Speaker: Host

Professor Strauss, Bedau treats R3 as “moral alchemy” and doubts that retributivism can offer a nonconsequentialist general justification. You’ve defended a neo‑Kantian reconstruction that frames R3 as a requirement of moral accounting and respect for agency. Could you spell out that justificatory structure: what is it, exactly, about responsible past wrongdoing that makes intentional hard treatment not only permissible but required, without smuggling in preventive goods? And why, on your view, may the guilty be intentionally harmed while the innocent may not?

On the nature of punishment, Bedau adds A1 and A2. Do you regard hard treatment (harm/suffering) as a constitutive element (A1), or can reparative sanctions without suffering still count as punishment within a retributive framework? Where, if anywhere, do you place communicative/restorative elements—side constraints, substitutes, or integral to retributive punishment?

Turning to proportionality, Bedau argues that once we abandon literal lex talionis, R2 becomes indeterminate or arbitrary. You propose a “moral equivalence metric” that integrates culpability and harm while forbidding instrumentalization. How does your metric avoid the Card and von Hirsch pitfalls he diagnoses (measurement opacity; collapsing seriousness into harm; arbitrary penalty schedules)? Could you walk us through a concrete case—say, provoked killing versus felony murder, or sexual assault with different autonomy losses—showing how culpability and harm are weighted and how the upper and lower bounds are set?

Finally, on capital punishment: you’ve argued narrow eligibility where culpability is maximal with near‑perfect procedures. Given Bedau’s epistemic skepticism and arbitrariness concerns, does your deontic reconstruction leave any real space for the death penalty in practice, or does it collapse into abolition under current conditions? What empirical or procedural thresholds would be necessary to clear your deontic hurdles?

---

Speaker: Prof. Miriam K. Strauss

Thank you. Let me take the questions in sequence.

1. Reconstructing R3 as moral accounting, not moral alchemy

Hart’s label “moral alchemy” makes R3 sound like a crude hedonic balancing: add suffering to wrongdoing and somehow we get moral gold. That caricatures the deontic structure. On my view, culpable wrongdoing is not just a bad state of the world; it is a normative act. In willing an assault or a fraud, the offender sets his agency against another’s standing as an equal. That past act generates directed second‑personal claims: the victim has a claim to be answered; the polity, as the public agent of equal freedom, has a claim to demand that answer; and the offender has a correlative duty to account for what he has done.

Punishment is the institutional form of that moral accounting. It has three elements:

• Authoritative censure that addresses the agent as the responsible author of the wrong.
• A non‑optional burden linked to the wrong, through which the agent bears the cost of his choice rather than externalizing it onto the victim and community.
• A status adjustment that restores the normative equilibrium: it marks that the wrong has been taken up and answered, and that the offender reenters the order of mutual accountability.

This does not smuggle in preventive goods. Even if no one else learns of the punishment, and even if the offender is undeterrable, we still owe the accounting to the victim and to the idea of equal agency. Conversely, because the point is answerability, it explains why only the guilty may be intentionally harmed: culpable wrongdoing changes one’s normative position. By one’s own past will, one has incurred liability to hard treatment as part of answering for that will. The innocent have incurred no such liability. To harm them intentionally for any further good would be to instrumentalize them. So R3, properly reconstructed, is not “suffering-for-its-own-sake”; it is a deontic requirement to settle a moral debt in a way that respects the offender as a responsible chooser.

2. A1 and A2; the place of repair and communication

I accept A2 outright: guilt is the gatekeeper for penal eligibility. On A1, I take “hard treatment” to be constitutive in the following modest sense: punishment must impose a non‑optional burdensome demand that communicates censure. But the burden need not be optimized for suffering. Reparative sanctions can count as punishment when and because they are imposed as authoritative, non‑optional duties of repair that carry the institution’s censure. A well‑designed restitution order, compulsory participation in a victim–offender conference, or a constraint that funds trauma care can be punitive if it is imposed as part of the moral accounting, not merely as a civil remedy or a voluntary good deed.

Communicative and restorative elements are not optional adornments; they are integral to retributive punishment’s point. Without communication and opportunities for re‑acceptance of the norm, hard treatment risks becoming mere pain delivery. At the same time, for grave wrongs, repair alone will rarely discharge the whole debt; some hard-treatment component remains required to register the gravity of the past choice. The structure is “censure plus accountable burden,” with repair able to discharge part of the burden when genuine and appropriately linked to the wrong.

3. Proportionality without talion: the moral equivalence metric

I agree with Bedau that literal talion is untenable, and I share his worry about arbitrariness once talion is gone. My “moral equivalence metric” tries to avoid both Card’s measurement opacity and von Hirsch’s collapse into harm by:

• Two independent cores:

  1. Culpability: reason‑responsiveness at the time of action (purpose, knowledge, recklessness, negligence), quality of reasons (malice, predation), partial excuses (provocation, duress), and impairments that undermine answerability. No strict liability: felony outcomes are graded by risk taken and foreseeability, not by doctrinal imputation.
  2. Harm and risk: not merely physical injury but rights-based losses, including autonomy deprivation, duration, multiplicity, creation of a credible-risk environment, and long-term capability losses. Here I take seriously feminist insights about lived, intersectional impacts in sexual and domestic violence.

• Aggravating relational wrongs: abuse of trust or public office, exploitation of vulnerability.

• Dignity constraints: penalties must not degrade or instrumentalize; they must be consistent with treating the person as an end. This rules out sexualized, mutilative, or terrorizing punishments regardless of “fit.”

• Ranges with uncertainty: we map seriousness to sentencing bands, not points, and we publish the weights and their confidence intervals. Following Prof. Chen’s work, the scale is empirical and revisable: we validate components against community judgments, victim-impact research, and comparative case review, and we flag cases where uncertainty is high so that appellate oversight and reason-giving are mandatory. Judge Villanueva’s structured discretion is then the institutional guardrail.

Two concrete applications

a) Provoked killing vs felony murder

• Provoked killing (adequate provocation; intentional but in heat of passion; no prior cruelty; immediate aftermath confession and assistance): Culpability: baseline “purpose to kill” is the highest level. Significant provocation is a partial excuse because it impairs self-control without negating reasons-responsiveness. I would reduce two levels (from 5 to 3 on a 0–5 culpability scale). Harm: death is maximal (5), with adjustments for multiplicity or torture not present here. Relational aggravation: none. Result: Seriousness places the case in a high but not maximal band. As a concrete guide, a jurisdiction might set a presumptive range of 8–12 years’ incarceration plus mandatory censure and reparative obligations (support to dependents of the deceased, memorial fund), with reasoned departures allowed only on written justification.

• Felony murder (armed robbery with a loaded weapon; death of a bystander during a foreseeable confrontation; no purpose to kill; substantial prior planning for the felony): Culpability: no intention to kill; but extreme recklessness regarding lethal risk in the service of predatory ends merits a culpability of 3–4 depending on the evidence of foresight. I reject strict liability imputation: the grade rises with the degree of risk embraced, not just the felony label. Harm: death (5). Relational aggravation: predation using weapons in public space adds modest aggravation. Result: Often this will sit in a band above the provoked killing when the risk-taking is extreme and chosen for predation (e.g., 12–20 years). Where the lethal outcome was a freak accident against moderate risk, it should fall below (e.g., 7–12 years). The key is transparent grading of risk and foresight.

b) Sexual assault with different autonomy losses

• Stranger rape with weapon, confinement, and threats of further harm: Culpability: purpose to sexually violate; planning; explicit terrorization (5). Harm: profound autonomy deprivation, enduring trauma and credible-risk environment (4–5), plus confinement and weapon use as aggravators. Result: Maximal non-homicide band (e.g., 12–18 years), plus long-term supervision and non-contact orders, alongside reparative obligations. No sexualized penalties; dignity constraints bind.

• Acquaintance assault involving grooming, exploitation of intoxication, and repeated incidents: Culpability: purpose to violate; exploitation of vulnerability and trust (5), with relational aggravation for grooming and repetition. Harm: possibly fewer physical injuries on a given occasion, but greater long-term autonomy loss from repetition and social entanglement; the metric must capture that (3–4 with upward adjustments for repetition). Result: Comparable or higher band than the stranger case if the pattern shows domination over time. Reparative duties are mandatory but cannot displace the custodial core; within-band mitigation is possible where the offender undertakes genuine, victim‑respecting repair and acknowledges wrongdoing.

These illustrations show three features I think Bedau rightly demands: culpability is assessed independently of harm; “seriousness” is not a black box; and mapping to penalties is by transparent bands with reason-giving, subject to appellate review and periodic recalibration.

4. Capital punishment under a deontic reconstruction

Conceptually, the metric leaves narrow space for capital punishment: only where culpability is maximal (purposeful, repeated, or mass killings; terrorism; genocide), no substantial mitigation, and where the community’s most severe hard treatment is necessary to complete the moral accounting without violating dignity. But two further constraints matter:

• Epistemic threshold: the system must achieve a demonstrably vanishingly low risk of wrongful conviction in capital‑eligible cases. At minimum: recorded interrogations; open‑file discovery; exclusion of uncorroborated jailhouse informants; validated forensics; robust defense funding; racially blind charging screens; multiple independent lines of inculpatory evidence (DNA or equivalent); and an institutionalized “residual doubt” safety valve. I would want credible, prospective audits showing an error rate well below one per thousand, with narrow confidence intervals across subpopulations.

• Non‑arbitrariness: capital charging and sentencing must be uniform statewide (or nationally), with centralized charging authority, guideline constraints, written reasons, and automatic, expedited review; documented elimination of race, county, and wealth effects; and execution protocols that avoid both degradation and long delays that themselves become cruel.

On present evidence, these conditions are not met. Error rates are non‑trivial; disparities are persistent; and the combination of delay and death row conditions undermines dignity constraints. So although my deontic account does not rule out the death penalty in principle, in practice it points toward abolition under current conditions. The maximal hard treatment consistent with agency‑respect will be lifelong incapacitation under humane conditions, coupled with an architecture of censure and structured opportunities for acknowledgment and partial repair.

A final note of challenge

Leif’s purely preventive frame cannot explain the asymmetry between guilty and innocent: if harm reduction is the only currency, scapegoating in rare high‑stakes cases is, in principle, eligible. Fingarette’s conceptual internalism, by contrast, risks erasing the difference between authoritative norms and just punishment of breaches. A legal order may conceptually call for censure; it does not thereby morally license the whole menu of hard treatments. The deontic accounting view secures both: it explains why punishment is owed to the past act and why its forms are constrained by dignity. And it gives us a disciplined way, through a transparent metric and institutional guardrails, to avoid the indeterminacy that rightly troubles Bedau.

---

Speaker: Host

Dr. Holm, Prof. Strauss argues that a purely preventive account cannot explain the asymmetry between guilty and innocent—if harm reduction is the only currency, scapegoating sometimes looks eligible in principle. From within a nonretributive framework, what grounds the categorical bar on intentionally harming the innocent and the corresponding permission to impose hard treatment only on those who are guilty (R1/A2), without quietly importing desert? Is it rule‑consequentialist legitimacy, epistemic error costs, or something else? Please be explicit about the normative structure.

