Social Security Taxation Reform
Draft manuscript for long-run policy analysis and impact assessment
This draft manuscript is the narrative companion to the interactive Social Security taxation-of-benefits dashboard. It documents the fourteen-reform menu, the long-run microsimulation and calibration strategy, the validation and benchmark framework, and the publication architecture for the current CRFB analysis package. The current build runs on PolicyEngine’s primary-source populace microdata and calibrates each 2026–2100 year to the 2026 Social Security and Medicare Trustees current-law projections, which already incorporate the 2025 reconciliation act. The dashboard remains the live current-results explorer; this paper is the stable long-form surface intended for citation, SSRN-style circulation, and formal external reference once shared exhibits are frozen.
Social Security, Taxation of benefits, Trust funds, Microsimulation, PolicyEngine, CRFB
1 Scope and framing
This manuscript is the long-form publication layer for the current Social Security taxation-of-benefits project. It sits alongside, but is intentionally separate from, two other project surfaces:
- the dashboard, which is the live current-results explorer
- the operational docs, which track rerun status, reproducibility, and audit decisions
The current analysis extends the original eight-option report to a broader fourteen-reform package over a 2026-2100 modeling window. The live release surface is the standard reform set, option1 through option12, plus two additional reforms — a reverse-Roth proposal and a 93% benefit-taxation benchmark — all generated from the current full-H5 contract on the populace microdata base.
1.1 Current-law baseline
Under current law, Social Security benefits become taxable once combined income exceeds statutory thresholds, with up to 50 percent of benefits taxable at the lower tier and up to 85 percent taxable at the higher tier (Internal Revenue Service 2024). Revenue from this taxation is split between the Social Security OASDI trust funds and the Medicare Hospital Insurance trust fund (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026; Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds 2026).
The baseline is the 2026 Trustees Report intermediate-assumptions current law, which already incorporates the 2025 reconciliation act (the One Big Beautiful Bill Act), including the temporary bonus senior deduction and its scheduled expiration after 2028 (Joint Committee on Taxation 2025). Because the Trustees baseline is OBBBA-native, this analysis requires no post-legislation bridge: the modeled current law and the calibration target describe the same world. The senior-deduction expiration still matters for interpretation because several reform options extend, repeal, or replace that temporary senior-specific tax relief.
1.2 Research questions
The paper is organized around three questions:
- How do alternative Social Security benefit-taxation reforms alter trust-fund revenue, household tax burdens, and long-run fiscal outcomes?
- What data and validation framework are required to make
2026-2100microsimulation estimates interpretable? - How should the interactive dashboard, the citable paper, and the operational audit record relate to one another without drifting?
1.3 Why a separate manuscript
The dashboard is optimized for exploration rather than formal citation. A separate manuscript is useful because it can:
- preserve the source and benchmark trail from the first report
- carry fuller methodological exposition than the live app should contain
- support SSRN-style circulation and external citation
- freeze narrative interpretation even when the dashboard continues to evolve
This draft therefore focuses on preserving substantive coverage and source structure ahead of final results, while reserving numerical exhibits for the shared artifact pipeline that will populate the release candidate.
2 Policy design
The current reform menu spans four broad families: direct benefit-taxation changes, senior-relief redesigns, employer-payroll-tax swaps, and a reverse-Roth restructuring. The release series shares one common current-law baseline.
2.1 Standard options 1-12
2.1.1 Direct taxability changes
option1: full repeal of Social Security benefit taxation beginning in2026.option2: tax 85 percent of all Social Security benefits by eliminating the current-law thresholds and applying the top statutory inclusion rate to all recipients.option8: tax 100 percent of benefits.option9: tax 90 percent of benefits.option10: tax 95 percent of benefits.tax93: tax 93 percent of benefits. This benchmark applies a uniform 93 percent inclusion rate across the taxability tiers; it sits between the 90 and 95 percent options and corresponds to the share the Social Security actuaries have used in illustrative full-taxation analyses.
These options primarily change the size of the taxable-benefit base while preserving the current-law trust-fund accounting framework.
2.1.2 Senior-deduction and credit redesigns
option3:option2plus permanent extension of the temporary bonus senior deduction.option4:option2, repeal of the bonus senior deduction, and a$500nonrefundable credit tied to Social Security tax liability.option7: eliminate the bonus senior deduction without otherwise changing the Social Security taxability structure.option11:option2, repeal of the bonus senior deduction, and a$700nonrefundable Social Security credit with a6%phase-out above$150,000for joint filers and$75,000for other filers.
This family matters because the baseline itself changes sharply once the bonus senior deduction expires after 2028, and because a flat credit is more progressive than a deduction whose value rises with marginal tax rate. In the long-run estimates, the option11 credit phase-out thresholds are treated as fixed nominal parameters of the modeled policy. Combined with the Trustees wage-indexing treatment for federal deduction thresholds after 2034, this means the remaining tax units with positive gross option11 credit in late years can have no pre-credit income-tax liability; the nonrefundable cap then makes the gross credit unavailable. This is why the far-horizon option11 revenue effect can converge to the option2 effect even though the gross credit variable is positive.
2.1.3 Employer-payroll-tax swaps
option5: eliminate Social Security benefit taxation and instead treat employer payroll-tax contributions as taxable compensation.option12: a phased Roth-style swap that taxes employer payroll contributions immediately but phases out the OASDI share of benefit taxation first and the HI share later.
These options are structurally different from pure benefit-taxation reforms because they change the tax base itself, shifting burden from retirees toward workers while preserving trust-fund accounting through explicit OASDI/HI allocation rules.
The raw artifact still retains option6, a shorter phase-in variant, for lineage and sensitivity checks. It is not shown as a separate CRFB-facing Roth option because the publication menu uses the longer option12 design as the phased Roth option.
2.2 Reverse-Roth proposal
reverse_roth: tax 100 percent of Social Security benefits while allowing an above-the-line deduction for the employee share of Social Security payroll taxes paid during working years.
The reverse-Roth reform inverts the timing of Social Security’s tax treatment. Current law is effectively a partial Roth structure: payroll contributions are made from after-tax wages (the employee share is not deductible), and benefits are then only partially taxable. The reverse-Roth design moves toward a traditional-account structure: it makes the employee payroll contribution deductible when paid, and in exchange taxes the full benefit when received. Medicare’s Hospital Insurance treatment is unchanged. Because the deduction accrues to workers and the full-inclusion change falls on beneficiaries, the net revenue effect depends on the relative size of the working-age and retired populations in each projection year.
