Reform Models & Programs

What the Evidence Actually Shows

The strongest evidence base in correctional rehabilitation belongs to cognitive-behavioral therapy. Meta-analyses consistently put CBT's effect at a 20–30% reduction in recidivism against control conditions (#4832), and a 58-study meta-analysis by Landenberger and Lipsey found mean recidivism 25% lower in treatment groups than controls — a drop from a mean rate of .40 to .30 (#7383). The effect is not uniform: representative CBT programs show 20–30% reductions, with larger effects for higher-risk participants and high-fidelity implementation, and the smallest — though still significant — effects in routine prison implementations (#13637). Independent moderators of larger effects include treatment of higher-risk offenders, high-quality implementation, and inclusion of anger-control and interpersonal problem-solving components (#7384). Notably, brand-name curricula — Reasoning and Rehabilitation, Moral Reconation Therapy, Thinking for a Change — do not outperform generic CBT (#7385). The Risk-Need-Responsivity framework that organizes most of this work holds that services should target higher-risk individuals, address criminogenic needs, and match intervention style to individual learning style; low-risk people may actually be harmed by intensive programming (#4854), and best practice indicates high-risk individuals need multimodal, longer-term, intensive services while low-risk individuals need minimal exposure (#4969).

Education carries the largest and most replicated returns. RAND's 2013 meta-analysis of 57 recidivism studies found correctional education participants had roughly 43% lower odds of recidivating than non-participants, translating to a 30% versus 43% three-year rate — a 13-percentage-point absolute reduction (#7373; also #4845) — and 13% higher odds of post-release employment (#4846), an estimate RAND itself flags as resting on a single higher-quality study (#7374). RAND also found $4 to $5 saved per dollar invested (#1216), with savings from reduced reincarceration of $8,700 to $9,700 per participant over three years (#1219), against direct program costs of $1,400 to $1,744 per inmate (#1218). Bozick and colleagues (2018) put the reduction at 32% with 12% higher employment odds across 1980–2017 research (#13632). An American Journal of Criminal Justice 2023 meta-analysis of 152 estimates from 79 papers found vocational education returns $3.05 per dollar invested (#13634) and college education the largest per-student benefit at approximately $16,908 (#13635). The Vera Institute estimates expanding postsecondary education in prison would save states $365.8 million a year in prison costs (#13650; also #4849). Vocational education is associated with a 15.6% decrease in recidivism odds (#4850) and 28% higher employment odds (#4840).

Beyond classroom and cognitive programming, the evidence is more qualified but real. Mitchell, Wilson and MacKenzie's Campbell review of 74 evaluations of incarceration-based drug treatment found therapeutic communities consistently produced modest reductions in post-release recidivism and drug use, with 30 of 35 TC evaluations showing a statistically significant treatment effect (#7379) and average odds ratios of about 1.37 for recidivism and 1.28 for drug use (#7380). Trauma-specific curricula have a substantial published base — Seeking Safety alone has more than 40 outcome studies (#4839) — and trauma-informed care is universal across the successful California programs GPS examined (#4924); SAMHSA's six principles are Safety, Trustworthiness and Transparency, Peer Support, Collaboration and Mutuality, Empowerment/Voice/Choice, and Cultural, Historical and Gender Issues (#4858). Mentoring by people with lived experience shows promising short-term recidivism reduction, especially in juvenile justice, though findings are largely correlational and depend on relationship length and intensity (#7716). A small randomized trial of peer-mentored reentry (n=55) found significantly lower recidivism in the mentored group controlling for risk and demographics (#7391), and New York's Arches Transformative Mentoring reported felony reconviction rates 69% lower at 12 months and 57% lower at 24 months against a propensity-matched comparison group (#7392). Poorly implemented mentoring can be harmful (#4855), and Georgia-specific design proposals point to education, CBT, and structured peer-mentoring continuity from inside to outside as the three components carrying the heaviest meta-analytic evidence (#7419). Program quality can be assessed with the Correctional Program Checklist, which evaluates evidence-based delivery, risk targeting, criminogenic needs, CBT approaches, staff qualifications and outcome tracking; high-scoring programs are significantly more likely to reduce recidivism (#4976).