On proportionality (R2), Bedau worries that once talion is abandoned, harm‑based scales either collapse into arbitrariness (von Hirsch) or measurement opacity (Card). You favor a pragmatic R2 anchored to social harm and risk, not desert. What would a determinate, preventive proportionality look like? Could you work through the same two contrasts Strauss discussed—(a) provoked killing vs. felony murder; (b) two rape scenarios—and show how a harm‑reduction metric would set ranges, emphasizing incapacitation, risk to reoffend, and marginal deterrence? Where would your recommendations diverge from her moral‑equivalence bands, and why?

Methodologically, do you accept Prof. Chen’s multidimensional seriousness indices and uncertainty intervals as a decision aid for a preventive regime? If so, which dimensions map to preventive goals (e.g., victim capability losses as predictors of social harm; culpability as a proxy for future risk)? If not, what is the alternative measurement strategy that avoids Bedau’s arbitrariness charge?

Finally, on capital punishment: given current evidence on certainty, swiftness, and severity, do you see any credible marginal deterrent or incapacitative case that survives error‑costs and legitimacy losses? If not, say so plainly. And take a stand on Strauss’s closing challenge: can a purely preventive theory vindicate the guilty/innocent asymmetry without help from R3’s moral accounting?

---

Speaker: Dr. Leif Anders Holm

Thank you. Let me be as clear as I can about the nonretributive architecture I’m working with.

1. Guilty/innocent asymmetry without R3

My framework has two layers:

• Telos: reduce future victimization (expected social harm).
• Legitimacy side‑constraints that are not “desert” claims but conditions for the preventive project to be effective and permissible in a liberal order.

Those constraints are justified on rule‑consequentialist and public‑reason grounds, not on moral accounting for past wrongs.

• No intentional harm to innocents. As a standing rule this is a dominance strategy for harm reduction. Once the public believes the state will sometimes sacrifice innocents “for the greater good,” you get cascading losses: less reporting and cooperation with police, lower compliance with the law, jury nullification, and decreased perceived fairness. The empirical literature on procedural justice and legal compliance is straightforward: legitimacy begets cooperation and self‑regulation. Even if you could concoct a “rare scapegoat” case where net harms fall locally, adopting such a practice as a rule destroys the informational and cooperative ecosystem that crime control depends on. Add the epistemic problem: because we can’t identify those “rare” cases reliably ex ante, a carve‑out becomes a mechanism for error and abuse. So the rule “do not intentionally punish the innocent” is not smuggled desert; it’s a system‑level constraint without which prevention fails in equilibrium.

• Why hard treatment may be imposed on the guilty. Not because they “owe” it, but because: (a) they are the locus where incapacitation and specific deterrence have nontrivial expected effects; (b) linking sanctions to actual wrongdoing is required to keep the deterrence signal informative. If sanctions decouple from culpable conduct, the signal‑to‑noise ratio collapses and general deterrence erodes. A2 (eligibility upon guilt) is thus a fairness constraint instrumental to credible, cooperative crime control. Proof standards, due process, and error‑minimizing procedures are part of that same legitimacy bundle.

Call this preventive legitimacy: a package of non‑desert constraints necessary for the state to reduce harm without provoking larger harms through mistrust, error, and abuse.

2. Preventive proportionality (R2) without talion

“Proportionality,” for me, is not a metaphysical fit; it’s the requirement that the incremental severity and modality of a sanction be justified by marginal prevention benefits net of costs. A determinate preventive R2 evaluates:

• Incapacitation benefit: expected averted harm from not being at liberty during the sanction.
• Specific deterrence/rehabilitation: expected reduction in that offender’s post‑sanction risk from programming, supervision, and credible consequences.
• General deterrence: elasticity of the offense to certainty, swiftness, and severity; we exploit the dimensions that actually move behavior (certainty and celerity generally dominate severity).
• Collateral costs: criminogenic effects of imprisonment, family disruption, labor‑market scarring, fiscal costs, legitimacy penalties, and error costs.

We then set ranges that maximize expected net prevention under constraints of equal treatment and basic rights.

Two contrasts, worked through:

a) Provoked killing vs felony murder

• Provoked killing (adequate provocation; no prior violence; immediate surrender; remorse): Risk profile: typically low base‑rate for future lethal violence; event‑driven. Validated tools (e.g., OxRec; HCR‑20) usually flag such cases as lower risk once acute precipitants are removed. Prevention calculus: long incapacitation yields diminishing returns and higher criminogenic costs. General deterrent payoff of severity is weak in heat‑of‑passion contexts; certainty (clear charging and swift adjudication) matters more than additional years. Range: 4–8 years’ custody with front‑loaded, evidence‑based programming (emotion/self‑control, substance abuse if relevant), followed by 8–10 years of intensive supervision, firearms prohibitions, and victim‑safety conditions. Structured restorative options may safely substitute for custody at the low end when validated risk is very low and victims consent. This will be below Prof. Strauss’s bands because my target is future risk, not moral equivalence.

• Felony murder (armed robbery with loaded weapon; death of bystander; substantial planning; no purpose to kill): Risk profile: lifestyle criminality; willingness to take extreme risks for predatory gain; higher forward risk of violent reoffending absent change. Prevention calculus: incapacitation yields more averted harm here; general deterrence is also more responsive: clear weapon/lethal‑risk enhancements reliably shift behavior at the margin (e.g., fewer guns carried during robberies). Range: 10–16 years’ custody, with strong weapon‑use enhancements tied to actual risk imposition (brandishing a loaded firearm in public robberies, not merely felony labels), plus long‑term supervision. Where the lethal outcome was genuinely freakish given modest risk (e.g., unarmed co‑felon in a low‑risk burglary), the range drops (6–10 years). I agree with Strauss in rejecting strict liability; it’s the risk embraced, not the label, that matters. I tend to place high‑risk felony murders above provoked killings because the forward‑looking risk and marginal deterrence effects are greater.

b) Two rape scenarios

• Stranger rape with weapon and confinement: Harm is grave; but for prevention, the key is risk type. Predatory stranger rapists with planning and coercive control cues are the small group with markedly higher sexual and violent recidivism. For them: Range: 12–18 years’ custody with specialized treatment, followed by 10+ years of intensive supervision, exclusion zones, no‑contact orders, and tailored controls. The sanction is not ratcheted for “expressive fit” but for incapacitation and targeted general deterrence (weapon use enhancements). We also fund certainty: investments in forensic capability and victim‑support to raise reporting and clearance.

• Acquaintance assaults involving grooming and repetition: These patterns predict continued offending within networks, even if per‑incident physical injury is lower. The prevention lever is long‑horizon control of access and opportunity, plus behavior change. Range: similar or higher than the stranger case if dynamic risk is high (patterned predation): 10–16 years’ custody plus lengthy supervision and strict contact and digital‑communication controls; mandated treatment. Where validated risk is moderate and victims approve, part of the custodial term can be replaced with intensive, enforceable community‑based regimes that actually reduce reoffending (swift‑certain‑fair supervision; alcohol/drug controls where relevant). Severity beyond these bands buys little prevention; raising certainty and celerity buys more.

Dividing line with Prof. Strauss: I will not pay “moral debt” with extra years when the expected prevention yield is flat. I will, however, maintain a severity floor for high‑harm crimes to preserve the credibility of the norm (legitimacy), but the heavy lifting comes from certainty, swiftness, supervision, and tailored controls.

3. Measurement: from “seriousness” to prevention impact

I’m broadly supportive of Prof. Chen’s multidimensional indices as part of the front end, provided we add risk and cost. For a preventive regime, the relevant dimensions and their roles are:

• Harm severity and capability loss: not to justify pain, but to weight averted harm in the objective function and to prioritize investments in certainty (we are willing to spend more to prevent high‑harm offenses).
• Culpability elements (purpose, planning) as weak proxies for stable propensities and hence forward risk (careful to debias for socio‑economic confounds).
• Risk/needs factors (static and dynamic): strongest predictors of incapacitation benefit and responsivity to interventions.
• Uncertainty intervals: when the scale says “under‑determinate,” we default to narrower bands, require written reasons, and prefer reversible sanctions (community‑based controls) to minimize error costs.

I would operationalize this as a Prevention Impact Score for each sanction option: PIS = expected averted harm (incapacitation + general deterrence + specific deterrence) – expected costs (fiscal + criminogenic + legitimacy + error), estimated with Bayesian models and updated via audits and policy trials. We publish the weights and the audits. That is how we answer Bedau’s arbitrariness charge: by making the tradeoffs explicit, empirically estimated, and revisable.

4. Capital punishment

On present evidence, there is no credible marginal deterrent effect of executions over long prison terms once you control for certainty and swiftness; any small deterrence claims are swamped by error costs and legitimacy losses (including racially patterned disparities) that themselves impede crime control. Incapacitation adds nothing relative to LWOP. There is suggestive evidence of brutalization effects in some jurisdictions. The fiscal cost is higher. So the preventive case fails. My view is unambiguous: abolish.

5. Reply to Prof. Strauss’s challenge

A purely preventive theory can vindicate the asymmetry without desert by committing, ex ante and publicly, to legitimacy‑preserving rules—chief among them the categorical ban on intentionally harming the innocent and the correlating of hard treatment with proved wrongdoing. Those commitments are not window‑dressing; they are part of what makes prevention work in a free society. The “rare scapegoat” hypothetical ignores equilibrium effects and the knowledge problem. If you force me into a catastrophe trolley case where every equilibrium consideration is bracketed, of course my framework looks less satisfying—that’s because it is designed for actual institutional practice, not for thought experiments.

Two gentle challenges in return:

• Deontic “moral accounting” still has to pick numbers. Once you leave talion behind, your bands are ultimately anchored in socially negotiated weights and legitimacy, too. If we both must be transparent about those weights, we should prefer the regime that demonstrably reduces victimization per unit of coercion.

• On capital punishment, your procedural thresholds are so demanding that, in practice, your view converges with mine. If that is right, then we should design our institutions for the world we have: close capital pathways, invest in certainty and swiftness, and reserve the severest deprivations for those subpopulations where risk and marginal deterrence make them preventive rather than symbolic.

---

Speaker: Host

Both Prof. Strauss and Dr. Holm lean on measurement to avoid Bedau’s charge of arbitrariness in R2, but in different ways. You’ve argued for multidimensional seriousness indices with uncertainty intervals. I’d like you to be concrete and rigorous here.

1. Model design. Specify a workable seriousness construct S with distinct components for (a) culpability (purpose/knowledge/recklessness/negligence; partial excuses; relational aggravators), (b) harm and risked harm (including autonomy loss, duration, multiplicity, credible-risk environments), and (c) abuse of trust or public office. How would you estimate weights (e.g., IRT/Bayesian hierarchical modeling)? What normative guardrails keep S from collapsing into mere harm (Sellin–Wolfgang’s problem) or into opaque judgment (Card’s Full Measure)?

