Brief written October 25, 2026 from GPS Intelligence System data.(DS)

The Segregation Georgia Does Not Count

Georgia operates one of the largest and least documented long-term segregation systems in the country, and it does so under its own written rules. The architecture is called the Tier system: Tier I, a short-term segregation program authorized at every facility type in the state; Tier II, long-term administrative segregation with no maximum duration written into policy; and Tier III, the Special Management Unit at the Georgia Diagnostic and Classification Prison in Jackson, the only piece of the system a federal court has ever forced open. Around those three sit disciplinary isolation, protective custody, moderated and stripped cells, and a juvenile program modeled on the adult one.

The documentary record assembled here — GDC's own standard operating procedures, its PREA audit corpus, its annual fiscal reports, federal court orders, the Justice Department's October 2024 findings, and the state's own bed-count production to Georgia Prisoners' Speak (GPS) in October 2026 — supports a set of findings that are unusual in how completely they rest on the agency's own words. GDC has published no count of the people it holds in Tier segregation for at least five consecutive reporting years. Its Tier II policy, unrevised since 2016, sets a floor of five hours out of cell per week and no ceiling on duration. The protections won in federal litigation attached to the one unit that was litigated and to no other part of the system. And the deaths that occur in these units — named below, one by one — are recorded by a state that has stopped publishing the manner of death.

What follows is an account of how the system is built, what the courts have found, what the state's own data does and does not show, and who has died.

The Three-Tier Architecture and the Policy That Has No Ceiling

GDC's segregation system was implemented in August 2013 as what the agency calls a "stratification-oriented pathway" from administrative segregation back toward lower security levels. Its governing policies are unusually explicit, and they are the spine of everything that follows.

Tier I, governed by SOP 209.07 effective April 30, 2015, is defined as "short-term segregated housing" and applies to "All State Prisons, County CIs, Private Prisons, Intensive Treatment Facilities, Transitional Centers, Probation Detention Centers, and Probation Boot Camp housing GDC offenders or probationers." It carries eight placement grounds, and only one of them presupposes a pending disciplinary charge; none requires a disciplinary conviction. A person can be placed in Tier I because a warden notes them as "a threat to the safe and secure operation of the Facility," because they are under investigation, or simply because they "ha[ve] not been classified due to their arrival at a Facility." Tier I is capped at thirty days — except for protective custody, voluntary or involuntary, for which the cap is disapplied entirely.

Tier II, governed by SOP 209.08 effective April 11, 2016, is the long-term program, and it is where the policy record becomes stark. The SOP's own introduction states that the program "is an offender management process and is not a punishment measure." Its out-of-cell floor is five hours per week of exercise, with no daily distribution specified. It contains no maximum duration. A person who "cannot return to general population because they pose a serious threat to the safety and security of the institutional operation shall remain in Phase 3," reviewed every ninety days — a review that may simply confirm retention. As GPS's analysis of the policy record puts it, this is indefinite administrative segregation as a matter of written policy.

The contradiction is documented inside GDC's own rulebook. SOP 209.01, the discipline policy, provides in a section whose stated premise is a finding of guilt that where "isolation time imposed exceeds thirty (30) days due to being necessary to control and encourage behavioral change, a recommendation for placement of the offender in the Tier II program is warranted." The policy that says it is not punishment is the designated route around the thirty-day limit on punishment.

Tier III, the Special Management Unit, is governed by SOP 209.09, revised effective April 23, 2025 and again effective June 18, 2026. It is a minimum thirteen-month, five-phase program — sixty days in E Wing, sixty in F Wing, ninety each in D, C and B Wings, which sums to the stated 390 days. It sets a general 24-month ceiling, subject to six enumerated retention criteria, and it guarantees single-cell occupancy, four hours out of cell per day Monday through Friday including at least one hour of recreation, daily recreation, showers three times a week, general-population food and bedding, tablet access, and weekly library access by mobile book cart.

The contrast between Tier II and Tier III is the central structural fact of Georgia's system. The protections that exist were won in litigation over the SMU; the program with no duration cap, no single-cell guarantee and a five-hour weekly out-of-cell floor is the larger one.

Gumm v. Ford: What the Court Found, and What It Did Not Reach

The litigation that produced those protections began in February 2015, when Timothy Gumm — then five years into isolation after a failed escape attempt, serving a life sentence for rape — filed a handwritten pro se complaint in the Middle District of Georgia. Sarah Geraghty of the Southern Center for Human Rights was appointed counsel in October 2016; Kilpatrick Townsend & Stockton joined as co-class counsel. The case is Gumm v. Ford, later captioned Gumm v. Jacobs and Daughtry v. Emmons, No. 5:15-cv-00041-MTT-CHW, before Chief Judge Marc T. Treadwell.

