Industries / Government, education and community / Government / Corrections
Running a facility safely
Running a facility safely means keeping a prison or jail secure and orderly while protecting the people who live and work in it: how the building is laid out, how staff supervise, how violence and abuse are prevented, and how heat and crowding are managed.
6
Milestones achieved, from 1829 to 2015.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1829
Isolation by design
Eastern State Penitentiary in Philadelphia opened in 1829. The British architect John Haviland gave it a radial plan, seven cellblocks around a central hub, built for the "Pennsylvania System": solitary confinement with labor, each person alone in a cell. Charles Dickens visited in 1842 and called the system "cruel and wrong". Eastern State officially abandoned it in 1913, and Pennsylvania stopped running the prison in 1970.
LessonA building designed around one idea of punishment can outlive the idea; Eastern State's cells stayed in use for 57 years after solitary confinement was given up.
Credit
John Haviland and the Commonwealth of Pennsylvania
1971
Attica's four days
On 9 September 1971 people held at Attica Correctional Facility in New York took over a yard and held 42 guards and civilian workers hostage. Their demands included better medical care, sanitation and food, more visits and an end to beatings by guards. On 13 September state police retook the prison, and 39 prisoners and hostages were killed by police gunfire; 43 people died in all. Officials first said the hostages' throats had been cut by prisoners; the Monroe County coroner found they had been shot.
LessonThe first official account of a prison crisis can be wrong, and only an outside examination corrected it at Attica.
Credit
the Monroe County coroner, whose findings contradicted the first official account
1970s
Officers inside the unit
In direct supervision, an officer works inside the housing unit among the people held, instead of watching from a control booth behind glass or bars. The US Federal Bureau of Prisons developed it in the early 1970s and tested it in three short-stay jails, its Metropolitan Correctional Centers in Chicago, New York and San Diego; counties followed from the late 1970s. A 1989 study for the US Department of Justice by Jay Farbstein and Associates surveyed 52 facilities and visited 7. Direct supervision facilities reported fewer violent incidents in a year, about 13 against 32, a difference at the edge of statistical significance, and cost the same or less to build, staff and run.
LessonThe same study found facilities' own incident records too poor to compare with confidence, so the safety case rested partly on surveys of staff and the people held.
Credit
US Federal Bureau of Prisons; measured by Jay Farbstein and Associates
2010
Halden's open design
Halden Prison in southeastern Norway, a maximum-security prison holding about 250 people, opened in 2010. A 25-foot wall surrounds it, but there are no guard towers, guns or razor wire; each person has a private room, and staff eat meals and play sports with the people held, according to an NPR report from 2015. Building safety on relationships between staff and the people they hold is called dynamic security, and the UN's Nelson Mandela Rules (rule 76) list it among the subjects every prison officer must be trained in.
Credit
the Norwegian Correctional Service and Are Hoidal, Halden's governor
2012
Surveys count sexual abuse
The US Prison Rape Elimination Act of 2003 required national surveys and standards. In the third National Inmate Survey, run from February 2011 to May 2012 in 233 prisons, 358 jails and 15 other facilities, 92,449 adults answered anonymously; 4.0% of people in state and federal prisons and 3.2% in jails reported sexual victimization by another person held or by staff in the past year. The Justice Department's standards, effective 20 August 2012, require every covered facility to be audited once every three years, and a state that does not comply can lose 5% of some Justice Department grants.
Credit
Bureau of Justice Statistics (Allen J. Beck) with RTI International; US Department of Justice
2015
Isolation limited to 15 days
The UN's first Congress on the Prevention of Crime and the Treatment of Offenders, held at Geneva in 1955, adopted the Standard Minimum Rules for the Treatment of Prisoners, a set of minimum conditions for prisons. On 17 December 2015 the UN General Assembly adopted a revised version, named the Nelson Mandela Rules (resolution 70/175). Rule 44 defines solitary confinement as 22 hours or more a day "without meaningful human contact", and rule 43 prohibits indefinite solitary confinement and prolonged solitary confinement, meaning more than 15 consecutive days.
Credit
United Nations General Assembly; UN Office on Drugs and Crime
Watch for
Air conditioning Texas prisons
On 22 September 2026 US District Judge Robert Pitman ordered Texas to air-condition all 104 of its state prison facilities by the end of 2029, after finding that the Texas Department of Criminal Justice's own logs showed indoor temperatures above 85 °F nearly every day from 1 May to 30 September in units without air conditioning. As of 1 October 2026 just over a third of the prisons were fully climate controlled, leaving about 90,000 people without air conditioning. The department says it expects 90,000 air-conditioned beds in 2028 and puts the full cost at $1.5 billion; it has appealed, and the judge refused to pause his order on 10 November 2026.