The caveats matter as much as the effect sizes. RAND itself flags that many underlying education studies have weak designs and that participants may differ from non-participants on unobserved motivation (#7376). The National Institute of Justice cautions that no reentry program yet meets the 'evidence-based' standard via replicated randomized controlled trials, and results are mixed (#7712); a DOJ meta-analysis of reentry programs found an average recidivism reduction of about 6%, with larger effects when treatment began in prison and continued into the community (#7710). Mental health and substance abuse treatment programming alone reduces recidivism by 6% (#2166). Almost all arts-program evaluations are observational, often small, and frequently program-reported; the strongest claim the arts literature can sustain is reduced in-prison infractions and improved engagement, not recidivism (#7390). The Good Lives Model, despite wide adoption, has not been confirmed: a 2021 systematic review screened 1,791 articles, found only 6 evaluable studies, all observational and none randomized, and concluded there is not enough peer-reviewed evidence to unequivocally confirm its efficacy (#7400); most rigorous GLM evaluation has been with men who had sexually offended (#7413). Attrition is the quiet failure mode: prison-based adult programs lose about 20% of participants and inpatient juvenile programs up to 60% (#4961), which is why dropout management — allowing re-entry after failure, early intervention, incentive systems, and peer accountability — is itself a design problem (#4977).

Georgia's Inventory: Programs, Beds, and Pennies

Georgia's official program inventory is short. The Department of Corrections' Reentry and Cognitive Programming unit lists Moral Reconation Therapy and Thinking for a Change as its core cognitive-behavioral programs across state prisons, private prisons, and the Metro Reentry Facility (#13705; also #13643). MRT is a 12-step, open-ended group format — participants enter at any time, progress at their own pace, and typically complete in three to six months (#4829) — supported by over 200 published outcome studies showing lower recidivism for periods as long as 20 years (#4827, #4828). Thinking for a Change is 25 lessons of one to two hours delivered to groups of 8 to 12 over 12 to 30 weeks, with a free scripted manual available through the National Institute of Corrections (#4826); a 2009 evaluation found 23% of participants recidivated versus 36% of controls at six months (#4825). Georgia also runs a 200-hour cognitive-behavioral program, Offenders Under Transition, delivered in three modules (#4958).

Statewide substance abuse treatment runs through Residential Substance Abuse Treatment programs. One GPS-tracked collection records 12 RSAT programs across state prisons and probation treatment centers (#70). A 2000 evaluation counted 7 RSAT programs at 4 prisons — Scott, Macon, Calhoun and Pulaski — with 310 beds (#19215, #19216), provided by Spectrum Health Services/CiviGenics (#19214). That evaluation found the program operating in a way reflective of its design and the intent of its designers, with a significant degree of fidelity between actual and proposed structure, setting, and content (#19217), and reported benefits cited by interviewees including lower rates of institutional misconduct and significant attitudinal and behavioral change (#19218). It also identified weaknesses in selection and referral processes and the lack of sufficient aftercare once graduates are released, including poor communication within GDC and among diagnostic and classification staff who did not know the program existed (#19219). That aftercare finding has not been retired: the 2024 Senate Study Committee record still attributes the highest recidivism among program participants to residential substance abuse treatment, which it attributes to addiction severity (#7560).

Georgia operates 12 Reentry and Cognitive Programming centers with 2,344 transitional beds, 346 designated for women (#13710; #4957), providing approximately six months of work release before parole (#13686) — a footprint GDC concedes is 'a limited amount of space,' with some people ineligible by criminal history (#13687). Three-year felony reconviction for transition-center releases in the FY2022 cohort was 11.4% against 33.7% for state prisons, per GDC data of July 9, 2025, though GPS notes selection effects are substantial (#13712). Postsecondary programs include the University of West Georgia's bachelor's program at Hays and Coweta (#13729); Life University's Chillon Project at Arrendale, the first Georgia-based degree program in prison since the 1994 Pell ban, whose first cohort of eleven graduated in January 2019 with an average GPA of 3.9 (#13730); Common Good Atlanta, privately funded with more than 100 professors from seven universities (#13732); and Georgia State University's Prison Education Project, opened in 2016, whose first nine graduates at Walker were awarded degrees on May 5, 2023 (#13733). Georgia State shut the program down in March 2024; it cost $180,000 a year and served 60 students (#5257). Federal Pell Grants for incarcerated students were restored on July 1, 2023, with programs expanding in 44 states (#5267); Georgia closed one. Central Georgia Technical College runs Transitional Career Centers in 13 correctional facilities (#4974). Newer inside programs include FORGE, a peer-led 36-week mentorship development program at Dooly State Prison launched in 2025, built on a self-sustaining cohort-trains-cohort pipeline (#13720), and the Forensic Peer Mentor program, run by the Georgia Mental Health Consumer Network with DBHDD and GDC since 2017 across five prisons, five day reporting centers, three mental health treatment courts, and two state hospital forensic units (#13718), which reported more than 550 people served in FY2017–2018 with two re-arrests, three psychiatric readmissions, and one revocation since inception — self-reported figures (#13719).