The record the court adopted came largely from Dr. Craig Haney's October 2017 inspection of the Special Management Unit. Haney described a cellblock "full of inmates with serious mental illness," a man "locked for months inside a pitch-black cell," and another man, "naked and psychotic, whose cell was covered in blood." He called the SMU "one of the harshest and most draconian" facilities he had seen "in decades of conducting evaluations," and its residents "among the most psychologically traumatized persons [he] ha[d] ever assessed in this context." He warned that some of the harm "may be irreversible and even fatal." The court adopted findings that residents were confined in cells "smaller than the average parking space," with as little as five hours per week out of cell, no outside light, no congregate programming, and frequently locked food-port flaps. Of roughly 180 people in the unit, 70 were designated mentally ill.

The numbers on duration were equally specific. As of July 2017, 182 people were held in the SMU; 78 percent had been there more than two years, 44 percent more than four years, and 26 percent more than five. Gumm himself was held continuously for seven and a half years despite fourteen separate transfer recommendations over four years. Robert Watkins was held eight to ten years; Johnny Mack Brown nine; Daniel Barfield eight at the time of SCHR's 2017 letter. The Southern Center for Human Rights wrote to GDC Commissioner Greg Dozier on July 31, 2017, quoting Haney's characterization of Georgia's solitary system as "one of the most draconian in the nation."

The settlement, approved May 7, 2019, imposed a general 24-month limit on SMU confinement, mandatory 60-day and 90-day reviews with out-of-cell mental health evaluations by a licensed professional, a prohibition on housing Mental Health Level III or above prisoners in Tier III, minimum out-of-cell time and programming, documentation requirements, and $425,000 in attorney's fees. The court's preliminary approval order found the relief "necessary to prevent violations of the inmates' constitutional rights," "narrowly tailored," and "the least intrusive means."

The scope limit matters as much as the terms. Every operative provision is written in SMU and Tier III vocabulary. The class arose from a single facility. GDC codified the settlement into SOP 209.09 alone. SOP 209.08, governing Tier II, has none of it and has not been revised since April 2016 — nearly three years before the settlement. The protections won in Gumm attached to the unit that was litigated, and to no other part of the Tier system.

Contempt: "No Desire or Intention to Comply"

Compliance did not follow. On April 19, 2024, Judge Treadwell issued a 100-page contempt order finding that GDC officials at the Special Management Unit were "obstinately disobedient, making no real effort to affect change," and that the defendants "were running a four-corner offense and had no desire or intention to comply with the Court's injunction; they would stall until the injunction expired." The order found that compliance documents were "not only insufficient but also unreliable," and documented that officials falsified therapy records — including recording that a man attended treatment sessions after he was already dead. In an August order, Treadwell wrote that "The Court has long passed the point where it can assume that even sworn statements from the defendants are truthful."

The sanctions were concrete: an independent monitor at GDC's expense, daily fines of $2,500 — $75,000 every thirty days for six months — additional attorney's fees, and an extension of the settlement past its initial three-year term. The Associated Press reported the finding. The Justice Department's October 2024 findings report independently noted that GDC had been held in contempt in April 2024 and that monetary sanctions were imposed until compliance.

Six prisoners testified in the contempt proceedings about being denied showers, out-of-cell time, programming, cell cleanout, and access to kiosks and book carts. One described a cell where the toilet was broken and filled with feces and urine from prior occupants; he was forced to urinate in a cup and pour it in the sink, or defecate on toilet paper and dispose of it on his food tray. He had no mattress, no clothing, and was held in freezing temperatures. The GDC attorney did not refute the testimony. The order also documented that GDC placed people in "strip cells" on arrival at the SMU, taking their clothing and leaving them naked or near-naked for hours or days. The plaintiffs documented violations stemming from chronic understaffing, including lack of officer rounds and the recent suicide of a class member. A later fee order states that the court found the defendants had purged their contempt, crediting "the guiding hand of court-appointed monitor Gary Mohr (and hefty daily contempt fines)" with fixing or beginning to fix most problems in the SMU.

The Justice Department's Findings and the Prong That Has Not Been Released

The Justice Department opened its statewide civil investigation of Georgia's prisons in September 2021 under the Civil Rights of Institutionalized Persons Act, 42 U.S.C. §§ 1997 et seq., expanding an inquiry that began in 2016. In April 2024, DOJ expanded the investigation again to include restrictive housing, disciplinary practices, and special education services. The 93-page findings report issued October 1, 2024 addressed only the violence and sexual-abuse prongs.

What it found on segregation was nonetheless direct: "GDC fails to control violence even in its segregated housing units and exposes incarcerated persons to an unreasonable risk of harm due to its inappropriate use of segregated housing." The report warned that conditions "pose a substantial risk of serious harm" and that the state had been "deliberately indifferent" to those risks. It found constitutional violations related to conditions of confinement, including lack of out-of-cell time for people in restrictive housing. It found that queer and transgender prisoners reported being placed in solitary confinement after reporting sexual assault or other violence, or because they were experiencing mental health crises. It recounted that a transgender woman placed in isolation after filing PREA complaints was denied a request to be moved and died by suicide in the isolation unit the very next day.