2. Uncertainty and bands. Show how to produce 95% intervals around S and then map S to sentencing bands rather than points. What trigger rules would you recommend when intervals straddle band thresholds (e.g., default to the lower band; require written reasons; appellate review)?

3. Worked examples (numbers welcome). Using stylized but realistic inputs, compute S (with intervals) for the two pairs discussed: a) Provoked killing (adequate provocation; immediate surrender; no prior violence) vs. felony murder (loaded weapon in a planned robbery; bystander killed; clear foresight of lethal risk). b) Sexual assault cases: (i) stranger rape with weapon, confinement, threats; (ii) acquaintance assault involving grooming, intoxication exploitation, repeated incidents. Make explicit how culpability and autonomy-loss components differ and how that affects S and bands.

4. Validation and democracy. Outline a validation pipeline: community surveys (including survivor samples), victim-impact research, recidivism forecasts, and cross‑jurisdictional calibration. How often should weights be re‑estimated? How do you prevent the scale from encoding racial/class bias (e.g., by excluding socio‑economic proxies and auditing disparate impact)?

5. Integration with preventive aims. Dr. Holm wants a Prevention Impact Score; Prof. Strauss seeks a moral‑equivalence ceiling. Can your S serve both—e.g., as a desert‑anchored upper bound and as an input to prevention calculus? If so, state the interface: S caps severity; risk/needs govern the point within the band. If not, say why.

Finally, address Bedau’s worry directly: why isn’t your banding still ultimately arbitrary? What features of transparency, empirical estimation, and periodic democratic revision make it the least arbitrary feasible scheme under non‑ideal conditions?

---

Speaker: Prof. Wei-Lun Chen

Thanks, Chair. Let me put my cards on the table: R1 is nonnegotiable; R2 is indispensable but underdeterminate; R3 doesn’t carry justificatory water. My project is to turn the “fit” problem into a measured constraint with error bars that institutions can use responsibly.

1. Model design: a workable seriousness construct S

Structure. S is a latent construct composed of three separable pillars, each itself multidimensional:

• Culpability C (what the agent was responsible for choosing)

  • Mental state: purpose, knowledge, extreme recklessness, recklessness, negligence (ordered categories).
  • Quality of reasons: predation, cruelty, opportunism vs. panic, mixed motives.
  • Partial excuses: provocation (adequate), duress (non-exculpatory), severe but non-exculpatory impairment.
  • Reasons-responsiveness: planning, preparation; availability of lawful alternatives.
  • Exclusions: no strict liability; prior record is not part of C (that’s risk, not instant culpability).

• Harm and risked harm H,R (what was imposed on or credibly risked to others)

  • Realized harm H: death; serious injury; sexual/autonomy invasion; confinement; psychological trauma; duration; multiplicity; capability losses (work, family, mobility).
  • Risked harm R: ex ante lethal/serious-injury risk embraced (weapon brandishing, discharge); creation of credible-risk environments; endangerment to multiple persons; public-space risk externalities.
  • Feminist/ victimological content: autonomy-loss is not a proxy for medical injury; it includes terrorization, grooming, and long-term constraints on life plans.

• Relational aggravation T (abuse of normative position)

  • Abuse of trust or public office; fiduciary/caretaker roles; exploitation of dependency or vulnerability (functionally defined).

Estimation. We estimate S with a Bayesian hierarchical IRT framework:

• Items. Each of the above subcomponents is operationalized as polytomous items (graded-response models). Example: “mental state” is an ordered item; “provocation” is a partially crediting item that reduces C.
• Latent dimensions. We fit a multi-dimensional IRT where C, H, R, and T are separate latent traits. S is then a weighted linear composite of the four latent scores, re-scaled to a 0–100 index.
• Weights. We place weakly informative priors on the weights α = (αC, αH, αR, αT), with normative guardrails:
  • αC ∈ [0.35, 0.60] (culpability must matter at least as much as harm alone).
  • αH ∈ [0.30, 0.50].
  • αR ∈ [0.05, 0.25] (risked harm has independent, nontrivial weight).
  • αT ∈ [0.05, 0.15].
  • αC + αH + αR + αT = 1.
• Identification/anchors. We anchor the scale with a small set of paradigmatic offenses (e.g., aggravated stranger rape, armed robbery causing no injury, intentional homicide without torture) to fix the metric. We also enforce monotonicity constraints (worse item responses cannot lower the latent trait).

Data sources. To calibrate item difficulties/discriminations and weights:

• Deliberative community panels (probability samples plus survivor oversamples) doing pairwise/best–worst seriousness rankings on standardized vignettes.
• Victim-impact research (standardized instruments for autonomy/capability loss).
• Comparative sentencing panels (judges/prosecutors/defenders) to check face validity and to stress-test invariance.

Guardrails against the classic failures:

• Against Sellin–Wolfgang’s collapse into harm: culpability is its own latent dimension with a minimum weight floor; we measure risked harm separately from realized harm; and we forbid proxies for outcomes (e.g., medical bills) that smuggle socioeconomic status into H.
• Against Card’s opacity: item content, thresholds, discrimination parameters, and α-weights are published; code is open-source; and sensitivity analyses show how S moves if you vary α within the permitted priors.

2. Uncertainty and bands

Uncertainty. For each case, the model yields a posterior for each latent (C, H, R, T) and hence for S. We report the posterior mean Ŝ and a 95% credible interval [S5, S95].

Banding. We partition S into seriousness bands with published cutpoints tied to anchor offenses. Illustratively:

• Band 7: 85–100 (maximal non-capital)
• Band 6: 75–85
• Band 5: 65–75
• Band 4: 55–65
• Band 3: 45–55
• Band 2: 35–45
• Band 1: <35

Trigger rules when intervals straddle thresholds:

• Lenity default: if [S5, S95] spans two bands, start in the lower band.
• Reasoned elevation: an upward move into the higher band requires written reasons that identify specific, model-recognized factors or documented miscodings; automatic appellate review.
• Indeterminacy flag: if [S5, S95] spans more than two bands, treat as underdeterminate—convene a panel, prefer reversible sanctions, and narrow the decision with additional fact-finding if possible.

3. Worked examples (stylized numbers)

For concreteness, set αC=0.45, αH=0.35, αR=0.12, αT=0.08 (within guardrails). Component scores are on 0–10 scales; S rescaled to 0–100.

a) Provoked killing vs felony murder

• Provoked killing

  • C: purposeful killing (10) minus adequate provocation (−2) minus immediate surrender/remorse (−0.2 to −0.5): C ≈ 7.8 ± 0.6.
  • H: death: H ≈ 9.6 ± 0.3.
  • R: low ex ante risk creation beyond the act itself: R ≈ 2.0 ± 0.7.
  • T: none: T = 0.
  • Ŝ = 0.457.8 + 0.359.6 + 0.122.0 + 0.080 = 7.11 → 71.1 (± 3.0 SE → 95% ≈ [65.2, 77.0]).
  • Band: 5 by lenity (interval straddles 5/6).

• Felony murder (planned armed robbery; loaded weapon; clear foresight of lethal risk; bystander killed)

  • C: extreme recklessness re: life, predatory plan: C ≈ 6.8 ± 0.7 (no intent to kill).
  • H: death: H ≈ 9.6 ± 0.3.
  • R: high ex ante lethal risk embraced (public gunpoint robbery with loaded firearm): R ≈ 8.5 ± 0.6.
  • T: none: T = 0.
  • Ŝ = 0.456.8 + 0.359.6 + 0.12*8.5 = 7.44 → 74.4 (± 3.4 SE → 95% ≈ [67.8, 81.0]).
  • Band: 5 by lenity (interval straddles 5/6), with transparent reason to sit toward the top of Band 5 or to depart upward with reasons: the R component is doing work that a pure harm scale misses.

Observation. The ex ante risk embraced pushes felony murder above provoked killing even though both have H=death, because we refuse to collapse seriousness into realized harm.

b) Sexual assault: stranger armed rape vs acquaintance grooming/repetition

• Stranger rape with weapon, confinement, threats

  • C: purposeful sexual violation with planning: C ≈ 8.5 ± 0.6.
  • H: autonomy deprivation severe; confinement; credible ongoing risk; trauma: H ≈ 8.8 ± 0.6.
  • R: high ex ante risk (weapon, threats): R ≈ 7.5 ± 0.6.
  • T: none: T = 0.
  • Ŝ = 0.458.5 + 0.358.8 + 0.12*7.5 = 7.81 → 78.1 (± 3.5 SE → 95% ≈ [71.2, 84.9]).
  • Band: 6 (interval may just nick Band 5; reasons can resolve).

• Acquaintance assault with grooming, intoxication exploitation, repeated incidents

  • C: purposeful violation; grooming; repetition: C ≈ 8.5 ± 0.6.
  • H: per-incident physical injury lower but autonomy loss amplified by repetition and social entanglement; enduring capability loss: H ≈ 8.2 ± 0.6 (with repetition modifier).
  • R: moderate ex ante risk to physical integrity; high risk of continuing victimization within network: R ≈ 6.0 ± 0.7.
  • T: abuse of trust/vulnerability: T ≈ 2.0 ± 0.5 (on a 0–3 scale).
  • Ŝ = 0.458.5 + 0.358.2 + 0.126.0 + 0.082.0 = 7.58 → 75.8 (± 3.6 SE → 95% ≈ [68.8, 82.7]).
  • Band: 6 or high 5; repeated domination and relational abuse can place this as serious as, or more serious than, a one-off stranger case—capturing Prof. O’Rourke’s concern.

Mapping bands to dispositional ranges (illustrative, not prescriptive):

• Band 7: 15–25 years (max non-capital)
• Band 6: 10–15 years
• Band 5: 7–12 years
• Band 4: 3–8 years Ranges are ceilings for retributive proportionality; within-band placement is governed by prevention (see 5).

4. Validation and democracy

Pipeline.

• Elicitation. Biennial community seriousness surveys with stratified random sampling; oversample survivors and marginalized communities; tasks: paired comparisons, best–worst scaling, and vignette rating.
• Victim-impact science. Longitudinal studies of autonomy and capability losses after different crimes; integrate as H indicators.
• Judicial calibration. Periodic structured elicitation from judges and practitioners to identify divergences between S and reflective practice.
• Cross-jurisdictional calibration. Multi-state pooling with hierarchical priors; allow local random effects while keeping national anchors.

Re-estimation cadence.

• Annual incremental updates (Bayesian updating on new data).
• 5-year major re-estimation and public hearing cycle; sunset clause on weights unless reauthorized.

Bias prevention.

• Exclude protected-class proxies from S (race, neighborhood, income, immigration status). S is about the instant offense, not the person’s social location.
• Define “vulnerability” functionally (custodial dependence, disability) and apply symmetrically to avoid valuing some victims’ harms less.
• Differential item functioning (DIF) tests across groups; drop or adjust items that behave differently without principled reason.
• Disparate impact audits: simulate sentencing under S; publish group-level effects; require justification and revision if disparities appear that cannot be normatively defended.

Governance.

