Disability: is the fraud number even fraud?

Two programs, not one roll — and do not sum them casually

SSDI (Title II) is social insurance for insured workers and eligible dependents. SSI (Title XVI) is means-tested cash for aged, blind, or disabled people with limited income and resources. December 2023: 7,365,987 disabled workers; 8,709,006 DI disabled beneficiaries including widow(er)s and adult children; SSI total 7,425,331, of which blind and disabled 6,264,542. Concurrent dual entitlement ages 18–64: 949,971. Adding SSDI workers to SSI blind/disabled double-counts that concurrent pool. Age-sex-adjusted DI incidence peaked at 6.4 per thousand exposed in 2010, fell to 2.9 in 2022–23, and stood at 3.3 in 2024 — a sustained post-peak decline, not a mystery surge in raw rolls.

The fraud number is not a fraud number

SSA’s Agency Financial Report is explicit: improper-payment findings come from stewardship reviews of the nonmedical aspects of OASDI and SSI. Terminating benefits after a medical Continuing Disability Review does not mean the original award was wrong — medical improvement means prior benefits were proper. FY2023 SSI improper payments: $6.5B / 10.62% (overpayment 9.18% / underpayment 1.44%). OASDI combined: ~$4.1B / ~0.30%. Leading SSI causes: financial accounts 28%, wages 24%, in-kind support and maintenance 9%. No AFR table equates these rates to intentional fraud.

The high-circulation claim that Social Security is losing on the order of $70–72 billion to fraud root-traces to one administrative source: an SSA OIG release summarizing nearly $72 billion in improper payments across FY2015–FY2022 — less than 1% of benefits — later restated in political and media framing as fraud. That is citogenesis: one root number, a label swap, peer repetition. Separately, CDR cessations (~39k disabled-worker initial cessations FY2019; DDS cessations can exceed 100k in a pre-COVID comparison year) dwarf CDI judicial actions (~74–77/year). Cessation is not a fraud finding; prosecution counts are not an IP-dollar series. A prosecution-centric “fraud theater” program is not among the ten scored rows — the protocol replaced it with integrity focused on CDRs and wage reporting.

The backlog is stage-resolvable — and still above the agency’s own goal

FY2024: initial allowance 38% (deny 62%) on ~2.09M decisions; reconsideration 16%; ALJ hearing allowance 51% (dismiss 33%, deny 16%) on ~289k hearings. Hearing average processing time 342 days (FY2023 450) against SSA’s 270-day goal; pending hearings ~262k (−19% YoY). Initial APT 231 days, up from 218; pending initials ~1.18M. Improvement at hearings is real; clearance of the goal is not. Initial and hearing pools are not independent draws — higher hearing allowances reflect selection through denial and appeal, not proof that ALJs are “soft” without case-mix controls.

GAO-18-37 found residual ALJ allowance dispersion after case-mix controls, with roughly 5 percentage points of narrowing under quality assurance and training — a separable administrative series, not a raw-allowance mislabel. Capacity-blind “get tougher” reforms under a million-plus initial pending queue inherit that caution.

The work cliff: BOND null, Ticket ≪2pp, 1619(b) at 2.6%

Benefit Offset National Demonstration (BOND): Stage 1 and Stage 2 returned null mean earnings effects; Stage 1 average benefits due rose about $143/year, Stage 2 about $450–500/year — much of it a windfall to people already at substantial gainful activity, not a poverty-reduction proof. Ticket to Work: intent-to-treat effects on employment or benefit exit for work are null or undetectable at population scale (≪2 percentage points); service enrollment rises a fraction of a point; assignment stays roughly 1–5% of eligibles. Section 1619(b) Medicaid continuation: 108,825 participants ages 18–64 in December 2023 — 2.6% of blind and disabled SSI recipients in that age band.

Cliff redesign, Ticket redesign, and Medicaid buy-in are three different levers. Buy-in attacks the coverage cliff — Washington’s matched MBI evaluation and the Mathematica MBI series support earnings/work effects the cash-offset arm did not clear. Forty-seven states already offer a buy-in (KFF). It is not “the cliff redesign that cleared the BOND bar.”

Awards ≠ stock — and childhood SSI is not adult DI in miniature

The circulating prior that mental disorders are roughly a third of working-age SSDI awards is false on awards. 2023 disabled-worker awards: mental disorders 12.7%; musculoskeletal 34.0%; neoplasms 13.6%. December 2023 disabled-worker stock: depressive/bipolar 12.4% + intellectual 3.8% + all other mental 12.4% ≈ 28.6%. SSI under 65: agency profile says about six of ten have a mental-disorder diagnosis — a different program and denominator. Gluing those into “one-third of awards” is the seam this filing shares with GBMT-11 (mental health). Childhood SSI is a separate track: adult IP-as-fraud and Ticket/SGA frames mis-travel; PRWORA already executed a child-specific standard at national scale.

The scorecard: ALJ QC leads two readers; Medicaid buy-in leads the third

Objectives: O1 measurement integrity · O2 adjudication timeliness/accuracy · O3 poverty and insurance protection · O4 work without a cliff · O5 administrability. Pass 1 authored the scales; red-team moved three cells; a structurally blinded re-score matched 42 of 50 cells and moved six more. Pass-2 tops:

Rank-stable top four under every weighting: #1 capacity surge, #2 ALJ QC, #4 Medicaid buy-in, #8 childhood separate track. #9 definition tightening is last under all three on this board — because #10’s unevidenced below-neutral cells returned to 3 under the evidence floor, not because #9 was cut further after red-team. That is a pass-2 arithmetic consequence; it is not a quiet reassertion of a withdrawn overclaim. #8’s integrity-hawk second place remains a construct artifact: refusing adult-fraud imports is an O1/O5 win, not the adult DI Trust Fund fix. Adult integrity readers should read #2 and #6 as the adult-facing instruments.

Honesty box → (what we are not claiming)