Assistant Attorney General Kristen Clarke said the report "lays bare the horrific and inhumane conditions that people are confined to inside Georgia's state prison system." GDC disputed the findings the following day, criticizing DOJ for issuing a "Notice Letter" rather than working cooperatively and asserting that "DOJ's track record in prison oversight is poor — often entangling systems in years of expensive and unproductive court monitoring." The agency said separately that it "recognizes the national reforms surrounding restrictive housing, and we are actively engaged in moving forward with internal procedures that address these trends."

The restrictive-housing findings themselves have not been released. As of GPS's most recent review, that prong of the investigation remains unissued — and it is the most likely external source of a per-facility Tier population count, because a CRIPA findings report on restrictive housing would ordinarily contain population, duration and per-facility figures obtained by compulsory process. Its absence is a live, dated gap, not evidence either way.

What Georgia Publishes, and What It Does Not

GPS's systematic review of the GDC-published report corpus found no published count of the Tier or restrictive-housing population. The review covered 110 non-monthly GDC reports and all 34 distribution sections of the monthly Inmate Statistical Profile series. The near-misses fail specifically: the "supervision" field reports only Close, Medium and Minimum; "institution_type" reports only facility classes; "disciplinary" reports counts of disciplinary reports, not sanctions or placements; "mental_health" reports care levels. The Daily Population by Facility series consists of eleven facility-type rows and three columns, with no housing type, no security level and no individual facility.

There is a structural reason the Tier population cannot be recovered from what GDC publishes. SOP 209.09 classifies every person in Tier III as Close Security; SOP 209.45 and SOP 209.55 do the same for both STEP populations. The monthly profile's only housing-adjacent field is "supervision," which reports Close, Medium and Minimum. The Tier population is contained within GDC's published Close Security count — more than 11,600 people — but cannot be isolated from it, because GDC publishes security classification and not housing status.

The agency has, at times, said more. Its FY2014 Annual Fiscal Report states that GDC "Developed and implemented the Tier Segregation Management System at all Level V facilities. As part of this system, 2,721 beds were converted to Tier beds." That is the only system-wide quantification of the Tier system GDC has ever published, and it is a bed count from 2014, not a population. Its FY2018 report states that "since January 2017, the number of offenders in extended restrictive housing has decreased by 41 percent (671)," achieved partly through the Segregated Transition Education Program, specialized protective custody, specialized mental health treatment units, administrative reviews and policy restructuring. Its FY2019 report gives a different figure for the same reduction — 40 percent, 675 people, since the same January 2017 baseline — and notes that the "TIER 2 SCRIBE module went live March 2019 to effectively track offenders in Extended Restrictive Housing." Its FY2020 report adds that the SCRIBE Warden's Dashboard was given a field for "number of days an offender has spent in restrictive housing Tier II."

Those disclosures have a distinctive shape: a delta with no endpoints. GDC reported how many people left and what percentage that was, and never stated the population before or after. Then the reporting stopped. A search of all eleven GDC Annual Fiscal Reports from FY2014 to FY2025 finds restrictive-housing and Tier narratives in FY2014, FY2017, FY2018, FY2019 and FY2020. From FY2021 onward the subject disappears as a program narrative; the only matches in FY2024 and FY2025 are "Special Management Unit" appearing in a list of facilities receiving the NoteActive electronic logbook. The last year in which GDC published any figure about the size or trend of its restrictive-housing population was FY2019.

One facility-level figure survives in the annual report series. GDC's FY2018 report records that Guard 1 Technology was installed at Valdosta State Prison and "captures automated segregation rounds in all tier units (178 Cells)." A single prison's Tier units comprised 178 cells — approaching the size of the entire Special Management Unit, which GDC's FY2017 report describes as consisting of 192 beds.

The only Tier population GDC does publish is the SMU's, and only as a side effect of the SMU being carried as a standalone facility in the monthly Inmate Statistical Profile. As of September 1, 2026, it held 154 people. Across 75 monthly snapshots from July 2020 to September 2026, the SMU population ranged from 149 to 195 — below its 192-bed capacity throughout the published series. That is one facility at Jackson. It says nothing about Tier I or Tier II, whose units sit inside ordinary prisons and are absorbed invisibly into those prisons' facility totals.

The Bed Count GPS Obtained, and the Facility GDC Left Out

On September 14, 2026, GPS filed open-records request R031710-091426 through GDC's portal, asking which facilities operate Tier II and Tier III programs and how many people were assigned to each. On October 2, GDC's Legal Services division uploaded a one-page table titled "Total_Beds_with_Locations_Tier_II_and_III.pdf" and stated that "No charges have been assessed." It was the first per-facility count of Georgia's Tier II population GPS holds from GDC.

The table reports beds rather than people, carries no as-of date, and does not split the STEP beds between Tier II STEP and Tier III STEP. Read as the nearest available measure, it shows a system substantially larger than the SMU. Hays State Prison holds the largest restrictive-housing program outside the SMU: 151 Tier II beds plus 59 STEP beds, 210 in all, every one occupied. Ware holds 138 Tier II beds, all occupied. Telfair 117, all occupied. Macon 121 beds with 119 occupied, plus 20 STEP beds at 100 percent occupancy. Smith 111 Tier II beds with 89 occupied and 20 STEP beds with 14 occupied. Hancock 83 Tier II beds with 82 occupied. Valdosta 129 Tier II beds with 80 occupied — the most room in the system. McRae Women's Facility, 34 Tier II beds with 28 occupied. The SMU's Tier III unit: 160 beds, 128 occupied.