• Independent Sentencing Metrics Commission with members drawn from survivors, formerly incarcerated persons, line defenders/prosecutors, judges, statisticians, and ethicists.
• Open data and code; redaction protocols to protect privacy.
• Notice-and-comment for any change to α-guardrails or band cutpoints; legislative ratification for major shifts.

5. Integration with preventive aims (Strauss ceiling; Holm PIS)

Yes—S can serve both.

• Step 1 (Desert cap/floor). Compute S and its band with intervals. That band is the permissible severity window as a matter of proportionality (upper bound; you may also set a desert floor for norm-signaling).
• Step 2 (Prevention choice). For options within the band, compute Dr. Holm’s Prevention Impact Score: PIS(option) = E[averted harm via incapacitation + general deterrence + specific deterrence] − E[costs: criminogenic, fiscal, legitimacy, error]. Choose the option maximizing PIS subject to dignity constraints (no degrading penalties) and the S ceiling.
• Step 3 (Reason-giving and review). Document which factors drove S and PIS; require appellate review for out-of-band moves or when uncertainty flags fire.

This interface preserves Prof. Strauss’s side-constraint role for desert, while enabling Dr. Holm’s forward-looking optimization inside the constraint set.

Why this isn’t “still arbitrary” (answering Bedau)

Three reasons.

• Structural transparency and constraint. We do not hide value choices: α-weights, item content, cutpoints, and code are public; α has guardrails that entrench the independent role of culpability and risked harm. Dignity constraints forbid certain penalty types regardless of S.

• Quantified uncertainty and lenity. We never pretend to precision we don’t have. Intervals are reported; when underdetermination is material, we default to the lower band, require reasons to go higher, prefer reversible sanctions, and flag cases for appellate scrutiny.

• Democratic revisability with audit. The scale is not a technocrat’s oracle; it is a measured instantiation of community judgments and victim research, subject to regular re-elicitation, bias audits, and legislative oversight. That institutionalizes contestation, which is the only honest cure for indeterminacy in a diverse polity.

Two friendly challenges back to colleagues:

• Miriam, once you publish your “moral equivalence” weights and bands, you will face the same underdetermination. My proposal is simply to own it statistically and govern it institutionally.

• Leif, if prevention alone sets severity, the temptation will always be to sneak priors about desert into “legitimacy costs.” Using S explicitly as a ceiling/floor keeps that accounting candid and democratically supervised.

Bottom line. We cannot abolish judgment from proportionality; we can domesticate it. A multidimensional, audited, and revisable S with uncertainty bands gives us the least-arbitrary feasible map from wrongs to permissible punishment under non-ideal conditions. It avoids Sellin–Wolfgang’s harm collapse, Card’s opacity, and von Hirsch’s point estimates—and it gives Judge Villanueva exactly what she asked for: structured discretion with reasons and review.

---

Speaker: Host

Judge Villanueva, you’ve asked for structured discretion with reasons and review. You’ve now heard Prof. Chen propose a multidimensional seriousness index S with uncertainty bands and lenity triggers, and Dr. Holm propose a Prevention Impact Score (PIS) to choose within-band sanctions. From a guideline designer’s perspective, please get concrete about institutionalization:

1. Guideline architecture. How would you translate S-bands into a workable grid? Specify: (a) default ranges per band; (b) departure rules when the 95% interval straddles thresholds; (c) the standard of proof and trier for S-affecting facts (e.g., autonomy-loss magnitude, risked harm)—jury beyond a reasonable doubt, or judge by clear and convincing?

2. Reason-giving and review. Draft the core elements you would require in a sentencing statement that uses S and PIS (e.g., identify the items driving C, H, R, T; disclose the credible interval; explain any upward move despite lenity; articulate the PIS tradeoff). What should be the appellate standard of review for (i) S computation; (ii) band selection; (iii) within-band choice?

3. Integrating reparative options. Prof. Strauss treats reparative duties as potentially punitive when imposed as censure-backed, non-optional burdens; Prof. Qureshi would go further and see repair as presumptive. Within your mixed theory, where and how would you authorize validated restorative dispositions within a band without violating equal-treatment expectations? What screening (victim consent? risk tools?) and what backstops (swift consequences for noncompliance) would you build in?

4. Managing bias and background injustice. Dr. Baptiste will press the Murphy critique: any proportionality scheme risks laundering structural inequality. What concrete auditing and remedial tools would you write into the guidelines to detect and correct disparate impact (e.g., periodic bias reports, automatic variance for poverty-driven offenses, forbidding certain proxies)? Who has standing to challenge the metric or its application on equality grounds, and what remedies are available (remand vs. systemic revision)?

5. Due process and democracy. Should S’s component findings (e.g., abuse of trust, relational vulnerability, autonomy-loss severity) be elements of the offense (jury), sentencing factors (judge), or some hybrid? If hybrid, draw the line and justify it. Would you require periodic public hearings and legislative ratification for changes to S weights and cutpoints, as Prof. Chen suggests?

6. Worked example. Please walk through a sample sentencing statement using S and PIS for the felony-murder scenario (armed robbery with loaded weapon; bystander killed; strong foresight of lethal risk; no intent to kill), showing: (i) S inputs and interval; (ii) band selection under the lenity rule; (iii) within-band choice based on PIS (incapacitation benefit, general deterrence from weapon-use enhancements, criminogenic costs), and (iv) any restorative conditions you would add.

7. Capital punishment. Given your mixed view—desert as side-constraint; harm reduction as justification—do Chen’s uncertainty intervals and Holm’s error-costs/legitimacy analysis push you toward a practical moratorium on capital charging? If not, what charging and review structure would be necessary to pass your error-cost and non-arbitrariness tests (e.g., centralized charging, heightened proof standards, residual doubt review)?

Finally, be candid about tradeoffs: where does institutional administrability force you to simplify Chen’s model, and what safeguards would you put in to keep simplifications from sliding back into Bedau’s arbitrariness?

---

Speaker: Hon. Elena P. Villanueva

Thank you, Chair. I’ll be concrete and, where necessary, blunt about tradeoffs.

1. Guideline architecture

• Grid design tied to S-bands

  • Map Prof. Chen’s S to seven seriousness bands. For administrability, each band has a narrow default range, plus a “supervision tail” and standard collateral orders.
    • Band 7 (85–100): 15–25 years custody; 10–15 years supervision.
    • Band 6 (75–85): 10–15 years custody; 8–12 years supervision.
    • Band 5 (65–75): 7–12 years custody; 5–10 years supervision.
    • Band 4 (55–65): 3–8 years custody; 3–7 years supervision.
    • Lower bands scale accordingly.
  • Within each band, the Commission publishes approved “modality menus” (custody; split custody; intensive community sanctions packages) so judges choose among equivalent options, not invent them ad hoc.

• Departure rules when the 95% interval straddles thresholds

  • Lenity default: if the credible interval crosses a band line, start in the lower band.
  • Upward move despite lenity requires: (a) identification of the specific, model‑recognized facts that would resolve the uncertainty upward; (b) clear and convincing proof of those facts; (c) written reasons; and (d) automatic appellate review.
  • If the interval spans more than two bands, treat as underdeterminate: convene a three‑judge sentencing panel or remand for further fact-finding; prefer reversible sanctions (community‑based controls) until the record is developed.

• Trier and standard of proof for S‑affecting facts

  • Hybrid consistent with due process and Apprendi/Blakely principles:
    • Any fact that increases the legally authorized ceiling (i.e., would permit sentencing in a higher S‑band than the offense of conviction otherwise authorizes) must be charged or admitted, and found by a jury beyond a reasonable doubt.
    • Facts that affect placement within the authorized band (e.g., degree of autonomy loss; risk embraced; abuse of trust when it does not raise the statutory maximum) are sentencing factors found by the judge under a clear and convincing standard.
  • To minimize Apprendi friction, the Commission should align major offense degrees with band ceilings, so that most S‑component findings operate within-band.

2. Reason‑giving and review

• Required contents of the sentencing statement

  1. Offense of conviction; applicable statutory maximum and the corresponding authorized S‑band.
  2. The computed S, its components (C, H, R, T), and the 95% credible interval; attach the worksheet showing the item codings used.
  3. Band selection: if lenity controlled, say so. If moving upward across a threshold, identify the specific facts, the evidence standard met, and why the lenity default is overcome.
  4. Prevention analysis (PIS): summarize expected incapacitation benefit, general deterrence considerations (e.g., weapon‑use enhancements with empirical support), specific deterrence/rehabilitation prospects, and collateral costs (criminogenic, fiscal, legitimacy). Explain why the chosen modality and duration maximize prevention within the S‑band and respect dignity constraints.
  5. Reparative conditions (if any): identify which are punitive (censure‑backed, non‑optional) and which are restorative, the victim’s position, and the monitoring/enforcement plan.
  6. Equality/bias checks: certify that no forbidden proxies were used; note any applicable structural mitigators and why.

• Appellate standards of review

  • S computation: legal questions de novo; item codings for which there is evidentiary dispute under clear error; adherence to lenity and trigger rules de novo.
  • Band selection: if within the authorized band, abuse of discretion bounded by the grid; if crossing a band threshold upward, “heightened reasonableness” with independent review of whether the evidentiary standard was met and reasons were adequate.
  • Within‑band choice: abuse of discretion, but the court must engage the PIS tradeoff; silence on prevention is reversible.

3. Integrating reparative options

• Authorization within bands

  • Within any band, the Commission authorizes “validated restorative packages” that are punitive when imposed as censure‑backed, non‑optional obligations (reparations funds, victim‑approved conferencing, community labor linked to harm, mandated treatment, letters of acknowledgment).
  • Equivalency schedule: publish conversion guidelines (e.g., X hours of verified, skilled community work or a structured restitution plan with wage‑garnish plus participation in a mediated process equals Y months of custody), based on evidence of compliance and prevention value—so equal‑treatment expectations are preserved.

• Screening and safeguards

  • Victim consent is required for direct encounters; victims may opt for indirect repair only.
  • Risk screening with validated tools; disqualify high‑risk violent offenders from fully community‑based packages unless additional controls render risk acceptable (electronic monitoring, exclusion zones).
  • Swift‑certain‑fair backstops: graduated, prompt sanctions for noncompliance; clear revocation routes back to custody; a single judge maintains jurisdiction to ensure continuity.

4. Managing bias and background injustice

• Auditing and remedial tools

  • Quarterly disparate‑impact dashboards (race, ethnicity, gender, indigence), published by offense type, S‑band, and modality.
  • Forbid proxies in S and PIS: race, ZIP code, education, employment status, and other socio‑economic markers cannot be used directly or as “risk” stand‑ins.
  • Structural mitigator: a narrowly defined “economic duress” mitigator available where credible evidence shows the offense was substantially motivated by acute material need rather than predation, never applicable to high‑harm violence; reduces C by one step with clear and convincing proof.
  • Automatic equity variance: when dashboards show a statistically significant unexplained disparity within an offense category/band persisting for two consecutive quarters, trigger Commission review; authorize judges to apply a temporary downward variance presumption for the over‑impacted group pending corrective revision of the metric or practices.