3 Data and methods
3.1 Estimand and unit of analysis
The analysis estimates annual federal revenue and trust-fund effects of each reform from 2026 through 2100. Let \(i\) index households, \(y\) index calendar years, \(w_{iy}\) denote the calibrated household weight for year \(y\), and \(P_0\) and \(P_s\) denote current law and reform scenario \(s\). The static revenue effect reported for scenario \(s\) in year \(y\) is
\[ \Delta^{static}_{sy} = \sum_i w_{iy} R_i(P_s, D_{iy}) - \sum_i w_{iy} R_i(P_0, D_{iy}), \]
where \(D_{iy}\) is the projected household record and \(R_i(\cdot)\) is the PolicyEngine federal income-tax liability calculation. Trust-fund effects are computed analogously for the OASDI and Hospital Insurance (HI) revenue components described below. The static estimand therefore holds the projected population, income, and benefit distribution fixed within a year and varies only the policy parameters.
The labor-supply response estimand uses the same projected data and baseline contract but lets the policy reform trigger PolicyEngine labor-supply response variables before revenue is measured:
\[ \Delta^{LSR}_{sy} = \sum_i w_{iy} R_i(P_s, D_{iy}, L_i(P_s)) - \sum_i w_{iy} R_i(P_0, D_{iy}, L_i(P_0)). \]
The behavioral component \(L_i(\cdot)\) is limited to the labor-supply response layer. The model does not endogenously reproject population, wage growth, interest rates, benefit claiming, mortality, or the Social Security trust-fund balance in response to a reform.
3.2 Base microdata
The microdata foundation is PolicyEngine’s populace 2024 database (populace-us-2024-9f1260b-20260611), a primary-source-built household file that supersedes the Enhanced CPS used in the original report. populace assembles its layers from the CPS Annual Social and Economic Supplement (U.S. Census Bureau 2024), the IRS Public Use File, the Survey of Consumer Finances, the Survey of Income and Program Participation, the CPS Outgoing Rotation Group, the Medical Expenditure Panel Survey, and the American Community Survey, with every imputed layer traceable to its source (PolicyEngine 2026a). PolicyEngine then applies the federal income-tax, payroll-tax, Social Security benefit-taxation, and benefit-program rules needed for this analysis. Relative to the Enhanced CPS, populace produces a more realistic income-tax baseline: federal income tax on the calibrated populace baseline sits near 8–11 percent of GDP across the projection, in line with historical experience, whereas the earlier reweighting-heavy construction pushed it well above 20 percent at the far horizon.
The projection and scoring stack uses households as the aggregation unit. For variables originally defined at person or tax-unit level, PolicyEngine maps values to households before calibration or national aggregation. This household-level contract matters because the taxation of Social Security benefits depends on tax-unit filing status, combined income, Social Security benefits, and other income sources, while the projection weights are stored on household records.
3.3 Long-run projection
Each projection year is built independently from the same 2024 populace cross-section, following one organizing principle: demographics enter through the household weights, and economics enter through the record values. The construction has four stages.
Stage A — grow incomes to the target year. Each income category is uprated from 2024 to year \(y\). Through 2034 the uprating follows the Congressional Budget Office long-term vintage carried in PolicyEngine’s parameter system; beyond 2034, any category whose cumulative growth would outrun the Trustees nominal-GDP path is capped at that path, so no income source grows faster than the projected economy. Tax parameters follow statutory indexing, with the threshold treatment described below.
Stage B — demographic reweight. A light positive-entropy adjustment shifts household weights to the 2026 Trustees single-year age distribution. The weights carry population level and age structure only; they are not asked to repair dollar aggregates.
Stage C — value calibration, given those weights. Holding the reweighted population fixed, three scalars align the economic aggregates with the Trustees targets: \(\alpha\) rescales earnings inputs so SSA taxable payroll (wages capped at the taxable maximum plus taxable self-employment income in remaining cap room) matches the Trustees taxable-payroll target; \(\beta\) rescales Social Security benefit inputs so total benefits match Trustees OASDI cost; and \(\gamma\) rescales beneficiary households’ non-earnings, non-benefit income toward the Trustees taxation-of-benefits target. Because the 85 percent inclusion cap saturates at the far horizon, total taxation of benefits becomes nearly inelastic to other income; \(\gamma\) is therefore bounded and the final calibration closes any remainder.
Stage D — final entropy calibration. A last weight adjustment hits the age distribution and every fiscal target exactly, with guard constraints holding investment and other non-payroll income to their Stage-A growth paths so the reweighting cannot inflate income to reach a revenue target.
Separating the two channels keeps both honest. Reweighting alone — the original report’s approach — can match aggregates only by distorting who the population is, which is what pushed the Enhanced CPS income-tax baseline far above realistic levels; value scaling alone cannot shift the age structure. The named target source is the 2026 Social Security Trustees current-law projections (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026), with Medicare HI context from the 2026 Medicare Trustees Report (Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds 2026).
The tax-parameter assumption preserves two distinct parts of the Trustees treatment. First, the Social Security benefit-tax combined-income thresholds remain fixed in nominal dollars, consistent with the Internal Revenue Code thresholds used in the Trustees taxation-of-benefits method. Second, the first ten projection years use current-law income-tax parameter indexing through 2034; beginning in 2035, the Trustees assume income-tax brackets rise with average wages rather than chained CPI-U to avoid indefinite bracket compression relative to income levels (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026). The public Trustees Report text names income-tax brackets. SSA/OACT separately clarified to PolicyEngine in May 2026 that the same long-range assumption applies to both income-tax rate brackets and federal income-tax thresholds such as standard deductions.1 We therefore implement the threshold bundle as fixed benefit-tax thresholds plus wage-indexed core federal income-tax thresholds beginning in 2035, covering ordinary income-tax brackets, standard deductions, aged/blind standard deduction additions, capital-gains thresholds, and AMT thresholds. Federal income-tax parameters outside this core threshold bundle, such as credit amounts and refundability limits, remain on current-law indexing unless they are directly changed by a modeled reform. For contributed CRFB reform parameters that are not part of current law, the policy specification controls indexing. In particular, the option11 Social Security credit phase-out thresholds remain fixed at $150,000 for joint filers and $75,000 for other filers. Because the credit is nonrefundable, any gross credit calculated for tax units with zero pre-credit income-tax liability is recorded as unavailable rather than reducing federal revenue.