Achieved whenevery housing area in all 104 facilities is cooled and a court monitor or other outside party confirms it from indoor temperature records through a full summer.
Claimed by
Texas Department of Criminal Justice, November 2026
An AI violence estimator
The UK Ministry of Justice said on 31 July 2025 that a "violence predictor" analyzes factors such as a person's age and past involvement in violent incidents, so staff can step in or move people before violence escalates, and that an AI tool had analyzed 8.6 million messages from 33,000 phones seized in prisons. On 24 September 2026 it said its Violence in Prisons Estimator is live in prisons. Neither statement gives an accuracy figure; the ministry says all its AI tools are tested for accuracy, bias and fairness before wide use.
Achieved whenan outside body publishes the estimator's error rates, overall and for different groups of people, and whether violence fell where it was used.
Claimed by
UK Ministry of Justice, July 2025 and September 2026
Open
01
A world prison heat record
A public, daily record of hazardous heat at every prison outside the United States. Take prison locations from OpenStreetMap, where volunteers tag prisons as "amenity=prison", compute each site's daily maximum wet bulb globe temperature (WBGT, a heat-stress measure that combines temperature, humidity, sun and wind) from the ERA5 global weather reanalysis, and count the days each year above 28 °C, the threshold the main US study used. It uses public locations and weather only, never information about any person.
The briefCredit · unclaimed02
Indoor heat from outdoor weather
A model that predicts the afternoon temperature inside prison housing from outdoor weather and simple building facts, fitted to the indoor temperature logs that prison systems have released and tested on prisons held back from fitting. Applied to the US heat dataset, it would turn outdoor exposure into an estimate of indoor exposure. The logs record buildings, not people.
The briefCredit · unclaimed
Open 01 · the brief
A world prison heat record
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, daily record of hazardous heat at every prison outside the United States. Take prison locations from OpenStreetMap, where volunteers tag prisons as "amenity=prison", compute each site's daily maximum wet bulb globe temperature (WBGT, a heat-stress measure that combines temperature, humidity, sun and wind) from the ERA5 global weather reanalysis, and count the days each year above 28 °C, the threshold the main US study used. It uses public locations and weather only, never information about any person.
Why it is still open
Tuholske and colleagues (Nature Sustainability, 2024) did this for 4,078 prisons, jails and detention sites in the continental United States, and published their code and daily values, but stopped at the US border. OpenStreetMap held 15,613 objects tagged as prisons on 18 November 2026. Searches found no worldwide equivalent.
Built with what exists
OpenStreetMap data through the Overpass API or Geofabrik extracts; ERA5 hourly data from the Copernicus Climate Data Store; Python with xarray; the US study's published code (GitHub: sparklabnyc/temperature_prisons_united_states_2024) as the template.
First step
Download the OpenStreetMap prisons for India, Brazil and Nigeria and compute one summer of WBGT for each site.
Done looks like
Daily WBGT for 1990 to 2025 at at least 5,000 sites in at least 100 countries, with each site checked against satellite imagery; for 200 US sites the yearly count of hot days agrees with the US study's values within 10%; data and code public.
Open 02 · the brief
Indoor heat from outdoor weather
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A model that predicts the afternoon temperature inside prison housing from outdoor weather and simple building facts, fitted to the indoor temperature logs that prison systems have released and tested on prisons held back from fitting. Applied to the US heat dataset, it would turn outdoor exposure into an estimate of indoor exposure. The logs record buildings, not people.
Why it is still open
The US study converted outdoor heat to indoor heat with an assumed formula (fixed wind, no radiant heat), and its authors call for direct measurement of indoor temperatures. Measured logs are now public. Reuters obtained indoor temperature logs from five states (California, Pennsylvania, Texas, Utah and Washington) through public records requests and published them in 2025; students at the University of Texas at Austin analyzed the 3 p.m. readings that Texas takes each day from April to September in units without air conditioning, for 2022 to 2025; and the September 2026 Texas ruling rests on the same kind of log. A 2026 heat-risk index for California's 31 state prisons, built for a University of California, Berkeley master's project, uses each prison's ratio of indoor to outdoor hot days but does not predict indoor temperatures. No model fitted to these logs and tested on prisons held back from fitting was found.
pmc.ncbi.nlm.nih.govgithub.comutdata.github.iomarybecica.com
Built with what exists
Reuters' published logs and air-conditioning records; hourly weather from NOAA's Integrated Surface Database; building age and type from prison systems' unit directories; Python with scikit-learn, or a building simulation in EnergyPlus.
First step
Download Reuters' Texas and Washington logs and pair three units' readings with their nearest weather stations.
Done looks like
Predictions for at least 20 housing units in at least two states, none used in fitting, with a mean error of 1.5 °C or less in the afternoon indoor reading over a summer; model, data and code published.