• Standing and remedies

  • Standing: defendants; state public defender; attorney general; designated civil rights organizations; victim advocates (for disparate minimization of high‑harm treatment).
  • Remedies: case‑specific remand for resentencing; injunctive relief requiring Commission to adjust items/weights; moratoria on use of particular factors; systemic revision with deadlines; independent monitor where necessary.

5. Due process and democracy

• Elements vs sentencing factors (hybrid)

  • Elements (jury BRD): facts that raise the statutory maximum/band ceiling—e.g., “abuse of public office” if it authorizes Band 7; “use of a loaded firearm in public robbery” if it authorizes entry into a higher band.
  • Sentencing factors (judge C&C): gradations of autonomy loss, risk embraced within the offense definition, relational vulnerability when it does not alter the statutory maximum, genuine remorse/assistance, provocation adequacy (if the legislature has created a partial‑excuse sentencing factor).
  • Justification: preserves the jury’s constitutional role over exposure while allowing judges to do principled within‑band calibration with a higher‑than‑preponderance standard.

• Democratic governance

  • Yes: periodic public hearings and legislative ratification for any change to α‑weights, band cutpoints, or dignity constraints every five years (with interim updates via notice‑and‑comment). Publish data and code. Require a supermajority of the Commission to alter guardrails that protect the independent role of culpability and risked harm.

6. Worked example: felony murder (planned armed robbery; loaded weapon; bystander killed; strong foresight; no intent to kill)

• S inputs (using Chen‑style scoring)

  • C: extreme recklessness regarding human life in service of predatory ends; planning; no intent to kill; no mitigation. C = 6.8 (95% CI 6.0–7.5).
  • H: death. H = 9.6 (9.0–10.0).
  • R: high ex ante lethal risk embraced in a public space with a loaded firearm; multiple persons exposed. R = 8.5 (7.8–9.1).
  • T: no abuse of trust. T = 0.
  • With αC=0.45, αH=0.35, αR=0.12, αT=0.08, Ŝ ≈ 74.4 with 95% CI ≈ [67.8, 81.0].

• Band selection

  • The interval straddles the 75 cutpoint between Bands 5 and 6. Under lenity, Band 5 applies (7–12 years’ custody; 5–10 years’ supervision). The State seeks Band 6; I deny because the upward facts (risk embraced) are already captured in R and do not, on this record, meet the clear and convincing standard to overcome lenity.

• Within‑band choice (PIS)

  • Incapacitation benefit: actuarial indicators (prior weapon carrying, lifestyle criminality) suggest elevated forward risk; incapacitation yields nontrivial averted harm during custody.
  • General deterrence: weapon‑use enhancements in public robberies have documented marginal deterrent effects (fewer guns carried when the certainty of enhancement is high). Severity beyond the top of Band 5 shows diminishing returns compared to investments in certainty/celerity.
  • Specific deterrence/rehabilitation: credible programming available (CBT, vocational training); post‑release controls (long supervision; exclusion zones; firearm prohibitions).
  • Collateral costs: long custodial terms carry criminogenic and fiscal costs; legitimacy costs are moderated by transparency and avoidance of strict liability.
  • Choice: 11 years’ custody (top of Band 5), plus 8 years’ intensive supervision with zero‑tolerance firearm conditions; mandatory violent‑offender programming; exclusion zones around the robbery corridor; no‑contact orders with co‑offenders; and compliance monitoring.

• Reparative conditions

  • Punitive‑reparative: 10% of prison wages and 10% of post‑release earnings for five years paid into a victim compensation fund designated by the decedent’s family (censure‑backed, non‑optional); a non‑contact acknowledgment letter subject to the family’s wishes; participation in a mediated impact panel if and only if the family consents. Swift‑certain‑fair consequences for noncompliance.

• Sentencing statement (sketch)

  • I would include the S components and interval, the lenity application, the PIS rationale (why 11/8 is the prevention‑maximizing point within the band), and an explicit acknowledgment that the absence of intent to kill places this case in Band 5 rather than 6 under our rules.

7. Capital punishment

Given our mixed framework, Prof. Chen’s quantified uncertainty and Dr. Holm’s error‑cost/legitimacy analysis leave me no responsible space for capital charging in practice. The error rate is not vanishingly low; disparities persist; delay and death‑row conditions raise dignity concerns. I would recommend a practical moratorium.

If a jurisdiction insists on retaining it, the minimum structure would be:

• Centralized statewide charging authority; racially blind charging screens; a requirement of at least two independent incriminating evidence streams (e.g., DNA or equivalent plus corroborated confession or reliable eyewitness).
• Heightened proof: beyond a reasonable doubt plus a “residual doubt” safety valve permitting juries to choose LWOP where any non‑trivial doubt remains.
• Automatic proportionality review; “comparative deserts” analysis statewide; uniform protocols to eliminate county‑level variance.
• Open‑file discovery; recorded interrogations; exclusion of uncorroborated jailhouse informants; validated forensics only.
• Time‑bound appellate review to avoid decades‑long limbo.

I remain skeptical that these conditions can be durably met; absent that, abolition is the candid course.

Tradeoffs and safeguards

• Simplifying Chen’s model
  • We cannot run a bespoke Bayesian model in every courtroom. The Commission should pre‑compute and publish a validated scoring instrument (structured worksheets and a digital tool) that implements the model with a limited, high‑signal set of items; freeze α‑weights for a five‑year cycle; and provide on‑bench guidance tables with pre‑calculated credible intervals for common fact patterns.
  • To avoid sliding back into arbitrariness:
    • Keep bands narrow; require written reasons for any upward movement across a threshold.
    • Mandate disclosure of the S worksheet in every case; prohibit “free‑text” aggravation not mapped to recognized items.
    • Institutionalize appellate policing of reasons; publish reversal rates by judge and county.
    • Require annual bias audits and public hearings; sunset the instrument unless reauthorized after review of accuracy, equity, and usability.
    • Train judges and counsel; certify risk‑assessment vendors; prohibit use of black‑box tools.

Two closing challenges to my colleagues

• To Prof. Strauss: your moral‑equivalence ceiling and my S‑bands converge operationally. The hard question is not whether to cap, but how to write caps so that line judges can apply them without reconstructing Kant on a Tuesday afternoon. Publishing weights, bands, and reasons is the price of legitimacy.

• To Dr. Holm: prevention does the heavy lifting within bands, but I will not allow “legitimacy costs” to become a back‑door desert proxy. Keeping S as a public ceiling/floor is how we keep our forward‑looking calculus from drifting into expediency.

The aim is Bedau’s: mitigate arbitrariness. The tools are structured discretion, narrow bands, written reasons, and review—with humility about what we don’t know, and institutions that can correct course.

---

Speaker: Host

Prof. Qureshi, Bedau’s A1 treats punishment as necessarily hard treatment (harm/suffering), and he worries that once we abandon talion, proportionality drifts. You reject A1 and R3, reframing desert as a claim to answer and repair. I’d like you to make that vision operational and to confront the hardest cases.

1. Conceptual core. If punishment need not be suffering, what makes a reparative disposition genuinely punitive rather than civil? Give necessary and sufficient features (e.g., authoritative censure, non‑optional duties, linkage to the wrong, communicative rituals), and say whether you accept A2 as the gatekeeper. Where, if anywhere, is hard treatment required for grave wrongs, or can repair plus censure fully discharge the state’s response?

2. Proportionality of obligations. Sketch your metric for “proportional repair”: what are the obligation dimensions (material restitution; time‑bound service; apology/acknowledgment; reintegration conditions; community benefit)? How do you calibrate weightings to autonomy loss and culpability without collapsing into Card’s opacity or von Hirsch’s arbitrariness? Could your metric be stated with the same transparency/uncertainty discipline Prof. Chen used for S (e.g., published weights; credible intervals; lenity triggers)?

3. Worked examples (same as others): a) Provoked killing (adequate provocation; immediate surrender; no prior violence) vs. felony murder (planned armed robbery with loaded weapon; bystander killed; clear foresight of lethal risk). What are the core reparative duties in each? What custodial, if any, remains once repair is undertaken? b) Sexual assault: (i) stranger rape with weapon, confinement, terror; (ii) acquaintance assaults involving grooming and repetition. Specify obligations to the victim (if they consent), to the community, and to survivor services. If a victim declines engagement, what substitutes preserve accountability without re‑traumatization?

4. Integration with S and PIS. Judge Villanueva authorized “validated restorative packages” within bands, and Dr. Holm wants prevention to drive within‑band choices. Do you:

• Treat reparative packages as presumptive within any band where validated risk is not high?
• Use Chen’s S as a ceiling and your obligations metric as the primary within‑band selector?
• Or do you propose a parallel track that can substitute for custody even in higher bands where repair is robust and risk is controllable? Please be explicit about screening (victim consent; risk tools), equivalency schedules (what counts as the punitive equivalent of Y months’ custody), and backstops for noncompliance.

5. Equality and net‑widening. Dr. Baptiste will warn that restorative options often favor the resourced and net‑widen the system. What design features prevent that (e.g., sliding‑scale restitution, guaranteed access to mediation/facilitators, exclusion of wealth‑based conditions, audits of disparate uptake)? Who gets to opt in, and how do you stop prosecutors from over‑charging to extract “repair bargains”?

6. Evidence. Point us to the best data on victim satisfaction, recidivism, and community safety from victim–offender mediation and conferencing in serious‑violence contexts. Where are the boundary conditions? What failures should make us revert to incapacitation?

7. Engage the critics.

• Respond to Prof. Strauss’s claim that for grave wrongs “some hard‑treatment component remains required to register the gravity of the past choice.” Can censure plus robust obligations suffice without custodial pain?
• Address Dr. Holm’s legitimacy concern: if reparative outcomes vary case‑by‑case with victim preferences, how do you preserve equal treatment and general deterrence signals?
• And to Judge Villanueva: how would your model fit the due‑process divisions she drew (what facts must a jury find; what counts as a punitive obligation that raises the penalty ceiling)?

---

Speaker: Prof. Saira Qureshi

Thank you, Chair. Let me make the reparative alternative fully operational and meet the hard cases head‑on.

1. Conceptual core: when repair is punishment

• Gatekeeper. I accept A2: penal liability attaches only upon proof of culpable offense.

• Necessary and sufficient features of punitive reparation

  1. Authoritative censure: a formal, public statement that the wrong was committed and condemned, addressed to the offender as answerable.
  2. Non‑optional duties: enforceable obligations imposed by the court; breach triggers swift, proportionate consequences.
  3. Linkage: duties are substantively tied to the specific wrong (its harms, risks, relationships).
  4. Dialogic accountability: structured opportunities to acknowledge, explain, apologize, and listen (directly only with victim consent; otherwise via surrogate/community).
  5. Status work: a defined pathway to reintegration upon completion.

If those features are present, the disposition is penal, not merely civil—even if the primary currency is repair rather than suffering.

• Is hard treatment ever required? Only instrumentally: where risk cannot be responsibly managed in the community or where liberty restrictions are needed to enable safe and credible accountability. Custody is not ruled out, but its point is safety and the conditions for repair—not pain as payback.