| Assumption area | Public Trustees source | OACT correspondence | Implementation in this analysis |
|---|---|---|---|
| Social Security benefit-tax combined-income thresholds | The Trustees taxation-of-benefits method treats the statutory benefit-tax thresholds as fixed in nominal dollars. | Our question to OACT stated this understanding; OACT’s reply did not contradict it. | Kept fixed in nominal dollars under current law. |
| Ordinary income-tax brackets | The Trustees Report states that, after the tenth projection year, income-tax brackets rise with average wages rather than chained CPI-U. | Confirmed as part of the long-range assumption. | Wage-indexed beginning in 2035. |
| Standard deduction and related deduction thresholds | Not named separately in the public Trustees paragraph. | OACT clarified that the long-range assumption applies to federal income-tax thresholds, including standard deductions. | Standard deduction amounts and aged/blind additions are wage-indexed beginning in 2035. |
| Capital-gains and AMT thresholds | Not named separately in the public Trustees paragraph. | Treated as threshold-like federal income-tax parameters under the OACT clarification. | Capital-gains thresholds and AMT bracket, exemption, phaseout, and separate-filing thresholds are wage-indexed beginning in 2035. |
| Credit amounts, refundability limits, and other non-threshold parameters | Not specified in the Trustees taxation-of-benefits paragraph. | We asked OACT a narrower follow-up on credits; no public or email clarification is incorporated in the current release contract. | Left on current-law indexing unless a reform changes them directly. |
The projection script builds a year-specific dataset only after it has calculated all calibration variables at household level. The saved H5 sidecar contains the year-\(y\) household weights and year-\(y\) input variables. Pseudo-input variables that are actually aggregates of calculated components are excluded before writing the H5, avoiding stale aggregate values in later rescoring. Each delivered dataset is accompanied by metadata recording the base dataset, calibration profile, target source, calibration audit, and validation status.
3.4 Calibration targets
The current long-run profile is ss-payroll-tob. It calibrates household weights to official demographic and fiscal targets while preserving the household-level tax calculation. The profile includes five target families.
Age distribution. The household age-composition matrix has 86 columns: single-year ages 0 through 84 and a top-coded 85+ category. Targets come from SSA single-year age population projections for the Social Security area (Social Security Administration 2024). This keeps the projected microdata aligned to official aging assumptions.
OASDI benefits. The model aggregates the household-level social_security variable and calibrates the weighted total to the 2026 Social Security Trustees OASDI cost projection (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026). This controls the benefit base affected by taxation reforms and benefit cuts.
OASDI taxable payroll. The model combines household-level taxable wages and taxable self-employment income subject to the Social Security payroll-tax base and calibrates the weighted total to the Trustees taxable-payroll projection (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026). This controls the wage base for payroll-tax and solvency comparisons.
OASDI taxation-of-benefits revenue. The household-level tob_revenue_oasdi variable is calibrated to the 2026 Trustees current-law taxation-of-benefits target, derived as the IV.B2 income-rate share of the VI.G1 taxable payroll. This controls the Social Security trust-fund component of baseline benefit taxation.
HI taxation-of-benefits revenue. The household-level tob_revenue_medicare_hi variable is calibrated to the CMS 2026 Medicare Trustees expanded-table HI taxation-of-benefits target, available annually through 2100 (Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds 2026). This controls the Medicare trust-fund component of baseline benefit taxation.
The production delivery path is exact-calibration only for delivered years. The runtime rejects aggregate-quality calibration, rejects IPF fallback, checks that the dataset metadata matches the required target source and profile, and requires all published files to come from validated year-level H5 artifacts.
3.5 Far-horizon support: no synthetic records
A reweighting-heavy construction on the Enhanced CPS can match the far-horizon taxation-of-benefits aggregates only by concentrating weight on a small number of contributors, which would force a choice between thin support and synthetic augmentation. The populace base plus value scaling removes that dilemma. Because the economics enter through record values rather than weights, the taxation-of-benefits-contributing population stays broad even at 2100, and every record in every published year is a real survey household — no synthetic rows.
We verified this directly by building the hardest year, 2100. It passes every publication gate with margin: taxation-of-benefits contributor effective sample sizes of 153 (OASDI) and 127 (HI) against the 50 floor, and a top-10 weight share of 3.4%. All six far-horizon years (2075–2100) pass with contributor effective sample sizes of 107–156.
The CRFB runtime still applies publication gates to both aggregate household weights and policy-target contributors. For OASDI and HI taxation-of-benefits contributors, the gates require at least 1,000 positive contributors, contributor effective sample size of at least 50, top-10 contribution share of at most 50%, top-100 contribution share of at most 95%, and max single-contributor share of at most 15%. The final year-level entropy calibration solves against the real rows and must hit Social Security benefits, taxable payroll, OASDI taxation-of-benefits revenue, and HI taxation-of-benefits revenue exactly. Ordinary nonpayroll-income and preferential-investment-income guard targets pin those income components to their projected growth paths so the reweighting cannot define its own income-tax target.
The data pipeline also carries Trustees covered-worker counts and beneficiary counts as diagnostic references rather than hard calibration targets. We do not treat GDP, AWI, or CPI as calibration targets in this microdata step: GDP is outside the microsimulation model, and AWI/CPI enter as exogenous projection inputs and as the post-2034 growth ceiling described above.
3.6 Calibration estimator
The current named profile uses positive entropy calibration, not the legacy GREG calibration path. Earlier internal notes used “GREG” as shorthand for the second-stage reweighting process, but the delivered ss-payroll-tob profile is implemented as positive entropy calibration. GREG remains in the data repo only as an explicit legacy option for flag-based runs.
Let \(w_i^0\) be the initial household weight, \(a_i\) be the vector of calibration variables for household \(i\), and \(t_y\) be the target vector for year \(y\). The estimator chooses strictly positive weights \(w_i\) that solve
\[ \min_{w_i>0} \sum_i \left[ w_i \log\left(\frac{w_i}{w_i^0}\right) - w_i + w_i^0 \right] \quad\text{subject to}\quad \sum_i w_i a_i = t_y . \]
This is the entropy-balancing analogue of survey calibration (Deville and Sarndal 1992; Hainmueller 2012). The implementation solves the dual problem. With scaled constraints \(a_i^*\) and targets \(t_y^*\), the dual implies
\[ w_i(\beta) = w_i^0 \exp(a_i^{*\prime}\beta), \]
and the solver finds \(\beta\) such that \[ \sum_i w_i(\beta) a_i^* = t_y^*. \]
The implementation uses a Newton/root solve with line search, records the achieved target errors, and reports the maximum age-bin error, maximum fiscal constraint error, negative-weight share, effective sample size, and top-weight concentration. Because the entropy parameterization is strictly positive, the publication target is a zero negative-weight share. If exact positive calibration is infeasible under the available survey support, the data pipeline can diagnose approximate feasibility, but the CRFB runtime is configured to reject non-exact publication artifacts.