The money is the tell. Georgia's vocational education budget is $172,000 statewide for FY2025 — $3.44 per person (#5252). Prison education represents 0.11% of the corrections budget, roughly $40 per person per year or $0.11 a day, less than the cost of 45 commissary ramen packets (#5266). No dedicated line items for comprehensive reentry programming, transition planning, or post-release support services are visible in publicly available GDC budget documents (#2111). In the FY2027 budget, the Senate Appropriations Committee eliminated all funding for Metro Reentry programming, taking it from the Governor's proposed $39,786 to $0 (#5553), while Transition Centers received $50,067,051 in total public funds ($49,847,398 State General plus $219,653 Agency) (#5573) and the approved budget added $32,637,565 to the physical health contract and $12,127,034 to the mental health contract to increase staffing ratios (GPS budget dataset, HB 974 Senate Appropriations Committee Substitute). Georgia does operate an inmate Performance Incentive Credit program: completing rehabilitative programming can move a tentative parole month earlier, with each PIC point equal to one month (#8662; #11548), a mechanism the Board describes as preparing offenders for success after release and improving utilization of prison beds at a cost savings to taxpayers (#11672).

Georgia's Own Evidence: Cognitive Skills, Parole Programs, and a Pattern of Routinization

Georgia's most rigorous test of a rehabilitation model was homegrown. The Board of Pardons and Paroles began implementing Reasoning and Rehabilitation in 1997, training parole officers to act as cognitive coaches (#16295; #19224), making it the first known statewide effort to employ cognitive skills training with parolees (#16294). The Georgia Cognitive Skills Experiment used discrete time survival analysis, controlling for relevant background characteristics, and found statistically significant treatment effects of 17% on returns to prison and 19% on re-arrests for class completers compared to controls in Phase I (#16342, #16343). The intent-to-treat picture was weaker: the 4% lower return-to-prison and re-arrest rates for the experimental group as a whole were not statistically significant (#16344). The program was administered as designed — rated high on adherence to modeling, reinforcement, practice, and interpersonal sensitivity (#16561), which the evaluators called exceptional in correctional programming history — and completers' reported use of taught skills rose from 71.0% at time 1 to 94.3% at time 5, a change researchers verified was not an artifact of attrition (#16512). Completion, though, was the binding constraint: 59.6% in Phase I (#16281) and 62.7% in Phase II (#16279), ranging from 33% to 100% across groups (#16282) with a mean of 72% among those who actually began (#16283). Groups ran from 6 to 28 members with a typical class of 11 (#16553) — well above the four-to-six members the curriculum specifies (#16326; #16384) — and observers rated 'general issues and class structure' less than satisfactory for 41.3% of groups at session 17 (#16286), though coaches met expectations on prosocial interpersonal skills in 78.3% of groups at session 17 and 91.1% at session 30 (#16288, #16289).

Two findings from the Georgia experiment cut against the field's orthodoxy and against one another's policy implications. The Phase I outcome study found that low-risk parolees also benefited from R&R (#16391), and analysis did not support the hypothesis that high-risk participants would benefit more than low-risk participants (#16390) — a direct tension with the Risk-Need-Responsivity principle that services target higher-risk individuals and that low-risk people may be harmed by intensive programming (#4854). The same Phase II report noted that if the 299 low-risk offenders in the experimental group were counted alongside those excluded for low cognitive ability (8%) and sex-offender status (3%), over half the experimental group (54%) could be considered inappropriate for the program — while acknowledging that Phase I found low-risk offenders benefited (#16548). Screening criteria nonetheless excluded parolees with IQ scores below 80 and those with histories of sexual offending or debilitating substance abuse, and targeted high-risk offenders for inclusion (#16325; #19230). The Board's operational version was a 60-hour cognitive skills program covering anger management, decision-making, and communication (#13493); separate meta-analytic work puts R&R's average reoffending reduction at about 14% (#4831), and Georgia's own FY2001 annual report cited an 18% recidivism reduction for class graduates measured against control groups (#11109; #11128).