2. Proportionality of obligations: a transparent metric

I propose an Obligations Index, O, calibrated to culpability and autonomy loss, with published weights and uncertainty intervals (mirroring Prof. Chen’s discipline).

Dimensions (scored on standardized scales; all duties time‑bound; sliding‑scale for ability to pay)

• Financial redress F: restitution to victims where apt; else payments to survivor services funds; assessed as a share of income to avoid wealth bias.
• Service S: verified, skilled community service linked to the harm (e.g., violence‑interruption, trauma‑support logistics).
• Accountability A: number and depth of facilitated sessions; written acknowledgment; where appropriate, a narrative accepted by the forum as owning the wrong.
• Liberty limits L: curfews, exclusion zones, electronic monitoring, reporting frequency—calibrated to risk and to prevent further harm.
• Change plan B: completion of evidence‑based programs (e.g., violence, sexual behavior, substance use), with relapse‑responsive supports.
• Reintegration R: education/work milestones; relational reparation to impacted communities.

Weights. Publish αF, αS, αA, αL, αB, αR; set guardrails so no single dimension dominates. Tie the O target to an offense‑seriousness profile derived from C (culpability) and A* (autonomy/capability loss), each estimated as separate latent traits with credible intervals. Lenity and written‑reasons rules apply when uncertainty straddles thresholds. Convert O into “burden points” with a public equivalency schedule (e.g., one month of custody equals X burden points = Y hours of service + Z% income share + defined liberty limits); the schedule is a bridge for mixed regimes, not the moral basis of the disposition.

3. Worked examples

a) Provoked killing vs felony murder

• Provoked killing (adequate provocation; immediate surrender; no prior violence; low forward risk) Core duties:

  • F: income‑based payments into a dependents’ support/memorial fund over 8–10 years.
  • S: 800–1,200 hours of service in violence prevention or victim‑support logistics across 3–5 years.
  • A: formal apology; 6–10 facilitated sessions if the family consents, otherwise community panel acknowledgment.
  • L: curfew, exclusion zones, reporting; firearm prohibition; 5–8 years.
  • B: targeted self‑regulation/emotion programming; grief/trauma counseling.
  • R: vocational/education completion. Custody: not presumptively required if validated risk is low and duties are secured; initial residential programming may substitute for a short custodial term where structure is needed.

• Felony murder (planned armed robbery; loaded weapon; clear foresight of lethal risk; bystander killed; elevated forward risk) Core duties:

  • F: income‑based payments to a homicide survivor fund and a gun‑violence prevention fund; longer horizon (10–12 years).
  • S: 1,200–1,800 hours in violence‑interruption/gun‑risk education under tight supervision.
  • A: acknowledgment and community forum; victim‑family options as above.
  • L: stringent liberty controls (curfew, GPS, exclusion of high‑risk zones) for 8–12 years.
  • B: intensive cognitive‑behavioral and employment programs. Custody: front‑loaded incapacitation is often warranted to manage risk and create space for change; but it should be the minimum necessary, with a clear conversion path to community‑based obligations once risk is demonstrably down.

b) Sexual assault

• Stranger rape (weapon; confinement; terror) Core duties:

  • F: payments to survivor services; victim‑directed restitution only if the survivor asks.
  • S: 1,200–1,800 hours in supervised roles not involving access to vulnerable persons; multi‑year horizon.
  • A: only with survivor consent; otherwise surrogate panels. Non‑contact acknowledgment is mandatory.
  • L: robust controls—GPS, exclusion zones, chaperone requirements in public venues, long supervision.
  • B: long‑term, evidence‑based sexual behavior treatment with external clinical audit. Custody: likely necessary initially for safety with high‑risk profiles; replaced by secure, therapeutic residential programs where available; the aim is to move to intensive, enforceable community controls as soon as risk allows.

• Acquaintance assaults (grooming; repetition) Core duties:

  • F: survivor‑services funding; if survivor consents, tailored restitution.
  • S: substantial service and prevention work focused on consent and power dynamics.
  • A: structured acknowledgment; no contact orders; if the survivor declines engagement, accountability is to a community panel.
  • L: strict relational boundaries, social‑media restrictions, exclusion from spaces with targets of prior grooming; long supervision.
  • B: specialized treatment; bystander‑intervention work as part of service. Custody: contingent on validated risk; many such cases can be managed with rigorous community regimes if controls and monitoring are tight.

If a victim declines any engagement, all accountability runs through community forums and written acknowledgment; the burdens remain—form varies, not magnitude.

4. How this integrates with S and PIS

• Presumption. Within any S‑band where validated risk is low or moderate, reparative packages are the default. In higher bands, they are authorized to substitute for some or all custody if O duties reach the band‑appropriate burden points and risk is controllable with L.

• Interface. Use Chen’s S as ceiling/floor for severity; set O targets by band; pick the within‑band mix by prevention (Holm’s PIS). Publish an equivalency table: e.g., Band 5 requires 1,200–1,800 burden points; a package might be 900 hours service + 10% income share for 5 years + curfew/GPS for 4 years + treatment benchmarks.

• Screening and backstops. Victim consent governs direct encounters; validated risk tools gate community placement; noncompliance triggers swift, proportionate sanctions, including conversion of unserved burden points to custody.

5. Equality and net‑widening safeguards

• Eligibility. Limit to cases otherwise facing custody; no adding RJ on top of minor sanctions.

• Access. Independent, publicly funded facilitators; defense and community referrals allowed; prosecutors cannot monopolize the gateway.

• Affordability. Income‑based payments with caps; service hours adjusted for caregiving and disability; never substitute cash for service.

• Over‑charging checks. “No‑worse‑off” rule: opting into repair cannot expose the defendant to a harsher custodial fallback than the standard guideline.

• Audits. Quarterly uptake and outcome audits by race, class, and neighborhood; corrective action if disparities persist without justification.

6. Evidence and boundary conditions

• Victims. RCTs and quasi‑experiments (e.g., UK Justice Research Consortium; Canberra RISE) show markedly higher victim satisfaction and reduced PTSD symptoms after conferencing compared to court; serious‑violence victim–offender dialogues conducted post‑conviction show strong mental‑health gains.

• Recidivism. For violent offenses, several trials found reductions relative to court processing; effects are program‑ and context‑dependent. Adult sexual violence evidence is thinner; what we have supports risk‑responsive treatment plus strict controls, with any dialogic element contingent on survivor consent.

• Boundary conditions for reverting to incapacitation. High, persistent risk despite compliant engagement; repeated breaches of core safety conditions; credible evidence of ongoing coercive control; refusal of all accountability duties. Incapacitation is then justified as a safety measure, not as a quantum of pain.

7. Engaging the criticisms

• To Prof. Strauss: registering gravity does not require calibrated suffering. It requires calibrated accountability. A multi‑year, life‑reordering set of obligations—public censure, sustained service tied to the wrong, income redirection, tight liberty limits, and face‑to‑face answerability where the survivor invites it—can communicate condemnation and restore normative equilibrium more truthfully than warehousing. Pain may occur as a by‑product; it is not the point.

• To Dr. Holm: equal treatment and deterrence are preserved by standardization. The O targets are set by S‑bands, not by victim preference. Victims may shape the form (direct vs surrogate), not the burden. Deterrence flows from certainty of censure and durable consequences, not from ever‑longer custody. We should invest in raising the certainty of reparative accountability.

• To Judge Villanueva: due process fits cleanly.

  • Elements for jury beyond a reasonable doubt: facts that authorize higher S‑band ceilings (e.g., weapon use; abuse of trust) and any aggravator that would permit higher O targets than the offense of conviction otherwise allows.
  • Sentencing factors for judge by clear and convincing: the specific composition of O within the ceiling (hours, payment shares, liberty‑limit intensity), validated risk level, victim consent to direct encounters.
  • Any increase above the default O target for the band requires written reasons and automatic review; victim refusal cannot raise the burden.

A final note. Bedau’s worry about drift once talion is gone is real. The way to avoid it is not to relapse into a proportionality of pain, but to discipline a proportionality of obligations: publish the weights and conversion rules, report uncertainty, default to lenity, and allow democratic revision. Where repair can deliver censure, safety, and restoration, carceral hard‑treatment is presumptively unjust. Where it cannot, we incapacitate to the minimum necessary—and still bind that deprivation to dialogic accountability rather than to suffering as an end.

---

Speaker: Host

Dr. Baptiste, Bedau channels Murphy’s critique that “unfair‑advantage” justifications presuppose a just background order and so cannot legitimate punishment under non‑ideal conditions. We’ve now heard an elaborate technocratic answer to arbitrariness (Chen’s S with guardrails; Villanueva’s structured discretion; Holm’s PIS) and two normative anchors (Strauss’s deontic ceiling; Qureshi’s reparative substitution). I want you to press the structural question and then get concrete.

1. Structural diagnosis. Do S‑bands, PIS, and guideline grids merely launder classed and racialized domination under a veneer of proportionality? Identify specific points where bias enters despite Chen’s exclusions: e.g., item content (autonomy‑loss valuations), offense definitions (property vs wage‑theft asymmetries), policing inputs (weapon use, public‑space risk), and “legitimacy costs” that may smuggle desert back in. Give us a couple of case comparisons where the same social harm distributes differently across race/class (drug markets; survival theft; tax and environmental crimes) and show how the proposed schemes would still misfire.

2. Replacement architecture. If R3 collapses under injustice and R2’s calibration is structurally tainted, what is the alternative justificatory frame? Sketch your preferred minimal penal footprint: which harms (if any) remain within penal sanction, and why? Do you retain A2/R1 as liability gates or move to non‑penal regulation entirely (fines, licenses, civil protection, administrative incapacitation)? Be explicit about the normative principle (e.g., anti‑domination, public health, capabilities) that does the justificatory work.

3. High‑harm violence. For intimate‑partner and sexual violence—the hardest cases—what does a decarceral but safety‑first regime look like in practice? Specify the mix: protective orders, electronic/geofencing controls, income supports and housing for survivors, specialized prevention and treatment, and when (if ever) you still authorize custodial incapacitation. How do you avoid reproducing patriarchal/racial harms through surveillance tools?

4. Restorative justice under non‑ideal conditions. Qureshi proposes “obligations of repair” as presumptive. You’ve warned elsewhere about net‑widening and access bias. What design features are non‑negotiable to prevent RJ becoming a status weapon (e.g., sliding‑scale monetary duties; independent gateways; banning over‑charging to extract repair; community governance)? Where do you draw a hard line that triggers incapacitation instead?

5. Institutional program. Put numbers on it. If you were chairing a reform commission, what decarceration target (percentage and timeline) would you set, which offense categories would you divert first, and how would you reallocate dollars to housing/health/education? What metrics—beyond recidivism—would you use to evaluate success (e.g., community safety, survivor well‑being, reduced racial disparities, material security indices)?