3.7 Policy simulation and trust-fund accounting
For each standard scenario and year, the scoring code loads the calibrated year-\(y\) dataset, runs a baseline PolicyEngine simulation, runs the reform simulation on the same calibrated records, and stores both levels and differences. The core output columns are:
| Output | Definition |
|---|---|
revenue_impact |
Reform minus baseline federal income tax. |
tob_oasdi_impact |
Reform minus baseline OASDI taxation-of-benefits revenue. |
tob_medicare_hi_impact |
Reform minus baseline HI taxation-of-benefits revenue. |
tob_total_impact |
Sum of the OASDI and HI taxation-of-benefits impacts. |
oasdi_net_impact |
Net OASDI trust-fund impact after reform-specific gains and losses. |
hi_net_impact |
Net HI trust-fund impact after reform-specific gains and losses. |
The trust-fund allocation depends on reform design. Direct benefit-taxation components use the model’s OASDI and HI taxation-of-benefits variables. Senior deduction or credit components can change federal income-tax liability without crediting all of that change to the trust funds. For option7, which only eliminates the temporary senior deduction, total revenue remains the full federal income-tax impact, while OASDI, HI, and general-fund columns show the accounting split. Employer-payroll-tax swap options add the modeled income-tax revenue from taxing employer OASDI and Medicare contributions, then subtract lost baseline taxation-of-benefits revenue. The legacy option6 short phase-in allocation assigns early employer-contribution gains to OASDI first until the OASDI share of the payroll-tax rate is filled, then to HI.
This allocation convention is why total federal revenue and trust-fund revenue do not always move one-for-one. A reform can raise individual income-tax revenue while weakening one trust fund, or it can replace taxation-of-benefits revenue with employer-contribution revenue allocated differently across OASDI and HI.
3.8 Scoring-year anchors and the budget-window infill
Reforms are scored exactly at a set of anchor years — 2026, 2030, and every fifth year from 2035 through 2100 — and intermediate years are filled by linear interpolation between adjacent anchors. Each delivered row records whether it is an exact anchor or an interpolated fill, and the long-run and present-value totals aggregate the complete annual series.
Linear interpolation is accurate where a reform’s annual impact is smooth, but it misrepresents a sharp within-window change. The 2025 reconciliation act created several temporary federal deductions — the enhanced senior deduction and the tip, overtime, and auto-loan-interest exclusions — that apply through 2028 and lapse in 2029, alongside a state-and-local-tax cap that reverts after 2029. A reform whose impact depends on those provisions therefore has a step in the budget window that a straight line from the 2026 anchor to the 2030 anchor cannot capture. The clearest case is option7, which only repeals the senior deduction: because the deduction is already absent from current law from 2029 onward, the reform’s true impact falls to approximately zero starting in 2029, yet interpolation between the 2026 and 2030 anchors shows it declining smoothly across the whole window, misstating the budget-window years by as much as $11 billion.
We correct this by adding exact anchors where the schedule actually bends, rather than by scoring every year. Querying the model for every parameter that changes discretely between 2027 and 2034 identifies two federal discontinuities relevant to these reforms: the temporary-deduction sunset at the end of 2028 and the state-and-local-tax reversion at the end of 2029. Exact anchors at 2028 and 2029 bracket both. A third reform, option6, phases its employer-payroll-tax inclusion in annually and caps it at 2033; exact anchors at 2032 and 2033 capture that corner. The remaining phased reforms are either immediate (option5, reverse_roth) or follow a smooth straight-line ramp (option12), so they need no additional anchors. The infill therefore adds four budget-window years for the affected reforms rather than the full annual panel.
3.8.1 Version-consistency requirement
Infill anchors must be built and scored with the identical microdata base, model version, and projection and scoring code as the rest of the panel. If any of those differs, the rebuilt years carry a calibration or code difference relative to their neighbors, producing a discontinuity — a seam — exactly where the new anchors meet the published ones, which would defeat the correction. The published panel was produced with the populace 9f1260b base (populace-us-2024-9f1260b-20260611), policyengine-us 1.700.2 under policyengine.py 4.5.1, and this repository at commit 3de8abb6. The infill years were therefore reproduced from a git worktree pinned to that commit and its locked environment (uv sync against the committed uv.lock), not from the current development head. The head of the repository carries a later, income-tax-calibrated revision of the projection that deliberately shifts the baseline income-tax level by roughly twelve percent; building the infill years there would not match the published anchors. The infill is a targeted interpolation correction on the existing panel, not a rebuild, so it adopts the panel’s own production version.
3.8.2 Seam validation
The reproduction is accepted only after two exact checks. First, a build gate: the rebuilt 2030 baseline must reproduce the published 2030 federal income tax — $3,231.9 billion — to within rounding; the observed deviation is 0.00 percent. Second, a scoring sentinel: published-year reform cells re-scored on the rebuilt baseline must reproduce their published impacts to the dollar. At 2030 the full-repeal, senior-deduction-repeal, and 100-percent-inclusion reforms reproduce −$159.4, $0.0, and $93.7 billion exactly. The sentinel is the comprehensive test, since a difference in base, build code, model version, or scoring code would all surface as a published cell that no longer reproduces. Only after both gates match are the four budget-window anchors scored and merged. The scoring code at the production commit differs from the current head only in cosmetic relabeling and guard logic for thirteen of the fourteen reforms, with one substantive change to the reverse_roth payroll-deduction definition; the infill cells are therefore produced from the pinned worktree end to end, which removes that single open question rather than certifying it.
3.9 Solvency-baseline scoring
Four options (full repeal, 85%, 100%, and the phased Roth-style swap) are also scored against a Social Security solvency baseline — a balanced “traditional fix” in which the program is brought into annual balance from 2035 onward before the reform applies. For each scored year, the solvent baseline is constructed on that year’s calibrated dataset in two stages:
- Benefit reduction. Every benefit is reduced proportionally by half of that year’s OASDI shortfall (income minus scheduled benefits), applied as a uniform benefit multiplier.
- Payroll-rate increase. The remaining OASDI gap is closed by equal employee and employer OASDI rate increases, and the Medicare HI gap by equal employee and employer HI rate adjustments (signed, so an HI surplus lowers rates).
The construction validates that both trust-fund gaps close within tolerance and that the current-law leg of the computation reproduces the published current-law row for the same year before any reform is scored. Reforms are then scored as deltas against the solvent system, with revenue split by baseline trust-fund shares and any residual assigned to the general fund; the split identity (components summing to the total) is asserted for every row.
Solvent scoring is exact in 2035, 2050, 2065, 2075, and 2100 (2065 was promoted from a validation spot-check because the phased Roth-style swap’s split materially differed from a four-anchor interpolation). Intermediate years are direct linear interpolations of the anchor rows. An earlier scheme interpolated the solvent/current-law ratio and scaled the live current-law row; it was replaced because the ratio diverges where current-law totals cross zero, which produced artifacts for the phased Roth-style swap. Years before 2035 are scored against scheduled benefits — the solvency fix coincides with current law until then — so the displayed series runs 2026-2100.