McRae is the finding inside the finding. GDC's public Tier Segregation System fact sheet names seven Tier facilities, all men's prisons: Hancock, Hays, Macon, Smith, Telfair, Valdosta and Ware. The current version, produced on October 5, 2026 in the same open-records request, still names only those seven. GDC's own bed table, produced three days earlier, lists an eighth Tier II location: McRae Women's Facility. The Tier II program does operate in a women's prison, and GDC's public description of the system leaves it out.

The bed table also confirms what the PREA audits had already shown: double-bunked segregation is not exceptional in Georgia. Hancock's G-1 "Tier 2" unit is described in two audit cycles in identical words as "a specialized management program housing a maximum of 96 inmates, in double occupancy cells located on two tiers." Dooly's J2 is "designated housing of Administrative Segregation, Isolation and Mental Health Observations ... 96 inmates in double bunks in a cell setting." Smith's dorms comprise "48 double bunked cells housing 96 offenders per dorm." At Georgia State Prison, before its 2022 closure, Tier II at G Building was double-bunked while Tier II at K Building was single-cell lockdown — two occupancy modes running inside the same program at the same prison. SOP 209.08 imposes no single-occupancy requirement at all; SOP 209.06 treats single-celling in administrative segregation as an enumerated exception requiring justification.

That matters for a definitional reason. The Mandela Rules define solitary confinement as confinement for 22 hours or more a day without meaningful human contact. GPS can establish the first element for Tier I and Tier II by arithmetic from GDC's own out-of-cell minimums: a five-hour weekly exercise entitlement leaves 163 hours per week in cell, or 23.3 hours per day. GPS cannot establish the second element from the SOPs, and does not assume it. Where a Tier II cell is single-occupancy, both elements may be satisfied. Where it is double-bunked, the person is in cell 23.3 hours a day but is not without human contact in the Rule 44 sense. Whether a person in Georgia's Tier system is held alone is determined by which unit they are assigned to — and GDC's Tier II policy neither specifies nor reports it.

The Segregation Cells the Tier Label Does Not Mark

The Tier designation marks where GDC concentrates long-term administrative segregation programming. It does not mark where segregation cells exist. A review of all 273 GDC-published PREA facility audit reports found named Tier program housing in the audits of exactly nine facilities, all close-security: Georgia State Prison, Smith, Hays, Valdosta, Telfair, Macon, Hancock, Georgia Diagnostic and Classification, and Ware. There is not one exception in 273 documents.

But segregation housing is documented at medium-security prisons throughout the system, sometimes at larger scale. Calhoun State Prison, whose Cycle 2 audit gives its security level as "4/Medium," reports 197 segregation cells — more than close-security Baldwin (167), Ware (125) or Phillips (75). Wilcox, described in its own audits as "housing medium security, general population inmates," reports 168 segregation cells in Cycle 3, up from 24 double-bunked in Cycle 1 and 72 in Cycle 2. Coffee Correctional Facility, a private medium-security prison, reports 170. Rogers ("Minimum/Medium") reports 95. Dooly reports 71 cells in Cycle 3 and describes a 96-person double-bunked administrative segregation unit in Cycle 4. Johnson's J-1 and J-2 are "an administrative segregation unit" holding 72 and 94 offenders. Also documented: Washington 48, Dodge 50, Coastal 74, Rutledge 32, Montgomery 16, Lee 16 cells and 32 beds, Central 24, Whitworth Women's 7, Burruss 10, Long Unit 4. Metro Reentry Facility reported zero.

The corpus itself is internally inconsistent, and GPS records the defects rather than reconciling them. Georgia State Prison's "Cycle 3" audit is the Cycle 2 audit republished — both 181 pages, identical text after whitespace normalization, both dated March 22, 2019. Telfair's Cycle 4 audit contradicts itself internally, stating that "the facility does not have administrative segregation cells" and, in the same report, that "the facility head reported having 360 administrative segregation beds." Georgia State Prison's Cycle 2 questionnaire says 12 segregation cells while its narrative itemizes 106 Tier I cells at E-South, 218 Tier II beds at G Building and 100 Tier II single lockdown cells at K Building. Any published figure drawn from this corpus must name the facility, the cycle and the field.

Mental Health, the Level III Door, and the Clinical Route Out

GDC's own mental health policy places Tier II among the most psychiatrically demanding settings it operates. SOP 508.01, effective July 27, 2023, lists "the highest needs" as Specialized Behavioral Therapeutic Units, Under Death Sentence, the Special Management Unit, Crisis Stabilization and Acute Care Units, "Specialized Mental Health Treatment Units (SMHTUs/TIER II, etc.)" and Level IV Supportive Living Units. It sets a ratio of "at least 1.0 FTE Mental Health Counselor per 20 offenders at designated SMHTU/TIER II facilities" — the second-richest counselor ratio in the system, behind only crisis stabilization at 1:15.