6. Engage the panel.

   • To Chen/Villanueva: if we keep an S ceiling, what additional anti‑domination guardrails must be added (e.g., automatic downward adjustments tied to measures of structural deprivation; presumptive diversion for poverty‑driven offenses)?
   • To Strauss: does “moral accounting” risk re‑inscribing social hierarchies when the material preconditions of agency are unequal? What would a deontic theory look like that conditions punishment on distributive justice audits?
   • To Holm: can “legitimacy costs” be operationalized without importing community sentiments that are themselves racialized? What’s the methodological fix?

7. Death penalty. Give us your structural abolition case in two moves: (a) why capital punishment is inseparable from racialized state violence even under refurbished procedures; and (b) why the resources consumed by capital systems should be reallocated, specifying which survivor‑support and community‑safety programs yield the largest returns.

Please don’t just critique—design. Offer at least two concrete rule changes or statutory provisions you would draft tomorrow that would move us toward your model while addressing Bedau’s charge that theories must speak to actual institutions, not just ideals.

---

Speaker: Dr. Nolan Baptiste

Thank you, Chair. Let me take Bedau’s invitation—really Murphy’s—and keep us in non‑ideal space. Under background injustice, “fair‑play”/unfair‑advantage stories (R3) don’t justify inflicting hard treatment; they presuppose the very distributive floor that’s missing. And a great deal of what we’re calling R2—however refined—travels with background power. I’ll diagnose where, then give you an alternative architecture with rules you can draft tomorrow.

1. Structural diagnosis: where “measured proportionality” launders domination

Even with Prof. Chen’s exclusions and Judge Villanueva’s guardrails, bias seeps in at design, data, and doctrine.

• Item content and valuations:

  • Autonomy‑loss scales are norm‑laden. Whose “credible‑risk environment” counts as severe? A woman in a gated suburb vs a woman in public housing? If panels are demographically skewed or culturally framed, S bakes in middle‑class white sensibilities despite best efforts.
  • “Abuse of trust/office” aggravates for a public official but not for an employer who steals wages or uses ICE threats to coerce. The latter is structurally trust‑violating, yet typically sits outside high bands.

• Offense definitions:

  • Street drug sales are felonies keyed to place, proximity to schools, and “public nuisance”; suburban delivery markets and cryptomarkets with the same volumes are less policed, coded as “non‑public,” and trigger fewer aggravators. S and PIS punish the over‑policed market as “riskier,” entrenching place‑based inequality.
  • Survival theft is a felony at low thresholds in many states; wage theft and tax evasion that dwarf it in aggregate harm are misdemeanors or civil. S can’t fix asymmetrical criminalization.

• Policing inputs and measurement:

  • Weapon‑use and “public‑space risk” aggravators are observed where police saturate. The presence of a gun is detected in Black neighborhoods; the same gun in a white living room is never measured. Risk isn’t just higher; measurement error is nonrandom and racialized.
  • “Community legitimacy costs” risks smuggling back desert through majoritarian sentiments that are themselves racialized. If you weight PIS by “what the public will accept,” you reinscribe who the public imagines as dangerous.

• Case comparisons (same social harm, different class/race):

  • Drug markets: Street‑corner seller (Black, surveilled) with a pistol → S spikes on R (risked harm) and H (credible‑risk environment); PIS favors severity for general deterrence of weapon carry. Private network seller (white, unarmed) with larger volumes → lower S and PIS. Social harm equal or higher in the latter, but the scheme punishes the former.
  • Survival theft vs wage/tax theft: A homeless mother shoplifts $200 of formula, scuffles at the exit (R goes up). A contractor steals $20,000 in wages across 10 workers; a high earner cheats $50,000 in taxes. The latter’s aggregate H is far higher; S (as currently conceived) centers immediate physical autonomy losses and underweights diffuse economic harms. The result is inverted seriousness.
  • Environmental crime: Lead emitter poisoning a census tract vs aggravated robbery with no injury. Population‑level capability loss dwarfs robbery harms. Yet S’s H/R items are built for acute incidents; environmental crimes slip into low bands.

Conclusion: you can sanitize inputs and publish weights, but under non‑ideal criminal codes and policing, S‑bands, PIS, and grids still allocate coercion along classed and racial lines.

2. Replacement architecture: anti‑domination/public‑health frame; a minimal penal footprint

• Justificatory principle: anti‑domination plus public health. The state may use coercion only to prevent ongoing subordination (interpersonal or structural) and serious, imminent physical harm, and only after it has met baseline duties (housing, income, health care) that enable lawful compliance. Coercion must be the least subordinating effective means.

• Liability gate: retain R1/A2 minimally—culpable conduct is necessary for any penal response—but add a distributive precondition: no custodial hard treatment absent a finding that the state provided baseline capabilities reasonably accessible to the person. Exhaustion of non‑penal measures is required.

• What remains penal: a narrow class—(i) ongoing lethal violence; (ii) demonstrable coercive control with high imminent risk; (iii) organized offenses that pose serious bodily harm where administrative controls fail. Everything else shifts to non‑penal regulation:

  • Civil protection and exclusion orders with swift, non‑custodial enforcement.
  • Licensing, bonding, and administrative incapacitation for dangerous practices (e.g., revoke corporate charters; bar licenses).
  • Fines and restitution keyed to income/wealth; debarment; corporate monitorships.
  • Public‑health interventions: overdose prevention, syringe access, MOUD for substance use; housing‑first.

3. High‑harm violence: a decarceral, safety‑first regime

• Core safety stack:

  • Immediate protective orders; firearm relinquishment within 24–48 hours with verified compliance.
  • Subsidized emergency relocation, lock changes, and guaranteed income/housing for survivors; legal advocacy.
  • Electronic/geofencing controls on the respondent, time‑limited and strictly overseen; real‑time response protocols co‑designed with community advocates.
  • Specialized IPV/SA units staffed with clinicians and culturally‑competent advocates; evidence‑based treatment for persons who offend (where they engage) and batterer intervention programs that meet standards.

• Custodial threshold: authorize incapacitation only upon judicial finding, on clear and convincing evidence, of high imminent risk (validated tool plus qualitative evidence) or repeated, willful violations of safety orders; impose the minimum term necessary to break the sequence; pair with mandated treatment; periodic review.

• Avoid reproducing harms:

  • Data minimization (no sharing EM data with immigration or predictive policing); strict sunsets; independent community oversight boards with survivor representation; audits of racialized EM deployment and violation responses; graduated, proportionate responses to technical violations.

4. Restorative justice under non‑ideal conditions: design to prevent status weaponization

• Non‑negotiables:

  • Sliding‑scale monetary duties and caps; service obligations capped per month; never allow wealth to buy down time.
  • Independent gateways: court‑appointed RJ coordinators with authority independent of prosecution; eligibility based on offense/risk criteria, not bargaining leverage.
  • Ban “RJ as leverage”: statutory prohibition on charging up to extract repair; “no‑worse‑off” rule—opting into RJ cannot produce a harsher custodial fallback than the guideline baseline.
  • Community governance: boards composed of survivors and formerly incarcerated persons to set curricula, select facilitators, and audit outcomes; publish demographics of uptake and completion.

• Hard lines for incapacitation: credible lethal threats; stalking with weapons; tampering with monitors tied to safety orders; repeated, escalating violations. Otherwise, repair plus controls are presumptive.

5. Institutional program: targets, sequence, dollars, metrics

• Decarceration target: 50% reduction in prison population in 5 years; 75% in 10.

  • Phase 1 (24 months): presumptive diversion for drug possession; low‑level sales under a threshold; non‑burglary property under a raised felony floor; technical violations of supervision; parole detainers for nonviolent conduct. Goal: 30% reduction.
  • Phase 2 (years 3–5): expand to burglary without injury; felony theft under $5,000; non‑weapon robberies with no injury; shorten lengths across remaining categories with earned‑time. Goal: another 20%.

• Reallocation: assume $50–80k per prison‑year saved. Redirect 60% to housing‑first slots; 20% to mobile crisis and mental‑health; 10% to IPV/SA survivor services; 10% to youth employment/violence‑interruption. Publish a line‑item “justice reinvestment” ledger.

• Success metrics (publicly reported quarterly):

  • Violent victimization rate per 100,000 (disaggregated by neighborhood and race).
  • Survivor well‑being indices (housing stability; PTSD symptom scales; satisfaction with system responses).
  • Racial disparity index in arrests, charging, sentencing, and sanctions.
  • Material security indices (housing stability; poverty rate; youth school attendance).
  • Clearance rates for serious violence (and time to charge).
  • Community trust measures (procedural justice surveys) with oversamples of marginalized groups.

6. Engagements with the panel

• To Prof. Chen and Judge Villanueva: if S stays as a ceiling, add anti‑domination guardrails:

  • Structural Deprivation Adjustment: when a defendant’s neighborhood deprivation index or individual deprivation markers exceed a statutory threshold (homelessness; food insecurity), automatic one‑band downward shift unless the state proves it met baseline provision duties; this is not “excuse,” it’s a distributive precondition.
  • Presumptive diversion for poverty‑driven offenses (codified definition).
  • Require an “Aggregate Harm Module” in S for corporate, environmental, wage, and tax crimes—so aggregate capability losses can place such cases into Bands 6–7.
  • Forbid police‑only data feeds for S items; require corroboration from non‑police sources (medical/clinical, community affidavits) for risked‑harm assessments. DIF tests and disparate‑impact audits must be binding: if bias persists, weights are suspended by rule.

• To Prof. Strauss: your moral accounting presumes agents had fair chances to comply. A deontic theory under non‑ideal conditions must condition punitive eligibility on a Distributive Justice Audit: did the state meet duties that make the norm one we can fairly expect compliance with? If the answer is “no,” only minimal liability (R1/A2) remains—and only reparative, non‑carceral duties are permissible, save imminent safety cases.

• To Dr. Holm: “legitimacy costs” can’t be measured by raw public sentiment. Methodological fix: vignette‑based, blinded surveys that mask race/class cues; reweight samples to correct for racial/SES imbalances; validate against behavior (reporting and cooperation rates) rather than stated attitudes. Publish the models; forbid use of “community fear” metrics derived from media or social media which are racially skewed.

7. Death penalty: structural abolition

• Inseparability from racialized state violence:

  • County‑level lynching rates predict modern death sentencing; the victim’s race drives capital charging; “heinousness” maps onto racialized narratives of dangerousness. Centralized charging and blind screens cannot erase the cultural sediment. Capital punishment functions as a status ritual of domination.

• Reallocation:

  • Each capital case imposes $1–3 million in marginal costs over LWOP. Abolish; reallocate to:
    • Homicide clearance squads (evidence‑based training; forensic capacity). Clearing homicides deters more reliably than executions.
    • Survivor support: long‑term financial stipends; trauma‑informed counseling; relocation funds.
    • Hospital‑based violence intervention; credible‑messenger programs; youth jobs—programs with demonstrated reductions in shootings.

Two draftable rules/statutes for tomorrow

• Distributive Preconditions for Custody Act:

  • “No court shall impose a custodial sentence unless it finds, by clear and convincing evidence, that: (a) the defendant, during the relevant period, had access to baseline housing, health care, and income supports sufficient to meet basic needs; and (b) non‑penal, less restrictive alternatives—including civil protection, reparative obligations, and administrative controls—have been attempted or are demonstrably inadequate to prevent imminent serious harm. Absent such findings, the sentence shall consist solely of non‑custodial measures. An exception lies for cases with a judicial finding of high imminent risk of serious bodily harm.”
  • Pair with a state duty: “The court shall order the responsible agency to provide the baseline supports as a condition of any supervision.”