3.10 Present-value convention
Seventy-five-year totals discount each year’s flow to the start of 2026 at each trust fund’s effective interest rates under the Trustees’ intermediate assumptions. OASDI flows use the annual effective yields implied by the compound effective trust-fund interest factors in Table VI.G1 of the 2026 OASDI Trustees Report (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2026), whose reciprocals are the discount factors SSA itself uses for summarized rates and balances. Medicare HI flows use the effective rates reported in Table IV.A4 of the 2026 Medicare Trustees Report (Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds 2026) through 2035, graded linearly to the 4.7 percent ultimate nominal rate by 2040 per that table’s convention, and held constant thereafter. General-fund flows and economy-wide denominators (taxable payroll and GDP) use the OASDI series, and the 75-year total is the sum of the discounted components, so headline figures remain additive across trust funds. Ten-year budget-window totals are nominal sums over 2026-2035. The headline figures the manuscript cites are generated server-side by scripts/build_headline_summary.py and regression-tested against the dashboard’s rendered values.
3.11 Supplemental labor-supply response scoring
Static estimates are the scoring surface displayed on the CRFB dashboard. Supplemental labor-supply response rows are reported in the published results data as a secondary estimand. They are generated from the same current-contract full-H5 reform lineage as the static rows, then aggregated from those durable H5 artifacts. Labor-supply response reform dictionaries merge the statutory reform parameters with the project’s age-based elasticity schedule:
- income elasticity base:
-0.05 - income elasticity multiplier above the model’s older-worker threshold:
2.0 - primary-earner substitution elasticities by decile:
0.31,0.28,0.27,0.27,0.25,0.25,0.22,0.22,0.22, and0.22 - secondary-earner substitution elasticity:
0.27 - substitution elasticity multiplier above the older-worker threshold:
2.0
The doubled older-worker elasticity reflects the literature review judgment used for this project: near-retirement and retirement-age workers are more responsive on the labor-supply margin than prime-age workers. Labor-supply response results therefore should be interpreted as partial-equilibrium revenue estimates under that specific elasticity schedule, not as macroeconomic forecasts. They are not official CBO or JCT scores.
The public dashboard labels these rows as behavioral. Earlier response files were removed because they were not generated from the current full-H5 production lineage; they are not part of the public release.
3.12 Reproducibility and validation
The publication workflow treats reproducibility as part of the method rather than as a post hoc implementation detail. A run is publishable only if the result artifacts can be traced to:
- a locked Python environment from
uv.lock - explicit local
policyengine-usandpolicyengine-us-dataworktrees - a calibrated projected-dataset snapshot
- a machine-readable baseline-dataset manifest and year-level metadata sidecars, each recording its H5 SHA-256
- the populace base release identifier and, when recorded, the resolved blob hash
- git SHAs and dirty-state records for this repo and both sibling repos
- patch artifacts for any dirty local overrides that affected the run
The Modal launch wrapper writes a run-level reproducibility bundle before submitting work. The runtime environment then requires the expected calibration profile, target source, tax assumption, and minimum calibration quality. That fail-closed design prevents a detached batch run from silently mixing lineages.
Validation occurs at three levels. First, each projected H5 must satisfy the metadata and calibration-audit contract. Second, reform rescoring is smoke tested in representative early, middle, and late years before full-panel publication. Third, the assembled dashboard, spreadsheet, and manuscript exhibits are generated from shared result artifacts rather than hand-maintained tables. The validation chapter reports the current status of those checks.
3.13 Limitations
Several limitations remain even after the clean rerun. The base microdata are survey-derived and therefore carry sampling error, item nonresponse, and imputation error. Calibration matches selected aggregate margins, not the full joint distribution of age, earnings, Social Security benefits, filing status, health spending, and assets. Long-run HI expenditures enter the dashboard as an external Trustees-derived denominator rather than as a micro-simulated spending model. Labor-supply response estimates are sensitive to the chosen labor-supply elasticities and do not include macro feedback, portfolio responses, claiming responses, mortality changes, or political feedback from trust-fund exhaustion.
4 Validation and external context
4.1 Internal validation
The long-run package is only interpretable if three layers are simultaneously clean:
- the projected datasets themselves
- the reform rescoring outputs
- the publication artifacts assembled from those outputs
The internal validation framework therefore tests:
- metadata conformance to the intended Trustees contract
- exact-calibration coverage for delivered years
- representative sentinel years across the horizon
- absence of known legacy anomalies in rebuilt standard outputs
4.1.1 Far-horizon calibration evidence
The projected datasets are checked against the calibration contract before reform results are treated as publishable. On the populace base with no synthetic support, all six far-horizon years (2075–2100) clear the exact-calibration and contributor support-gate contract: OASDI taxation-of-benefits contributor effective sample sizes run 131–161 and HI contributor effective sample sizes run 47–121 against the contributor-gate floor, which is applied uniformly across every projection year rather than only the late horizon. In a local static reform smoke test, full repeal of benefit taxation reduced revenue by the baseline taxation-of-benefits amount in every anchor year, and taxing all benefits raised that revenue with the expected sign.
The production run delivers the full fourteen-reform static panel (option1 through option12, the reverse-Roth proposal, and the 93% benchmark) across the annual budget window, five-year long-run sentinels, and the reform-transition junctures, with labor-supply-response multipliers derived from 2026 and 2100 endpoints and interpolated across years. Dashboard annual rows between computed years are display interpolations flagged as such; the computed anchor-year rows remain the cited microsimulation outputs. Legacy special-case series are excluded from the current dashboard, spreadsheet, and paper release surface.
The budget-window infill adds exact full-H5 anchors where federal tax schedules have discontinuities that would otherwise be smeared by interpolation. The largest validation case is option7, which only repeals the temporary senior deduction: the annual panel previously interpolated its static revenue impact from $8.35 billion in 2028 to $4.18 billion in 2029, while the exact full-H5 cells show $18.22 billion in 2028 and approximately zero in 2029, as expected once the deduction has already expired under current law. The same check moves option4 from $24.23 to $34.35 billion in 2028 and from $20.96 to $17.78 billion in 2029; option11 from $25.39 to $35.86 billion and from $22.67 to $19.50 billion; and the option6 payroll-ramp corner from $133.45 to $160.63 billion in 2032 and from $143.82 to $177.51 billion in 2033. Each new row is marked exact_full_h5 and carries its own R2 scenario.h5, metadata, completion marker, and SHA-256 lineage.