Scale has always been modest. Georgia trained 30 parole officers to teach R&R and 22 coaches to lead Thinking for a Change in FY02 (#13495); 456 parolees graduated from Parole-delivered cognitive-skills programs that year, not counting those mandated to anger management and parenting classes run by community providers (#13497). In FY04, 389 parolees graduated from cognitive skills programs (#13109) and 1,925 completed substance abuse programs (#13110). By FY05, 1,050 parolees participated in cognitive skills programs with 135 active coaches, all parole officers or chief parole officers (#10702), and by FY07 the agency had trained 186 officers as MRT facilitators through three in-house classes of 25 per year (#12664; #12931). Substance abuse treatment volume ran far higher than cognitive programming: 10,449 parolees were referred to drug treatment in FY2011 and 10,268 enrolled (#9954, #9955); 7,740 were referred and 7,905 enrolled in FY14 (#9622, #9623); and 7,726 Recovery Groups were conducted in FY14 alone (#9629). Reported completion rates were 67% (FY2011), 64% (FY2012), 66% (FY13), and 60% (FY14) (#9957, #10337, #12496, #9625). No Turning Back, a five-month, 60-slot intensive addiction program for higher-risk, higher-need parolees at the Whitworth Parole Center (#10124; #10913; #10828), reported 142 completions against only 2 non-completions in FY06 (#10829) — a figure that sits awkwardly beside later years, when the same program reported 44 unsuccessful discharges against 154 graduates in FY14 (#9632, #9631), 63 discharges against 142 graduates in FY16 (#9012, #9011), and 23 discharges against 74 graduates in FY13 (#12503, #12502). The Board's FY2002 annual report claimed that each month a parolee attended programs, the likelihood of arrest declined 2%, producing a 24% reduction for a year of attendance (#13534, #13535) — a self-reported figure with no described design.

Georgia's reentry-side programs have published counts, and Georgia's earlier experiments counsel caution about how such initiatives age. M.O.R.E. (Max-Out Reentry), for inmates within 12 to 18 months of their maximum release date, including people statutorily ineligible for parole (#9681; #9856), placed 494 inmates in FY14 (#9593), with 326 active as of June 30, 2014 (#9594) and 241 successful completions released from transitional centers at the end of their court-ordered terms (#9600); 42 offenders previously denied parole were granted parole after referral through M.O.R.E. in FY14 (#9598). In FY13, 367 inmates passed through the program in the seven months it operated, 71 failed and were returned to a secure prison, and 46 maxed out (#12472, #12471, #12470). Reentry Partnership Housing placed 488 parolees with 28 providers in FY14 (#9611, #9610), paying providers $600 per month for up to three months — a maximum of $1,800 per parolee (#9615) — with 86% of placements since FY11 either on active parole or successfully discharged (#9614) and $1,966,316 in cost avoidance generated in FY13 (#12484). The Georgia Prisoner Reentry Initiative, launched in November 2013 around 'Getting Ready, Going Home, and Staying Home,' set a stated goal of a double-digit recidivism reduction (#9707, #9861, #9709). But Intensive Probation Supervision — implemented in 1982 in 13 of 45 judicial circuits and expanded to 33 circuits by the end of 1985 (#13971, #13972, #13973) — supervised 2,322 probationers, of whom 1,952 were successfully diverted from prison and 370, or 16%, absconded or were revoked (#13994, #13979, #13978); drug offenders posted a 90% success rate, better than under regular probation (#14043). The statewide Youthful Offender Program fared worse over time: one-year return-to-prison rates rose across exit cohorts — roughly 7%, 12%, and 13% for FY74, FY75 and FY76 (#15479) — and its vocational rehabilitation clients went from 10% to 13% to 20% one-year return rates over the same cohorts, described in the evaluation as 'perhaps the most alarming' change (#15480), with specialized programs 'losing ground' relative to the rest of the prison population in that age group (#15483). The evaluator's conclusion was structural, not individual: the program had become routinized in a traditional prison mold that significantly limited its effectiveness (#15519), and its contracting process was oriented to whatever programs the prisons already ran rather than to assessed need (#15415). The same arc appears in the most recent cycle. H.B. 265 in April 2011 formed the Special Council on Criminal Justice Reform for Georgians (#14364), and the 2011–2016 justice reinvestment work — driven by the Georgia Council on Criminal Justice Reform with technical assistance from Pew and the Texas Public Policy Foundation (#8344) — produced a 6% prison population reduction and $57 million reinvested into accountability courts, substance abuse treatment, and community supervision without increasing crime rates (#2112, #2114, #2115); commitments to prison and probation for offenses affected by H.B. 1176 fell 10%, from 28,604 in 2012 to 25,784 in 2015 (#14330). But the Council's influence waned after Governor Deal left office in 2019 (#3891).