The placement rules then route people with the highest classifications away from the reformed unit. SOP 209.09 excludes both Mental Health Level III and Level IV from Tier III. SOP 209.08 excludes only Level IV and admits Level III on mental health recommendation: "MH Level III must be screened and recommended by MH staff before consideration for placement in the Tier II Program." The effect written into policy is that prisoners with the higher mental health classifications remain eligible for the program whose floor is five hours a week out of cell, and are routed away from the one that guarantees four hours a day.

The historical record is worse than the policy. Thirty-nine percent of prisoners in Georgia's Special Management Unit had a diagnosed mental illness by GDC's own classification. The Southern Center for Human Rights alleged in its September 2021 class action that over 70 percent of the approximately 300 people in Georgia State Prison's Tier II program experienced serious mental illness. Georgia State Prison's Cycle 2 audit records "D East houses 49 Tier II Mental Health Offenders." SOP 209.08's own admission criteria list a "Tier II-Mental Health program" as a distinct placement and provide that people transfer from the SMU into "Tier II-Phase 3 or Tier II-Mental Health."

There is a clinical route out of restrictive housing on health grounds, and it generates records. SOP 507.04.33, effective January 31, 2022, provides that "If following the review of the health record, the nurse believes that the offender's health will be adversely affected by continued placement in restrictive housing, the nurse will notify the physician. If the physician concurs, security personnel will be notified and alternate arrangements made." The same SOP requires daily medical rounds in restrictive housing — a standard that conflicts with the older Tier policies, which require health care rounds only three times per week "excluding weekends and holidays." Three visits a week is fewer than half of seven. GDC has not harmonized the two, and the older, less protective standard is the one written into the policy that governs the Tier unit itself.

The research base on what this does to people is not seriously contested. A June 2025 meta-analysis in PLOS One synthesizing data from 171,300 inmates found significantly greater psychological distress, more psychiatric symptoms including self-harm, and greater need for mental health services among prisoners in disciplinary confinement. A 2019 systematic review of 13 studies covering 382,440 inmates found higher-quality evidence associating solitary confinement with increased adverse psychological effects, self-harm and mortality. Kaba et al., analyzing 244,699 incarcerations in the New York City jail system, found that although only 7.3 percent of admissions included any solitary confinement, 53.3 percent of self-harm acts and 45 percent of potentially fatal self-harm acts occurred within that group. Nationally, roughly half of prison suicides occur among people in solitary confinement, who comprise an estimated 6 to 8 percent of the prison population.

The People Who Died

The registry entries below are the deaths in GPS's records that the evidence on this page discusses. They are not a complete accounting of deaths in Georgia's segregation units; they are the ones GPS can name.

James Merritt Byrd died January 23, 2022, at Effingham County Prison; GPS's registry lists the cause as Medical Neglect. He was 30.

Reginald Jerome Jacobs Jr. died February 6, 2023, at Calhoun State Prison; GPS's registry lists the cause as Deliberate Indifference. He was 24. He was found dead on the floor of an isolation cell; the coroner determined he died of dehydration with renal failure, and the body was rigid, wrapped in mattress padding, leaning against the door. No one had entered his cell for two days. The flap in the door had been locked shut; staff had shut off the water supply to his room, closed the flap, and did not deliver meals. The Justice Department documented the case in its 2024 findings.

Arthur Wimbush died April 2, 2023, at Phillips State Prison; GPS's registry lists the cause as Homicide. He was 46 and weeks from release. GPS's reporting describes accounts that the facility's Chief of Security attempted to move him to safety two days earlier and found every segregation bed full.

Sherry Elaine Joyce died April 27, 2024, at Arrendale State Prison; GPS's registry lists the cause as Homicide. She was 61. Hallie Marie Reed died May 5, 2024, at Arrendale State Prison; GPS's registry lists the cause as Homicide. She was 23, and had asked in writing for protective custody eight days after Joyce's death, citing fear; the request was denied.

Stephen Prochaska died January 21, 2025, at Augusta State Medical Prison; GPS's registry lists the cause as Suicide. He was 70. Justin Waymon Hollingsworth died June 26, 2025, at Rogers State Prison; GPS's registry lists the cause as Suicide. He was 43. Sheqweetta Vaughan died July 9, 2025, at Arrendale State Prison; GPS's registry lists the cause as Unknown/Pending. She was 32. GPS's reporting describes accounts that she was found in a segregation cell in triple-digit heat, and that a neighboring prisoner reported hearing her call for medical help more than 28 hours before she was discovered.