4.1.2 Independent estimation cross-check
As an additional guard against systematic error, we cross-checked a sample of headline cells against independent back-of-the-envelope estimates derived only from public aggregates and statutory parameters, with no access to the model’s output. The 2026 current-law taxation-of-benefits total reproduces the published 2026 Trustees and CMS figures directly (the Hospital Insurance component is the CMS expanded-table value to the dollar); the reverse-Roth net effect and the rising share of beneficiaries who owe tax on benefits both landed inside the independently derived ranges, and the share trajectory reproduced the same near-universal-but-capped asymptote from first principles.
The one cell that diverged from a simplified hand calculation — full (100 percent) benefit inclusion — did so for an instructive reason. The hand calculation assumed low-income beneficiaries are shielded from tax by the senior standard deduction and the temporary OBBBA bonus deduction. The microsimulation instead finds the revenue is broad-based across the lower and middle of the income distribution: benefit-reliant retirees with little other income have low adjusted gross income under current partial inclusion, but once the full benefit enters income it exceeds even the enlarged senior deductions, so they cross their deduction floor and owe tax. The household-level calculation captures this composition effect that an aggregate shortcut misses. The cross-check is a sanity bound on order of magnitude rather than an alternative scoring method, but it is a standing pre-publication step.
4.2 External benchmark framework
The original report compared selected reforms against outside estimates from the Congressional Budget Office, the Social Security Trustees, Tax Foundation, and the Joint Committee on Taxation (Congressional Budget Office 2024a; Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2024; Tax Foundation 2024; Joint Committee on Taxation 2024, 2025).
That external spine remains useful in the updated manuscript, especially for:
- full repeal of Social Security benefit taxation
- expanded taxation of benefits relative to current law
- treatment of the temporary bonus senior deduction
- pension-style alternatives to current Social Security taxability rules
Direct comparison still requires care because benchmark organizations often use different budget windows, scoring conventions, and baseline assumptions. The paper therefore treats external comparisons as orientation rather than as a point-estimate horse race.
4.2.1 External comparison
The benchmark layer is best treated as orientation rather than a point-estimate competition. The current rebuilt static package continues to sit within the broad range established by the first report’s benchmark set, but windows and policy baselines are not identical.
4.3 Option 1: full repeal benchmark
| Source | Policy | Scoring | Window | Revenue Impact ($B) |
|---|---|---|---|---|
| PolicyEngine | option1 full repeal | Static | 2026-2035 | -1,658.8 |
| CBO (Congressional Budget Office 2024a) | Full repeal | Conventional | 2025-2034 | -1,600.0 |
| SSA Trustees (Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2024) | Full repeal | Conventional | 2025-2034 | -1,800.0 |
| Tax Foundation (Tax Foundation 2024) | Full repeal | Conventional | 2025-2034 | -1,400.0 |
| Tax Foundation (Tax Foundation 2024) | Full repeal | Macroeconomic | 2025-2034 | -1,300.0 |
4.4 Options 2 and 8: broader taxation benchmark
| Source | Policy | Scoring | Window | Revenue Impact ($B) |
|---|---|---|---|---|
| PolicyEngine | option2 tax 85% uniformly | Static | 2026-2035 | +419.0 |
| PolicyEngine | option8 tax 100% of benefits | Static | 2026-2035 | +896.2 |
| JCT (Joint Committee on Taxation 2024) | Current SS tax expenditure | Conventional | 2024-2028 | +318.4 |
| CBO (Congressional Budget Office 2024b) | Pension-style basis recovery | Conventional | 2021-2030 | +458.7 |
4.5 Option 7: bonus senior deduction benchmark
| Source | Policy | Scoring | Window | Federal revenue (\(B) | TOB impact (\)B) | |
|---|---|---|---|---|---|
| PolicyEngine | option7 eliminate bonus senior deduction | Static | 2026-2035 | +52.4 | +9.7 |
| JCT (Joint Committee on Taxation 2025) | Bonus senior deduction | Conventional | FY2025-FY2034 | -66.3 | n/a |
4.6 Artifact contract
To keep the paper and dashboard from drifting, manuscript exhibits are not hand-maintained inside the prose. Instead, the paper is wired to ingest shared artifact-derived Markdown files under paper/exhibits/. During drafting those files may contain placeholders; before release they should be overwritten by generated content from the same checked artifacts that feed the dashboard and comparison spreadsheet.
The release package should only cite current-contract artifacts that can be traced to the full-H5 selected-panel manifest and dashboard metadata.
5 Results framework
The original public report devoted separate space to household impacts, revenue effects, and external comparisons. The updated manuscript keeps that structure, but it binds those sections to generated exhibits so that the final release can use cleaned current artifacts without rewriting the paper body.
5.1 Release logic
The publication rule is:
- dashboard-facing files should contain current rerun values only
- comparison spreadsheets should not retain legacy rows on the public path
- the manuscript should cite shared generated exhibits rather than duplicate results manually in prose
5.2 Core exhibit families
5.2.1 Headline summary
Headline revenue effects per reform under the dashboard’s default trust-fund split. Ten-year figures are nominal 2026-2035 sums; 75-year figures discount each fund’s 2026-2100 flows at its own Trustees effective interest rates (OASDI: Table VI.G1 factors; Medicare HI: Table IV.A4, graded to the 4.7 percent ultimate rate by 2040), with general-fund flows at the OASDI series. Totals sum the discounted components; the general-fund component is not shown separately and is zero under the default split for every option except the no-senior-deduction option.
Scored against scheduled benefits:
| Reform | 10-year (nominal \(B) | 75-year PV (\)B) | OASDI PV (\(B) | Medicare HI PV (\)B) | ||
|---|---|---|---|---|
| Full repeal | -1,659 | -10,879 | -6,206 | -4,673 |
| Full MAGI inclusion | +201 | +812 | +467 | +345 |
| 85% taxation | +419 | +1,313 | +757 | +556 |
| 85% + deduction | +182 | +801 | +457 | +344 |
| No senior deduction | +52 | +50 | +10 | -1 |
| $500 credit | +221 | +653 | +374 | +278 |
| $700 credit | +248 | +957 | +548 | +409 |
| 90% taxation | +576 | +2,230 | +1,280 | +949 |
| 93% taxation | +671 | +2,785 | +1,598 | +1,188 |
| 95% taxation | +735 | +3,159 | +1,811 | +1,348 |
| 100% taxation | +896 | +4,100 | +2,349 | +1,752 |
| Roth swap | +426 | -110 | +2,115 | -2,226 |
| Short phase-in Roth | +1,173 | +857 | +3,009 | -2,152 |
| Phased Roth | +1,945 | +2,600 | +3,179 | -579 |
| Reverse Roth | -643 | -4,164 | -7,836 | +3,672 |
Scored against the Social Security solvency baseline (2026-2034 against scheduled benefits, solvent baseline from 2035):
| Reform | 10-year (nominal \(B) | 75-year PV (\)B) | OASDI PV (\(B) | Medicare HI PV (\)B) | ||
|---|---|---|---|---|
| Full repeal | -1,638 | -9,640 | -5,494 | -4,146 |
| 85% taxation | +412 | +1,031 | +597 | +434 |
| 100% taxation | +883 | +3,479 | +1,993 | +1,485 |
| Phased Roth | +1,989 | +6,290 | +5,751 | +532 |
These figures are generated server-side by scripts/build_headline_summary.py from the published results panel and the effective-interest-rate series, reproduce the dashboard’s default view, and are regression-tested against it.