National Models: California, Texas, Maine, Michigan, and the Attribution Problem

California is the largest test case. Under the California Model, peer mentorship sits alongside normalization, dynamic security, and trauma-informed care as one of four pillars (#4899). Across the state's successful programs, the credible messenger model — using formerly or currently incarcerated facilitators — is identified as a key design principle (#4923), trauma-informed care is universal (#4924), and self-directed options such as in-cell CBT are credited with reaching people in restrictive housing who cannot access group programs (#4933). Reported outcomes are striking and largely program-reported: Project Rebound, the nation's longest-running prison-to-college pipeline, founded in 1967 at San Francisco State (#4893), reports a 0% recidivism rate among its students against California's 41.9% general rate (#4888), a system-wide 3.0 GPA since 2016 (#4889), a 71% three-year graduation rate (#4890), and 87% of graduates securing employment or postgraduate admission (#4891), across 19 CSU campuses (#4892). GRIP, a 52-week coursebook-based program rooted in restorative justice, trauma-informed care, and neuroscience at San Quentin (#4881), reports 421 graduates released since 2012 with only 2 returning to prison — a 0.5% recidivism rate (#4878, #4879, #4880) — and was the subject of a 2025 peer-reviewed study (#4882). The Last Mile, operating 17 classrooms across 8 states with six-month web development and audio/video production tracks (#4886, #4887), reports a 75% employment rate among graduates (#4883). Hustle 2.0, an in-cell CBT program co-authored by incarcerated people at Pelican Bay (#4911) with 1,266+ pages across 32 courses (#4910), reports 91% of participants avoiding further violence, 93% with no further misconduct, and 82% not returning to restrictive housing, at 85% less cost than group-facilitated CBT (#4912, #4913, #4914, #4915); participants can earn college credit from Southern Utah University (#4932). Certification pipelines exist too: California's Offender Mentor Certification Program requires about a year of coursework, a state exam, 255 hours of practicum, and 2,000+ supervised hours (#4962), with a 4,000-hour supervised counseling internship in Phase 3 (#4895), after which graduates serve as paid co-facilitators and hold portable certification (#4898). San Quentin's redesign into a rehabilitation center — vocational hubs, podcast studio, farmer's market, self-serve grocery — carries an approximately $239 million price tag and is scheduled to open in January 2026 with capacity for upwards of 2,500 people (#6065, #6066, #6067, #6068). Staff buy-in remains, per the Sacramento Bee, the biggest obstacle to rollout (#6079), the state correctional union has offered only guarded support (#6078), and some officers allege the new freedoms created more dangerous situations (#6080).

Texas's justice reinvestment reforms invested $241 million in evidence-based programs and swift, graduated sanctions (#1221), cut the parole revocation rate 46% (#1222), and avoided $3.00 billion in projected prison construction and operating costs (#1224). Its Prison Entrepreneurship Program, a Baylor-certified in-prison Mini-MBA founded in 2004 (#4980, #4981) that reaches over 6,000 men across 80 TDCJ units annually (#4939, #4940) and runs 128 participants per class (#4947), reports about 7% of graduates returning to prison within three years against 23% of Texas male releasees, from a highly selective intake of roughly 2,000 applications and 500 acceptances a year (#13645); 90% of PEP staff are program graduates (#4946). Maine reorganized its entire correctional system starting in 2022 around normalization, humanization, and destigmatization, building on two decades of incremental reform (#4717), and reports a three-year recidivism rate falling from 30.5% in 2017 to 21.4% in 2022 (#4695), a 40% reduction in resident-on-resident assaults (#4696), a 36% reduction in assaults on staff (#4697), a 69% reduction in staff use-of-force incidents (#4698), an 84% decrease in self-inflicted injuries at Maine State Prison (#4699), nearly 50% of its incarcerated population on medication-assisted treatment for opioid use disorder (#4702), and education from GED through master's degrees via the University of Maine (#4706). Michigan's Vocational Villages — three facilities, roughly 600 students, 13 trade programs, open to people within 12 to 24 months of release (#4722) — saw 2019 graduates recidivate 6.5 percentage points below the state rate (#7755). Per-inmate education spending benchmarks frame Georgia's position: Florida spends about $1,028 per person per year with 21% recidivism (#5296); Texas about $508 to $585 with 15% recidivism (#5298); South Carolina runs a dedicated prison school district awarding approximately 8,300 credentials a year (#6138); Mississippi, the poorest state in the nation, enrolls 80% of its 10,549 inmates in programming and spends $4 million or more on prison education (#5299).

Vera's Restoring Promise initiative has established 9 dedicated housing units in 6 states since 2016, pairing adults 25 and older as mentors with young adults aged 18 to 25 (#4959), and published an implementation toolkit in January 2024 (#4721). A randomized trial at South Carolina sites found a 73% reduction in the odds of violent incidents (#7756) and an 83% reduction in the odds of restrictive-housing stays (#7757), with the design controlling for custody level, education level, pre-treatment outcomes, time in study, race, and age — and no significant difference between applicants who did not get spots and non-applicants, meaning outcomes are attributable to the program rather than self-selection (#4677). GPS's own cross-collection comparison flags an attribution conflict worth stating plainly: the same 73% and 83% figures appear in another collection attributed generically to 'South Carolina's programming reforms' for violence write-ups and restrictive-housing stays (#5025, #5026). The underlying numbers are identical; the causal attribution to one program is not established by the second source. That is the normal condition of this literature. Nordic-inspired pilots report early, mostly non-causal results, as researchers themselves stress (#7775), and the Nordic comparison carries a transferability caveat: those outcomes reflect much shorter sentences, far smaller populations, and stronger social-welfare baselines, establishing that measurable alternatives exist rather than that they transplant one-to-one to Georgia (#7678).

