest_cell
dtm_special_population_actions	468	Annual action counts for special populations; nullable country applies only to country tables	latest_cell
dtm_vulnerable_population	114	Quarterly segregation placement counts and average consecutive/cumulative days for vulnerable populations	latest_cell

Source, scope and limitations

This frozen source has maximum file_date July 20, 2026 and pull_date July 9 or 11, despite an October 27 repository commit. It is not current November statistics. Observation periods, source extraction dates, workbook publication dates and commit dates are different concepts.

Twelve tables retain release snapshots; ten retain the latest reported cell for their series key. Never sum FY-to-date snapshots, stock populations across dates, averages, or totals together with components. Totals coexist with detailed rows, and missing categories are not zero. Use the declared series dimensions and a single pull-date snapshot for comparisons. fiscal_year can describe the workbook; data_fiscal_year, alos_fiscal_year or date identifies the measured period when present.

Units vary among counts, USD, days and fractions. share_of_total divides each arresting agency's count by the combined ICE+CBP population within that criminality category and pull date. It is not the criminality composition within an agency. All 1,024 nonmissing shares match this denominator within 0.005, reflecting source rounding. The bond_pct series is labelled percent upstream but stores fractions (observed 0.0234378 to 0.279019); ATD court pct values are also fractions (observed about 0.00543 to 1). Convert to percentage points explicitly for display. Seven upstream ATD average-length-in-program values are negative (minimum -7); they are preserved as source anomalies, not meaningful negative durations. Flag or exclude them explicitly for duration analysis.

The upstream parser skips some corrupted sheets and omits zero ICLOS/bond cells as presumed placeholders. These omissions are retained; absence is not a measured zero. dtm_facilities is facility-by-release statistics, not the geocoded directory. No name-based join to microdata or the directory is supplied.

All 21 source-defined candidate keys are unique under NULL-aware grouping; dtm_facilities has no verified natural key and retains _source_row. All source rows/columns are preserved, including anomalies. Independent checks find no file_date before pull_date or observation date after pull_date. ICE plus CBP equals Total in the inspected 129 YTD and 94 monthly book-in series.

Every source byte hash, immutable URL and repository commit is in source-lock.json. Every row is retained; _source_row is added as an ordinal within the pinned member. Geometry, when present, is stored as WKB in geometry_wkb, with all original metadata in _geometry_metadata. _provenance, _metadata and _columns document the build. DICTIONARY.md records types, exact NULL counts, units and time semantics. Original source flags and text are not coerced into new numeric categories.

Publisher: Deportation Data Project. Data license: CC0-1.0, as labelled by the publisher; see the retained license notes. Cite government data collated by the Deportation Data Project and the pinned source commit. Repository commit dates do not promise current observations or automatic refreshes.

Local queries

import duckdb
con = duckdb.connect('ice-detention-management.duckdb', read_only=True,
                     config={'memory_limit':'128MiB','threads':'1'})
try:
    print(con.sql("SELECT fiscal_year, pull_date, n_book_ins_ytd FROM dtm_book_ins_by_arresting_agency_fy_ytd WHERE arresting_agency='Total' QUALIFY row_number() OVER (PARTITION BY fiscal_year ORDER BY pull_date DESC, file_date DESC)=1 ORDER BY fiscal_year DESC LIMIT 3").fetchall())
finally:
    con.close()

library(DBI)
con <- dbConnect(duckdb::duckdb(), dbdir='ice-detention-management.duckdb', read_only=TRUE)
tryCatch({
  dbExecute(con, "SET memory_limit='128MiB'")
  dbExecute(con, 'SET threads=1')
  print(dbGetQuery(con, "SELECT fiscal_year, pull_date, n_book_ins_ytd FROM dtm_book_ins_by_arresting_agency_fy_ytd WHERE arresting_agency='Total' QUALIFY row_number() OVER (PARTITION BY fiscal_year ORDER BY pull_date DESC, file_date DESC)=1 ORDER BY fiscal_year DESC LIMIT 3"))
}, finally=dbDisconnect(con, shutdown=TRUE))

Public datapond identifier ice-detention-management requires a future approved catalog release. The smoke scripts use an isolated localhost registry, never a global registry edit.

Reproduce and refresh

Use Python 3.11 or newer in a dedicated environment. Install the exact tested dependencies with python -m pip install -r requirements.txt (or use the pinned uv source in pyproject.toml), then:

python acquire_sources.py
python build_database.py --validate-only --report fresh-validation.json
python build_database.py --output fresh-ice-detention-management.duckdb

Downloads are pinned, bounded and SHA-verified; existing source files are checked and reused. The importer preserves all source fields and compares every stored row with the input using symmetric EXCEPT ALL. It uses one thread, 128 MiB DuckDB memory, no spill, a 32 MiB working-file cap and a 512 MiB physical free-space reserve. A watchdog interrupts work when a budget is crossed. Builds are private until complete; hard-link publication refuses an existing final database, manifest or orphan WAL path. Synchronous checks enforce the working-file cap and physical reserve after close and before publication; hash checks remain enabled under optimized Python. Failed/private files remain for inspection, with no cleanup of unrelated work. Peak process memory can exceed DuckDB's buffer limit.

To refresh, review the publisher's changed files and definitions, update pins and expected schemas/counts, run full source checks and Python/R smoke tests, and audit both data and documentation. No scheduled or unattended refresh is configured.

Source transformation semantics were checked against the pinned publisher parser. Facility composition and inspection fields retain changing source text layouts; no unsupported numeric scale or crosswalk is imposed.

Validation commands: python -m unittest -v test_safety.py and python run_smoke.py --python /path/to/python --rscript /path/to/Rscript --r-library /path/to/R/library. The latter needs the released Python and R datapond packages and an already installed HTTPFS extension. It uses only localhost and this database. See VALIDATION.md for the executed versions and results.

Snapshot files

Download the database, browse its files, or inspect the reproducible source. Verify the snapshot SHA-256 b29a199fe88141973912af8b62a3c4d26dc0bf51f7ccc097a528176e385e9201 (10,498,048 bytes). The registry PR pins immutable Hugging Face revisions; a later main may refer to a newer snapshot.

Once the separate catalog entry is merged, use the released clients:

import datapond
con = datapond.connect("ice-detention-management")
try:
    print(con.sql("SELECT fiscal_year, pull_date, n_book_ins_ytd FROM dtm_book_ins_by_arresting_agency_fy_ytd WHERE arresting_agency='Total' QUALIFY row_number() OVER (PARTITION BY fiscal_year ORDER BY pull_date DESC, file_date DESC)=1 ORDER BY fiscal_year DESC LIMIT 3").fetchall())
finally:
    con.close()

library(datapond)
con <- dp_connect("ice-detention-management")
tryCatch(print(DBI::dbGetQuery(con, "SELECT fiscal_year, pull_date, n_book_ins_ytd FROM dtm_book_ins_by_arresting_agency_fy_ytd WHERE arresting_agency='Total' QUALIFY row_number() OVER (PARTITION BY fiscal_year ORDER BY pull_date DESC, file_date DESC)=1 ORDER BY fiscal_year DESC LIMIT 3")), finally=DBI::dbDisconnect(con, shutdown=TRUE))

Until catalog integration, the local Python/R examples above work with the downloaded snapshot. The smoke scripts exercise both released clients with an isolated registry and this same database.