5.2.2 Overview results
The current release surface contains the fourteen contract-standard reforms: option1 through option12, the reverse-Roth proposal, and the 93% benchmark. All rows come from the June 12 full-H5 panel on the populace baselines, or display interpolation between those exact anchor-year H5 outputs. Legacy non-contract artifacts are excluded from the dashboard, paper exhibits, and release package.
5.3 Ten-year static revenue impacts
| Reform | 2026-2035 Revenue Impact ($B) |
|---|---|
| option1 | -1,658.8 |
| reverse_roth | -643.5 |
| option7 | +52.4 |
| option3 | +182.4 |
| option4 | +221.0 |
| option11 | +248.1 |
| option5 | +413.5 |
| option2 | +419.0 |
| option9 | +575.9 |
| tax93 | +670.8 |
| option10 | +734.6 |
| option8 | +896.2 |
| option6 | +1,148.3 |
| option12 | +1,908.5 |
In the ten-year window, the largest revenue raisers are option12 (+1,908.5 $B) and option8 (+896.2 $B). The largest revenue reduction is option1 (-1,658.8 $B).
5.4 Terminal-year static impacts
| Reform | 2100 Revenue Impact (\(B) | 2100 TOB Impact (\)B) | |
|---|---|---|
| option1 | -3,319.1 | -3,319.1 |
| reverse_roth | -1,045.8 | +988.4 |
| option12 | -547.8 | -3,319.1 |
| option5 | -547.8 | -3,319.1 |
| option6 | -547.8 | -3,319.1 |
| option7 | +0.0 | +0.0 |
| option4 | +187.1 | +187.1 |
| option3 | +231.3 | +231.3 |
| option11 | +231.3 | +231.3 |
| option2 | +231.3 | +231.3 |
| option9 | +493.3 | +493.3 |
| tax93 | +652.5 | +652.5 |
| option10 | +759.1 | +759.1 |
| option8 | +1,026.6 | +1,026.6 |
At 2100, the strongest positive revenue effects come from broader benefit taxation options such as option8 and option10. The largest revenue reductions are repeal or structural swap options such as option1, option5, and option12.
5.4.1 Full-H5 selected-panel results
5.5 Full-H5 selected-panel static results
The May 22 full-H5 production run completed all 276 exact selected standard-option cells: 12 reforms by 23 years (2026-2035 annually and every fifth year from 2040 through 2100). Exact completed cells have durable R2 scenario.h5, metadata.json, and complete.json markers. Aggregation is downstream from those H5s and is labeled by row in the public CSV.
Dashboard-facing standard-option rows use the exact full-H5 microsimulation result in modeled years. The 624 non-modeled static annual rows after 2035 are linearly interpolated only so charts and 75-year totals remain continuous. The public results surface is results.csv, with scoring_type distinguishing static from labor-supply response rows. No reference baseline substitution, display normalization, or post-hoc TOB calibration is applied in results.csv.
5.5.1 Ten-year static revenue impacts
| Reform | 2026-2035 revenue impact ($B) |
|---|---|
| option1 | -1,905.5 |
| option7 | +55.7 |
| option3 | +81.7 |
| option4 | +113.0 |
| option11 | +161.1 |
| option5 | +214.1 |
| option2 | +244.4 |
| option9 | +398.0 |
| option10 | +553.0 |
| option8 | +709.5 |
| option6 | +1,114.5 |
| option12 | +1,937.6 |
5.5.2 Milestone static revenue impacts
| Reform | 2035 | 2050 | 2075 | 2100 |
|---|---|---|---|---|
| option1 | -257.9 | -507.5 | -1,503.5 | -3,863.5 |
| option2 | +27.1 | +44.1 | +121.5 | +366.7 |
| option4 | +5.4 | +22.8 | +102.7 | +346.9 |
| option5 | +1.5 | -52.6 | -324.5 | -1,401.5 |
| option8 | +87.7 | +155.3 | +470.0 | +1,275.9 |
| option10 | +67.4 | +118.1 | +352.7 | +961.8 |
| option12 | +203.7 | +97.9 | -324.5 | -1,401.5 |
5.5.3 Current baseline diagnostics
Baseline income tax is intentionally shown as the direct full-H5 microsimulation aggregate, not normalized to an external revenue baseline. That raw series is visibly high: it peaks at 251.1% of GDP in 2045. The TOB baseline is much smoother: total TOB is 3,863.5B in 2100, or 2.06% of OASDI taxable payroll. The generated post-OBBBA TOB target remains a diagnostic comparison only; current gaps are exposed in baseline_aggregates.csv.
5.5.4 Late-horizon Roth-family note
The structural employer-payroll swap family (option5, option6, and option12) worsens materially from 2095 to 2100 in the exact static full-H5 rows. For option12, the static revenue impact moves from -597.4B in 2095 to -1,401.5B in 2100. This is not an interpolation artifact; both years are exact selected-year H5 outputs and should remain a release caveat until the late-horizon baseline/provenance issues are fully closed.
5.6 Behavioral endpoint results
The current behavioral endpoint run completed all 28 endpoint cells: 2026 and 2100 for all fourteen current reform rows. Each endpoint cell saved durable R2 scenario.h5, metadata.json, and complete.json artifacts before aggregation. The 1,022 non-endpoint behavioral display rows use documented linear interpolation of behavioral/static ratios between those exact endpoints.
The public combined results.csv keeps static and behavioral rows in one file and distinguishes them with scoring_type. Behavioral endpoint rows use the current v2 populace/TR2026 baseline H5s; non-endpoint behavioral rows are derived only from the documented endpoint-ratio interpolation.