Calvin Earl Noble died August 26, 2025, at Macon State Prison; GPS's registry lists the cause as Suicide. He was 25. Marcus Walker died January 10, 2026, at Wilcox State Prison; GPS's registry lists the cause as Homicide. He was 20. Christopher Lee died January 31, 2026, at Georgia Diagnostic and Classification State Prison; GPS's registry lists the cause as Natural/Medical. He was 19. Miguel Angel Duran died March 1, 2026, at Central State Prison; GPS's registry lists the cause as Suicide. He was 44. Denecia Nichelle Randall died March 30, 2026, at Pulaski State Prison; GPS's registry lists the cause as Suicide. She was 28.

Two further deaths in the record are not in the registry lines above but belong to this account. In February 2023, an incarcerated person was found dead in his restrictive-housing cell at Calhoun State Prison — the Jacobs case as the Justice Department described it. In April 2023, a man at Smith State Prison was discovered dead in a segregated housing unit, possibly strangled by his roommate; the local coroner noted the body was badly decomposed and the man likely had been dead for over two days. In 2017, two men died by suicide in the Special Management Unit at Georgia Diagnostic and Classification Prison.

Beyond the named dead, GPS has independently tracked 1,927 deaths in GDC custody since 2020.

The Records That Exist and the Records That Do Not

The mechanism of the Tier system's invisibility is documented in GDC's own rules. A review of the 209 series found no SOP naming any Tier roster, periodic report, or Central Office aggregate submission. The term "Central Office" does not appear in SOP 209.08 at all. Instead, Tier placements flow upward as individual approvals: for Tier II, Classification Committee to Warden, with notice to the Regional Director, and release requiring the Director of Field Operations; for Tier III, facility Warden to Tier III Program Classification Committee to SMU Warden to North Regional Director, with an appeal to the Deputy Director, Field Operations and a final decision by the Director, Field Operations. Every approval travels as a single case. The reason no Tier population figure is published is not that the data does not exist — GDC's own annual reports say a SCRIBE module tracks it — but that no policy instrument requires anyone to compile it into a report.

The records that would answer the question are named, dated and requestable. The Accountability Log required by SOP 209.06 and SOP 209.09 records "cell number, bed number, offender name, offender number, race, in or out time, assignment of offender, and destination of offender," and is "printed daily after 1800 hours." SOP 209.07 and SOP 209.08 impose the same requirement for Tier I and Tier II. The segregation admissions and releases register records name, housing location, date and time of action, reason for admission, tentative release date, and authorizing official. The Restrictive Housing Rounds Log, form M40-01-01, is required weekly at every facility by SOP 508.20. The Sick Call Encounter Log for Restrictive Housing, form P-03-0005-11, is one of seventeen mandatory tracking logs required by SOP 507.04.04 at all facilities "including private and county prisons." Board of Corrections Rule 125-3-2-.10(f) requires that "No confinement to an isolation cell of greater than thirty (30) days shall be imposed without the review and approval of a Division Director" — counting those approvals for a year would give a floor on long-stay isolation independent of any Tier population figure. SOP 209.09 Attachment 12 is the quarterly over-24-month panel review form. And SOP 508.01 names a SCRIBE report called "Daily lock down" as a routine mental-health oversight instrument — a named, existing, daily electronic report on lockdown status.

GDC has demonstrated it will run and produce a per-facility housing count query. GPS open-records request R024615-101925, filed in October 2025 and completed, requested a "Current count of inmates by security designation (Close, Medium, Minimum) housed at each Georgia Department of Corrections state prison," and GDC produced per-facility count tables. That precedent materially undercuts a burden or new-record objection to a Tier placement count. As of GPS's review, none of the 43 GDC requests in the registry was segregation-related before R031710-091426.

There is one obstacle worth naming. SOP 219.01 asserts that "Offender records are confidential state secrets under O.C.G.A. § 42-5-36(c)" and directs that all requests for offender-related records be forwarded to the Office of Legal Services. A request framed around individual records will meet that exemption; a request framed around aggregate counts, unit rosters stripped of names, and designation and capacity documents will not.

The Grievance That Cannot Be Grieved

SOP 227.02 prohibits retaliation for filing a grievance in the strongest terms: "Retaliation against an Offender for filing a grievance is strictly prohibited. The prohibited retaliation includes, but is not limited to, disciplinary action against the Offender for filing a grievance." A person may file a grievance alleging retaliation "regardless of the form of the alleged retaliation or harassment."

The same SOP then removes the alleged act from the grievance system. Involuntary assignments to administrative segregation are non-grievable, routed instead to the SOP 209.06 classification appeal. Housing assignments, program assignments and security classifications are non-grievable "unless there is an alleged threat to the Offender's health or safety." The classification appeal tests whether the placement criteria are met; it does not test motive; and SOP 209.06 provides that "The Warden's/Superintendent's review and decision on the appeal is final." The first 96 hours of segregation rest on a 24-hour decision that the policy makes unappealable. There is no neutral or external decision-maker anywhere in the process, and no right to counsel or to a non-staff advocate.