Several design principles recur across the programs with credible results. Most successful prison programs use a cohort model rather than rolling enrollment, because cohorts create peer accountability and shared identity; rolling enrollment works for ongoing support and senior-to-junior mentoring models (#4968). Mentor development takes a minimum of 6 to 18 months and requires demonstrated behavioral change, ranging from formally certified models (California's OMCP) to organic ones (San Francisco's RSVP) (#4970). Programs that work with corrections administration show the greatest sustainability (#4971), and identity-affirming language is treated as a success factor — Defy Ventures deliberately calls participants 'Entrepreneurs-in-Training' (#4972). Reported results include RSVP's up to 80% reduction in violent rearrests at 50 hours a week intensity (#4944, #4943), Bard Prison Initiative's recidivism rate under 4% with 400 students enrolled full-time across seven New York prisons (#4950, #4941), and Defy Ventures' 85% employment within six months of release (#4948) and under-10% one-year recidivism against a 40%+ national average (#4949). Gang exit is the weaker link. Texas's GRAD and RP-GRAD programs offer a nine-month, three-phase renouncement pathway and a six-month return pathway for gang-validated offenders who would otherwise be automatically placed in restrictive housing (#3687, #3719), and the National Institute of Justice called step-down and gang-exit programs a positive move toward jointly reducing gang influence and overuse of restrictive housing (#3698). But NIJ also noted that no program to date has been rigorously shown to permanently remove people from prison gangs (#3699), and the leading academic study of Texas GRAD explicitly does not seek to answer whether renunciation works — no rigorous recidivism or before/after violence evaluation exists (#7668). Georgia has no structured gang renouncement or exit program at all (#3715); its state-level gang strategy is prosecution, not correctional management, with GDC's internal work limited to intelligence gathering and reactive lockdowns (#3709).

Step-Down Programs: Reform Inside Segregation

Georgia's step-down model is the Segregated Transition Education Program, embedded in a Tier Segregation System GDC says was implemented in August 2013 'to create a managed and measured stratification-oriented pathway for offenders to transition from Administrative Segregation to lower-security levels successfully' (#20803). Tier II STEP policies were completed in FY2019 (#20782). An offender exiting Administrative Segregation review may be assigned to STEP (#20802), and the Tier III Special Management Unit Review Committee can recommend STEP assignment to the Warden following its at-least-every-90-days review of an offender's status (#20804). The Tier III Program is a minimum 13-month (390-day) sequence requiring at least 60 days in E Wing, 60 in F Wing, and 90 each in D, C, and B Wings if the person is successful at every phase (#20784); the Special Management Unit at Georgia Diagnostic and Classification Prison runs a five-phase Tier III program, level five being the most restrictive, with privileges increasing for demonstrated compliance (#20808). Successful completion leads to consideration for reassignment to Tier III STEP with Close Security status (#20787). Smith State Prison is designated a Tier I & II facility with a STEP DOWN program (#20796); Hays State Prison is listed with a Step Down Program alongside its Tier I & II designation (#20799).

The 2025 Tier III Special Management Unit policy sets out a floor that reads, on paper, like reform: single-occupancy cells furnished consistently with general population, hygiene items on the same basis as general population, showers and shaves three times a week, food identical to general population with substitutions capped at seven days, the same bedding, laundry, and barbering, a minimum of one hour of outdoor recreation daily Monday through Friday, and — the operative number — 'a minimum of four (4) hours out of cell time per day, Monday through Friday,' composed of table time, recreation, out-of-cell programming, or visitation but including at least one hour of recreation; legal visits do not count, and weekend visitation cannot be used to reduce weekday out-of-cell time (#19281). Tier II STEP, written in April 2019, guarantees four hours out of cell per day seven days a week and general-population privileges — but expressly counts 'shower, feeding ... etc.' toward the four hours, so the four hours are not four hours of association or programming (#19283). Tier III STEP is textually near-identical to Tier II STEP, with one substantive difference: all cells are single occupancy, where Tier II permits double occupancy — meaning whether a person in a Georgia step-down unit is held alone or with a cellmate is determined by which tier they stepped down from (#19284). The one accountability mechanism in the corpus is narrow and specific: if out-of-cell time is canceled or shortened for three or more consecutive days, the SMU Warden must promptly notify the North Region Director in writing, explaining the circumstances, the steps taken, and an estimated timeline for resuming, and denial of out-of-cell time as punishment is prohibited unless the person committed a 'Great' or higher severity disciplinary offense (#19282). No equivalent requirement exists for Tier I, Tier II, or the juvenile program (#19282).