5.6.1 Ten-year behavioral revenue impacts
| Reform | 2026-2035 revenue impact ($B) |
|---|---|
| option1 | -1,918.4 |
| option7 | +56.1 |
| option3 | +81.8 |
| option4 | +113.3 |
| option11 | +161.7 |
| option5 | +225.0 |
| option2 | +244.7 |
| option9 | +399.2 |
| option10 | +555.1 |
| option8 | +712.5 |
| option6 | +1,130.4 |
| option12 | +1,967.9 |
5.6.2 2100 behavioral vs static revenue impacts
| Reform | Static (\(B) | Behavioral (\)B) | Difference ($B) | |
|---|---|---|---|
| option1 | -3,863.5 | -3,892.9 | -29.4 |
| option7 | -0.0 | -0.0 | +0.0 |
| option4 | +346.9 | +347.1 | +0.2 |
| option2 | +366.7 | +367.1 | +0.4 |
| option3 | +366.7 | +367.1 | +0.4 |
| option11 | +366.7 | +367.1 | +0.4 |
| option9 | +657.9 | +660.7 | +2.8 |
| option10 | +961.8 | +967.2 | +5.3 |
| option8 | +1,275.9 | +1,283.8 | +7.9 |
| option5 | -1,401.5 | -1,378.3 | +23.2 |
| option6 | -1,401.5 | -1,378.3 | +23.2 |
| option12 | -1,401.5 | -1,378.3 | +23.2 |
The largest endpoint behavioral movement versus static is option1 in 2100 at -29.4B; the employer-payroll swap family (option5, option6, option12) moves by about +23.2B in 2100. The corrected option7 2100 endpoint is effectively zero, confirming the previous common late-year labor-response drag was an artifact of the wrong behavioral baseline branch.
5.6.3 Cost estimate
Fresh cost reporting should come from the Modal usage page for the current v2 populace/TR2026 apps. The dashboard release artifacts no longer carry the stale May 2026 cost estimate.
5.6.4 Household impacts
The cleaned static rebuild finalized the aggregate reform package before the distributional refresh. The manuscript therefore does not cite legacy household-burden or distributional point estimates as current results.
Distributional exhibits should be regenerated only after a current-contract household aggregation pass is complete.
5.6.5 Revenue and trust-fund impacts
5.7 Milestone revenue impacts
| Reform | 2035 | 2050 | 2075 | 2100 |
|---|---|---|---|---|
| option1 | -223.5 | -442.2 | -1,334.0 | -3,319.1 |
| option2 | +48.9 | +54.2 | +103.4 | +231.3 |
| option4 | +17.5 | +19.4 | +61.3 | +187.1 |
| option5 | +39.3 | +17.2 | -218.0 | -547.8 |
| option8 | +110.0 | +168.1 | +431.0 | +1,026.6 |
| option10 | +89.3 | +129.3 | +320.7 | +759.1 |
| option12 | +215.0 | +150.8 | -218.0 | -547.8 |
The rebuilt standard series shows the expected late-horizon split: repeal or TOB-reducing options (option1, option5, option12) become increasingly costly relative to current law, while broader taxation options (option2, option8, option10) continue to raise revenue.
5.8 Terminal-year trust-fund decomposition
| Reform | 2100 Revenue (\(B) | 2100 TOB (\)B) | 2100 OASDI Net (\(B) | 2100 HI Net (\)B) | ||
|---|---|---|---|---|
| option1 | -3,319.1 | -3,319.1 | -1,805.4 | -1,513.7 |
| option12 | -547.8 | -3,319.1 | +299.4 | -835.4 |
| option2 | +231.3 | +231.3 | +10.4 | +220.9 |
| option4 | +187.1 | +187.1 | -28.6 | +215.7 |
| option8 | +1,026.6 | +1,026.6 | +10.4 | +1,016.2 |
5.8.1 Supplemental labor-supply response
Labor-supply response rows are generated under the current full-H5 reform contract and published as the dashboard’s supplemental scoring surface. Each endpoint cell saves a durable reform H5 first, computed at 2026 and 2100 for all fourteen reforms; aggregates are then derived from those H5s using PolicyEngine/MicroSeries operations, and intermediate annual rows are interpolated from the endpoint ratios. The rows appear in results.csv under scoring_type = behavioral. Static scoring remains the primary surface; labor-supply response results are partial-equilibrium estimates under the project’s age-based elasticity schedule and are not official CBO or JCT scores. Earlier non-contract response artifacts were removed from the release surface.
In the current release, behavioral rows are supplemental to the static scoring surface. They are included only when generated from saved full reform H5 artifacts produced under the current contract.
5.9 Interpretation boundary
The paper structure is now wired to generated static exhibits rather than manual placeholder prose. The remaining manuscript work is narrower:
- refresh the exhibit fragments as artifacts evolve
- regenerate household exhibits once that track is clean
- tighten the abstract and conclusion around the frozen release package
6 Publication boundary and companion surfaces
This repo now has three distinct publication roles.
6.1 1. Interactive dashboard
The Next.js dashboard is the current-results explorer. It is the right place for:
- reform-level comparison
- current static outputs and supplemental labor-supply response outputs after baseline-match validation and full-H5 generation
- direct access to downloadable current data products
It is not the right place to carry the full narrative burden of methods, bibliography, and citation framing.
6.2 2. Citable manuscript
This Quarto paper is the stable narrative layer intended for:
- SSRN submission
- CRFB citation and external reference
- methods exposition
- benchmark framing
- appendices and bibliography
The manuscript should preserve the source trail and substantive chapter coverage of the first report while updating scope to the current CRFB-facing fourteen-reform menu.
6.3 3. Operational documentation
Operational documentation answers a different question: not “what does the analysis conclude?” but “what exactly did we run, validate, and decide?”
Those materials include:
- current methodology and pipeline notes
- reproducibility details
- audit findings and sentinel evidence
- delivery rules governing dashboard versus spreadsheet outputs
6.4 Citation architecture
The clean long-run setup is:
- cite the paper for narrative interpretation and methods
- use the dashboard to explore current scenario outputs
- use the operational docs to audit provenance and rerun decisions
That split keeps the public-facing interpretation surface stable and citable without forcing the dashboard to double as the full methodological record.
6.5 Release discipline
That separation only works if each surface is honest about result status.
- If a result track is not generated from the current full-H5 contract, the paper should say so directly rather than implying it is part of the release.
- Operational notes should carry the detailed rerun state, launch identifiers, and recovery details needed to reconcile those two publication surfaces.
For the current rebuild, that release discipline means the dashboard-facing panel uses the hardened exact-only path on real (no-synthetic) populace records and saved full reform H5 artifacts before appearing in the public result surface.
For the supplemental labor-supply response package, the fourteen-reform panel is public only where generated from durable full reform H5 artifacts under the current contract. Multiplier-derived sample panels and other non-contract response artifacts are excluded from the release surface.
7 References
Footnotes
Email from SSA Office of the Chief Actuary Actuarial Services to PolicyEngine, May 5, 2026, replying to “Clarification on Trustees long-run income-tax threshold assumption.” OACT wrote that the long-range assumption applies to income-tax rate brackets and federal income-tax thresholds, with standard deductions given as an example.↩︎