The record-keeping rules compound it. SOP 227.02 provides that "A Grievance must not be placed in the Offender's file or referred to in SCRIBE case notes." SOP 209.08 conditions release from Tier II on the Unit Manager's review in SCRIBE of the offender's disciplinary reports, profiles, case notes, Security Threat Group history, incident reports, progress notes, movement history, mental health status, personal history sheet, medical profile, court production orders, and county of conviction and residence. A keyword search of SOP 209.08 for "grievance" returns nothing: the Tier II policy does not mention grievances at all and contains no anti-retaliation provision. The one record that would show a grievance preceded a Tier placement is, by GDC policy, absent from both the institutional file and the SCRIBE case notes — and SCRIBE case notes are among the records that determine whether a person leaves Tier II.

GPS has no document establishing that Georgia's Tier programs are used to retaliate against people who file grievances, contact lawyers or the press, report staff misconduct, or otherwise speak out. No court has made such a finding. GPS's case-law corpus returns nothing for retaliatory segregation in Georgia, but the corpus is post-conviction and habeas weighted and does not hold section 1983 conditions litigation, so its silence is not itself evidence of absence. GPS's internal case file records alleged courses of retaliation in which segregation was the method, but the coding schema has no Tier value — Tier I, Tier II, Tier III, disciplinary isolation and administrative segregation all collapse into a single code. The data model cannot answer the question as asked.

What GPS records at scale is different and broader. Across the twelve months from December 2025 to October 2026, GPS records 78 due-process-violation signals from 14 facilities, drawn from 33 distinct cases, with the heaviest concentrations at Georgia Diagnostic and Classification, Coastal, Washington, Calhoun and Wilcox. The external bodies named across those signals include the Eleventh Circuit Court of Appeals, the DOJ Civil Rights Division, the State Board of Pardons and Paroles, the U.S. Department of Justice, and federal and appellate courts. In the same period, GPS records 24 mental-health-crisis-unattended signals from six facilities, concentrated at Johnson, Augusta State Medical Prison, Georgia Diagnostic and Classification, Hancock and Wilcox.

The Juvenile Program and the Step-Down

Children in GDC custody are held under a program modeled on Tier II. SOP 209.11, "Restrictive Housing Assignment - Juvenile Offender Administrative Segregation," effective April 11, 2016, governs juvenile offenders under 18 and uses the same multi-phase structure, the same three-phase visitation ladder, the same moderated-cell cross-reference to SOP 209.05, and the same Warden modification clause as Tier II. Its out-of-cell floor is identical: five hours per week of exercise. On release review, a juvenile may be "Reassigned to the same phase of Tier II as the current Phase of RHA-JOAS, if of age." Children carry their phase into adult Tier II when they turn 18.

The juvenile policy is, in two respects, more protective than the adult one. It provides contact visits at every phase where the adult Tier II policy provides non-contact visits only. And it forbids open-dormitory housing: "The housing unit shall not be open dormitory." No parallel prohibition appears in SOP 209.08.

The step-down programs are the system's stated answer, and GDC classifies them as something other than restrictive housing. SOP 209.45, effective April 25, 2019, governs Tier II STEP and provides that "Cells may be single or double occupancy," with four hours out of cell per day seven days a week, one hour of exercise, fifteen-minute calls, two-hour non-contact visits, general-population tablet access, and thirty-minute checks. SOP 209.55, issued the same day, governs Tier III STEP and is textually near-identical except for one substantive difference: "All cells are single occupancy." GDC wrote two step-down policies on the same day with the same conditions and made cell occupancy the one point of divergence. Both SOPs state affirmatively that the STEP program "is not a restrictive housing unit" — a definitional carve-out that will exclude the step-down population from any GDC production framed on the term "restrictive housing."

The October 2026 bed table shows 119 STEP beds across four locations: Hays 59, all occupied; Macon 20, all occupied; Smith 20, 14 occupied; Georgia Diagnostic and Classification 20, 3 occupied. It contains no count of STEP participants, completions, failures, or returns to Tier.

The Comparative Record and the Eighth Amendment Gap

The Eleventh Circuit, which governs Georgia, has not issued a published opinion squarely holding prolonged solitary confinement unconstitutional under the Eighth Amendment. In Marbury v. Warden, 936 F.3d 1227 (11th Cir. 2019), the court found deliberate indifference shown by "pervasive staffing and logistical issues rendering prison officials unable to address near-constant violence." In Hope v. Harris (2021), the Fifth Circuit held that solitary confinement does not violate the Eighth Amendment "no matter how long it is imposed for." The Supreme Court denied certiorari in 2023 in a case involving Dennis Wayne Hope, held in continuous solitary for 27 years in a 54-square-foot cell.

Other circuits have moved. In Porter v. Clarke, 923 F.3d 348 (4th Cir. 2019), the Fourth Circuit became the first federal court of appeals to hold that long-term solitary on Virginia's death row violated the Eighth Amendment, finding the conditions deprived inmates of "the basic human need for meaningful social interaction and positive environmental stimulation." In Williams v. Secretary Pennsylvania DOC, 117 F.4th 503 (3d Cir. 2024), the Third Circuit held that "It was clearly established that someone with a known preexisting serious mental illness has a constitutional right not to be held in prolonged solitary confinement without penological justification," and rejected qualified immunity; the Supreme Court denied certiorari in 2025. Madrid v. Gomez, 889 F. Supp. 1146 (N.D. Cal. 1995), likened prolonged solitary for the mentally ill to "the mental equivalent of putting an asthmatic in a place with little air to breathe."