GDC reports that since January 2017 the number of offenders in extended restrictive housing decreased by 41% — 671 people — achieved partly through STEP, along with Specialized Protective Custody housing, Specialized Mental Health Treatment Units, administrative reviews, changes in the discipline process, and policy restructuring (#19298; #20780). The shape of that disclosure is itself a finding: GDC reports the size of the decrease and the percentage but never states the population before or after, and the Commissioner's letter attributes the same 41% reduction to FY2018 alone while the body text attributes it to roughly eighteen months since January 2017 (#19298). GDC's published table does not split its 119 STEP beds between Tier II STEP and Tier III STEP, leaving the tier composition of step-down capacity unknown (#20854). The seven-facility Tier I and II designation rests on GDC fact sheets whose most recent editions GPS holds are from 2018 and 2019; material changes have occurred since, including the closure of Georgia State Prison at Reidsville on February 19, 2022 (#19366; #6599).

Other states provide the comparison. Virginia developed its Restrictive Housing Reduction Step-Down Program in 2011 at Red Onion and Wallens Ridge, creating a pathway from Security Level S, the most restrictive, to lower security levels (#20827); the proportion of people in the program with a mental health diagnosis dropped from 47% in February 2016 to 32% in July 2017 (#20829). Oregon's randomized study of the Step Up Program against the Intensive Management Unit found 10.0% of the SUP group maintained an MH-R mental health classification compared with 53.9% of the IMU group (p < .001) (#20825). North Carolina's RDU program operates with roughly 500 participant capacity at Marion Correctional Facility since 2016, aimed at reducing prison violence among people who would otherwise be in restrictive housing for control purposes and providing a structured rehabilitative pathway to safe reintegration (#20824). California's Ashker settlement ended indeterminate solitary confinement, ended gang affiliation alone as a basis for SHU placement, capped continuous Pelican Bay SHU stays at 5 years, required immediate review of everyone held more than 10 years, and created a four-step, two-year step-down program (#3721, #6667); by June 2015 CDCR had conducted 1,274 reviews of SHU inmates, releasing or endorsing release to general population for 910 and placing 321 in the Step-Down Program (#3693). Arizona's Security Threat Group program produced declines of over 50% in assault, drug violations, threats, fighting, and rioting among segregated gang members, and a system-wide 30% reduction in total violations that researchers estimated may have prevented as many as 22,000 rule violations (#3682, #3683, #3684) — but among gang members who formally renounced membership, disciplinary violations decreased by amounts that were not statistically significant, suggesting renouncement needs robust post-exit support and monitoring (#3714). An Ohio study found that gang affiliates leaving restrictive housing fared worse afterward, engaging in higher levels of misconduct, suggesting that separation without programming and structured reintegration may simply delay rather than prevent violence (#3713).

The Delivery Gap: DOJ Findings, a Named Death, and Where the Money Goes

The reform models above describe what works. Georgia's delivery system describes what happens instead. The DOJ's investigation found 142 or more homicides in Georgia prisons from 2018 to 2023, a figure it noted was likely an undercount due to misreporting (#1614). It found that educational and vocational programming had been 'slashed rather than expanded,' and that conditions in most facilities were so chaotic and violent that meaningful programming participation was effectively impossible (#2125; #5256). Understaffing was the mechanism: 'Understaffing affects programs... prisons do not have enough staff to prevent or even respond to the most blatant gang activities, let alone provide programs' (#5261). DOJ recommended 82 remedial measures and noted the Attorney General may sue under CRIPA (#8088). GDC denies the violations and claims DOJ 'misunderstands' prison operations (#1631). The exception proves the rule: among the 17 prisons DOJ visited, Walker State Prison had fewer people reporting they feared for their lives, a much higher proportion of security staff positions filled, more rehabilitative programming, and no reported homicides in several years (#6658; #13717).

Programming failure is not abstract. Marquis Reshawn Jefferson died May 11, 2022, at Washington State Prison; GPS's registry lists the cause as homicide and records his age at death as 26. Reporting and documents obtained by his brother established that the prison was so understaffed that no one was watching the dorm when he was attacked in an incident involving multiple incarcerated people, and that no one noticed until other incarcerated people carried his body to the door (#1308). The staffing collapse DOJ documented is the same resource base on which every program in this page depends.