Wilkinson v. Austin, 545 U.S. 209 (2005), supplies a due-process argument that has never been applied to Georgia's Tier II. The Court held that Ohio's supermax conditions — near-total solitary, indefinite duration, and parole disqualification — imposed an "atypical and significant hardship" requiring informal procedural protections. The factor the Court identified as central, indefinite duration, is present in Tier II, where SOP 209.08 sets no outer limit, and absent from Tier III, where SOP 209.09 sets a 24-month ceiling with enumerated exceptions and a quarterly panel. On Wilkinson's own reasoning the due-process argument is stronger for Tier II than for the program the litigation actually reformed. That is an argument from a holding, not a holding.

The comparative record is not ambiguous about what is possible. California's Ashker settlement ended indeterminate solitary, ended gang affiliation alone as a basis for SHU placement, capped continuous Pelican Bay SHU stays at five years, and created a four-step, two-year step-down program; the Pelican Bay long-term isolation population fell from 513 to 2. Colorado reduced its administrative segregation population from roughly 1,500 in 2011 to under 200 by 2017, and after banning solitary in two mental health prisons, assaults, forced cell entries and the use of heavy restraints declined by 40 percent. North Dakota achieved a 74.28 percent reduction in solitary use between 2016 and 2020. Pennsylvania's Disability Rights Network v. Wetzel settlement removed approximately 800 seriously mentally ill people from restrictive housing statewide and capped any such placement at 30 days. New York's HALT Act caps solitary at 15 consecutive days. Connecticut and Nevada enacted 15-day maximums in 2023. Georgia, by contrast, considered and did not pass House Bill 714 in 2019, which would have prohibited consecutive periods in restrictive housing and capped total restrictive housing at no more than 90 days per year.

What the Record Supports

Three findings rest entirely on GDC's own documents and are not seriously contestable. First, Tier II has no outer duration limit in written policy, and the policy that says it is not punishment is the designated route around the thirty-day limit on punishment. Second, the protections won in Gumm v. Ford — the 24-month ceiling, the mental health exclusion, the quarterly panel, the four hours out of cell — were codified into the Tier III SOP alone and appear nowhere in the Tier II SOP, which has not been revised since 2016. Third, GDC has published no Tier population count since FY2019, and the structure of its published data makes the Tier population arithmetically unrecoverable rather than merely unpublished.

A fourth finding is newer and rests on the October 2026 production: the Tier II system is substantially larger than the SMU, with 210 occupied restrictive-housing beds at Hays alone against an SMU that has held 149 to 195 people across six years of published snapshots. And a fifth: GDC's public description of the Tier system omits the women's prison that runs a 34-bed Tier II unit.

What the record does not support is any claim that Georgia's Tier placements are used to retaliate for protected activity. GPS has no document establishing it and no court has found it. What GPS has is a grievance system that prohibits retaliation in the abstract while making the segregation assignment itself non-grievable, a Tier II policy that does not mention grievances at all, and a records rule that keeps grievance filings out of the very SCRIBE case notes that determine release. The sequencing question — was the Tier assignment before or after the grievance — is answerable from two documents that exist: the SCRIBE Tier assignment record and the Classification Committee 90-day review forms, set against that person's grievance filing dates. Nobody has asked for them yet.

The system that produced these rules has also produced the named dead on this page, and a state that stopped publishing the manner of death in February 2024. Both facts belong in the same sentence, because they are the same fact: a system that does not count what it does cannot be shown what it has done.

Sources

This analysis draws principally on the Georgia Department of Corrections' own published record: the segregation and restrictive-housing standard operating procedures in the 209 series (SOP 209.01, 209.03, 209.05, 209.06, 209.07, 209.08, 209.09, 209.11, 209.45, 209.55), the health services and mental health policies governing restrictive housing (SOP 507.04.33, 508.01, 508.20, 219.01, 227.02), the Board of Corrections rules on isolation and medical checks, all 273 GDC-published PREA facility audit reports across Cycles 1 through 5, eleven Annual Fiscal Reports from FY2014 to FY2025, the monthly Inmate Statistical Profile series, and the Tier II and Tier III bed table produced in October 2026 under GPS open-records request R031710-091426. Court records include the April 19, 2024 contempt order and May 8, 2019 decision in Gumm v. Ford, No. 5:15-cv-00041-MTT-CHW (M.D. Ga.), and the published opinions cited above. The Justice Department's October 1, 2024 findings report on Georgia's prison system supplies the federal findings on segregated housing, staffing and violence. GPS's own investigative reporting and its mortality registry supply the named deaths and the accounts of individual cases described here; where those accounts rest on family or witness allegation rather than court or agency documentation, they are framed as accounts. GPS's intelligence records supply the aggregate signal counts for the December 2025 to October 2026 period.