Georgia's corrections budget has grown even as its population fell. State general funds for the department rose from $1.07 billion in FY2020 to $1.42 billion in FY2024, with another $100,652,921 in other funds in FY2024 (Governor's Budget Reports, Amended FY2024 and FY2025, and Governor's Budget Report, Amended FY2026 and FY2027 — GPS budget dataset). The FY2027 approved budget adds $32,637,565 to the physical health contract, $12,127,034 to the mental health contract, and $1,498,347 to the dental contract to increase staffing ratios, while reducing funds for opioid abuse programs by $6,094,804 in State Prisons and $2,178,619 in Detention Centers and shifting those functions to Opioid Settlement Trust Funds (HB 974 Senate Appropriations Committee Substitute — GPS budget dataset). The programming line that disappeared entirely was Metro Reentry, cut to $0 from a proposed $39,786 (#5553). For scale, an estimated phased post-conviction counsel program serving 1,000 to 2,000 cases a year would cost $3 million to $10 million — 0.17% to 0.56% of GDC's $1.78 billion budget (#4337). Nothing in the budget record contradicts the DOJ finding on programming; the line items are simply absent (#2111).

Oversight Models and the Data Gaps That Block Evaluation

The oversight models that would catch any of this exist elsewhere. Across jurisdictions, the levers that work are consistent: independence from the corrections agency, statutory access to records and people, confidentiality of communications, and time-bound public reporting — retaliation is sustained by opacity (#6562). Deitch's effectiveness checklist for oversight bodies requires independence and externality, a routine inspection mandate, unfettered 'golden key' access including unannounced entry, adequate resourcing, public reporting, varied information-gathering methods, a matched agency duty to respond, and a focus on treatment, health, safety, and civil rights (#8054). The Federal Prison Oversight Act of 2024 created an independent DOJ Inspector General inspection regime and a Bureau of Prisons ombudsman with public reporting to Congress (#7986); models discussed in Georgia's 2024 Senate study committee included Indiana, New Jersey, Pennsylvania, and Virginia, where ombudsmen independent of corrections agencies typically report to attorneys general, governors, and legislatures (#7581). Hawaii's five-member commission uses a capture-resistant appointment structure with one member each appointed by the Governor, Senate President, Speaker, Chief Justice, and the Office of Hawaiian Affairs (#8064). Virginia's model makes grievance data a statutorily mandated, published data stream held by a body independent of the corrections agency (#7984). Georgia's own history shows what court-supervised reform can look like: over thirteen years, Judge Alaimo's orders in Guthrie v. Evans mandated changes in virtually every aspect of prison operations, including rehabilitation and educational programs, medical, dental, and mental health programs, classification, disciplinary due process, grievance procedures, and exercise (#4752).

Georgia already has an internal audit standard that could support evaluation. GDC SOP 107.14 (Office of Reentry Services Audit Process, effective March 2, 2022) requires Quality Assurance Evaluations confirming programs are 'delivered in the manner in which it was designed,' including completeness of documentation and group participation, plus a Georgia Program Assessment Inventory measuring compliance with evidence-based principles at least every two years, with Corrective Action Plans for critical findings due within 30 days; audit records are retained five years on the GDC intranet (#13707). Whether any of those audits have been released, or any Corrective Action Plan produced, is not established in the record GPS holds.

The data gaps are concrete. There are no dedicated reentry line items in publicly available GDC budget documents (#2111), which means the state's reentry investment cannot be tracked against its reentry claims. Program counts conflict across time and source: one collection records 12 RSAT programs (#70) while a 2000 evaluation counted 7 at four prisons (#19215). GDC's STEP bed table does not split 119 beds between Tier II and Tier III STEP (#20854), and the Tier facility designations rest on fact sheets from 2018–2019 (#19366). Georgia's own cognitive skills experiment contains an unresolved internal tension between an intent-to-treat result that was not statistically significant (#16344) and a completer analysis showing 17% and 19% effects (#16342), and between a finding that high-risk participants did not benefit differentially (#16390) and the field-standard RNR position that they should (#4854). None of these gaps are reasons to discount the evidence that exists; they are reasons the state cannot say with confidence which of its programs work.

What the comparative record suggests for Georgia is not exotic. The Washington State Institute for Public Policy's benefit-cost model ranks correctional education — especially postsecondary — as the highest-return adult corrections program, with CBT, vocational education, and correctional industries showing positive net benefits (#13638). Georgia's own data already shows the same gradient: vocational completers recidivate at 13.64% against a 26% general rate (#3229), and vocational programming in particular cuts recidivism, in the Senate Study Committee's own framing, by half (#7560). The state's foundational audit rule already requires fidelity measurement every two years (#13707). What is missing between those two facts is not evidence or statutory authority — it is the money, the staff, and the independent reporting that would let the state act on either.