• Equal Harms Enforcement and Corporate Accountability Act:

  • “For offenses defined by aggregate harms (environmental pollution causing measurable health losses; wage theft above $5,000; tax fraud above $10,000), the seriousness index shall include an Aggregate Harm score reflecting population‑level capability losses; such offenses are eligible for Bands 6–7. The Attorney General shall establish a Corporate Crime Unit; felony liability extends to responsible executives where willful or reckless endangerment is shown. Annual reports shall compare enforcement intensity and S‑band usage across violent and aggregate‑harm categories.”

Add two more clean‑up provisions:

• Presumptive Diversion for Poverty‑Driven Offenses: codify a rebuttable presumption of non‑custodial, reparative sanctions where credible evidence shows material need as primary driver and no injury occurred.
• Technical Violations Reform: prohibit incarceration for technical supervision violations absent a new arrest for violent conduct; require graduated, non‑custodial responses.

Closing thought

Bedau asked for a theory that speaks to institutions. The through‑line of my proposals is simple: under non‑ideal conditions, retribution (R3) is not available, and proportionality (R2) is not neutral. If we keep any penal power, it must be subordinated to anti‑domination and public health; it must be minimized; and it must be conditioned on the state doing its distributive part. If we retain Prof. Chen’s S at all, embed it in that frame—with structural adjustments, distributive audits, and enforcement parity for aggregate harms—lest “doing justice” become the latest language in which we reproduce the very hierarchies punishment claims to condemn.

---

Speaker: Host

Prof. O’Rourke, Bedau flags the opacity of Card’s “Full Measure” and the harm-collapse in Sellin–Wolfgang. You’ve argued that any proportionality metric must capture lived, gendered autonomy losses and credible-risk environments. Given Prof. Chen’s S (with separate latent dimensions for culpability, realized harm, risked harm, and relational aggravation), what’s missing to make it adequate for sexual and domestic violence? Please be concrete:

• Measurement: How should S operationalize autonomy loss (e.g., duration of control, freedom-of-movement constraints, social/financial entanglement, repeat victimization) and credible-risk environments (weapons, stalking, threats to children)? What validated instruments or indicators would you import, and what α-weight adjustments or new items would you mandate to avoid underweighting long-term capability losses vs acute injury? How would you embed intersectional modifiers without reintroducing bias or victim blaming?

• Worked comparisons: For the two scenarios on the table, when (if ever) should the acquaintance assault with grooming/repetition surpass the stranger rape with a weapon in S, and on what dimensions? Offer indicative ranges or item scores and explain the moral salience (e.g., patterned domination, erosion of agency over time). Do the same for IPV with coercive control but low acute injury versus a single severe assault—how should the metric rank them?

• Carceral feminism risks: Where do you draw the line on custodial hard treatment for sexual violence in light of recidivism risks and survivor-defined justice? Specify screening criteria (validated risk tools; indicators of predation vs situational harm), and articulate a presumptive package of non-carceral controls (exclusion zones, EM/geofencing, communication bans, income redirection to survivor services, long-term treatment) that can substitute for imprisonment in higher S bands while preserving safety and censure.

• Integration with restorative/reparative approaches: Engage Prof. Qureshi’s obligations model. In serious sexual violence, what forms of dialogic accountability are ethically permissible without re-traumatization, and when should surrogate/community panels replace direct encounters? Propose equivalency rules (hours, payments, supervision) that reflect gendered harms without commodifying trauma.

• Structural critique: Respond to Dr. Baptiste’s warning that public-space risk and weapon-use aggravators are policed and measured differently across communities. What safeguards (data sources beyond police; DIF testing; aggregate-harm modules for coercive control) would you require so S doesn’t reproduce racialized patterns? Where should the guideline explicitly upgrade harms like wage-theft-for-sex, immigration-based threats, or digital sexual coercion that standard codes underweight?

• Evidence and metrics: Point us to the best empirical bases on survivor outcomes for non-carceral interventions in sexual/IPV cases (protective orders with enforcement, advocacy + housing, specialized treatment), and specify what success metrics you would write into the guideline audit (e.g., autonomy restoration indices, fear reduction, time-to-violation, recidivism in coercive control).

Bottom line: what concrete revisions to S, PIS, and Judge Villanueva’s grid would make our scheme genuinely answer to gendered harms while avoiding the overreach of carceral feminism that you warn against?

---

Speaker: Prof. Tamsin O’Rourke

Thank you, Chair. I’m broadly sympathetic to Wei‑Lun’s architecture—separate latent dimensions, public weights, uncertainty bands—but as Bedau anticipated in his critique of Card and Sellin–Wolfgang, the devil is in what we count as “harm.” Sexual and domestic violence are quintessential cases where acute injury is a poor proxy for the lived harm. If S is to do real work here, three upgrades are non‑negotiable: (a) autonomy/capability loss must be directly measured and given protected weight; (b) “credible‑risk environment” must be operationalised beyond weapons to include patterned coercive control; and (c) intersectional leverage must be coded as the offender’s aggravating conduct, not as victim traits.

1. Measurement: make autonomy and coercive control visible and weighted

• New latent dimension: Patterned Coercive Control (P)

  • Why: Coercive control is not a side‑note to “harm”; it is a distinct mode of wrongdoing—eroding agency over time. Folded into H it gets swamped by medical injury and episodic frames.
  • Items (polytomous/graded; scored with validated instruments):
    • Duration/frequency of control (Composite Abuse Scale–Revised; Women’s Experience with Battering; Psychological Maltreatment of Women Inventory).
    • Isolation, surveillance, financial control (CAS subscales; bank account control; device/credential takeovers).
    • Stalking and monitoring (SAM: Stalking Assessment and Management guide; technology‑facilitated coercive control indicators).
    • Threats to children/pets/immigration/employment (coded as offender’s use of structural leverage).
  • Guardrails: add αP ∈ [0.15, 0.30]. This protects P from being diluted by H’s acute injury subcomponents.

• Re‑specify H (realised harm) for gendered autonomy loss

  • Within H, create protected subcomponents:
    • Physical injury (Hphys).
    • Autonomy/capability loss (Hauto): post‑offence restriction of movement, work/education withdrawal, social withdrawal, sleep disruption, long‑term hypervigilance (measured by validated scales: WEB; IES‑R/PCL‑5 for fear/avoidance; capability‑loss questionnaires developed in victim research).
    • Confinement/false imprisonment.
  • Guardrail: in SV/IPV categories, require Hauto to be at least 40% of H’s internal weight (so H cannot collapse to Hphys).

• Risked harm (R): expand “credible‑risk environment”

  • Items: weapons access/brandishing; strangulation (“non‑fatal strangulation” flag); stalking behaviours; threats/escalation markers; presence of children; public/private trapping; digital threats (image‑based abuse).
  • Sources/validation: Danger Assessment (Campbell) and DA‑LE items for lethality risk; ODARA/B‑SAFER/SARA V3 for IPV recidivism risk; integrate “non‑fatal strangulation” as a high‑discrimination item.

• Relational aggravation (T): add Structural Leverage Aggravators

  • Offender’s exploitation of immigration status, employer power/wage theft/“sex for shifts,” landlord threats, outing threats for LGBTQ+ survivors. These are offender conduct items, not victim characteristics.

• Intersectionality without blame: do not code victim demographics. Code offender’s exploitation of structural vulnerability. Apply Differential Item Functioning (DIF) tests to ensure items behave equivalently across groups; if not, adjust or drop.

• Weight adjustments for SV/IPV categories

  • Set αC ∈ [0.40, 0.55], αH ∈ [0.25, 0.40], αR ∈ [0.15, 0.25], αT ∈ [0.05, 0.15], αP ∈ [0.15, 0.30]. This ensures P and R have real bite alongside culpability.

2. Worked comparisons

Illustrative α: C 0.45; H 0.30 (with Hauto≥40%); R 0.15; T 0.05; P 0.20. Component scores 0–10; S rescaled 0–100.

• Stranger rape with weapon, confinement, threats

  • C: purposeful sexual violation; planning: 8.5 (±0.6)
  • H: autonomy loss severe; confinement; trauma: Hphys moderate to high; Hauto high → H ≈ 8.8 (±0.6)
  • R: weapon, threats, confinement: 7.5 (±0.6)
  • T: none: 0
  • P: short‑duration coercive control (event‑bound): 3.0 (±0.8)
  • S ≈ 0.458.5 + 0.308.8 + 0.157.5 + 0.050 + 0.20*3.0 ≈ 73.3 (95% ~ [66, 80])

• Acquaintance assault with grooming and repetition

  • C: purposeful violation; grooming; repetition: 8.5 (±0.6)
  • H: per‑incident injury lower; Hauto amplified by repetition/social entanglement: 8.2 (±0.6)
  • R: moderate immediate bodily risk; high ongoing credible risk in network: 6.0 (±0.7)
  • T: abuse of trust/vulnerability: 2.0 (±0.5)
  • P: patterned domination over months: 7.5 (±0.7)
  • S ≈ 0.458.5 + 0.308.2 + 0.156.0 + 0.052.0 + 0.20*7.5 ≈ 79.3 (95% ~ [72, 86])

Interpretation: with robust P, the grooming/repetition case can surpass the stranger case because the moral reality is sustained domination and erosion of agency. Bedau’s worry about “fit” evaporates only if our metric can see that.

• IPV: long‑term coercive control, low acute injury vs single severe assault
  • Coercive control case: C 7.5; H 7.8 (Hauto heavy); R 7.0 (DA flags: strangulation, threats to children); T 1.5; P 8.5 → S ≈ 78–82.
  • Single severe assault (hospitalisation) with no prior control: C 7.5; H 8.8 (Hphys heavy; Hauto moderate); R 5.5; T 0; P 2.0 → S ≈ 70–73. The coercive control case rightly outranks the single assault because the social meaning and forward risk are worse.

3. Carceral‑feminism risks: when custody, and what instead

• Screening for custody reserved use

  • Use validated tools (Danger Assessment, ODARA/B‑SAFER/SARA V3 for IPV; Static‑99R/STABLE‑2007 cautiously for sexual recidivism, recognising limitations).
  • Custody is presumptively reserved for: (a) high imminent risk (DA high; strangulation; weapon access; escalating stalking); (b) serial predation with non‑responsiveness to controls; (c) violations of safety orders that create immediate danger.
  • Not for “expressive” severity alone in acquaintance SA or IPV where risk is moderate and controls can make survivors safe.

• Presumptive non‑carceral safety package (Band 5–6 equivalency)

  • Mandatory firearm relinquishment with third‑party checks and audits; DV‑specific geofencing/GPS with exclusion zones; zero‑contact orders across all channels; tech access restrictions and forensic phone audits; workplace/school safety plans.
  • Income redirection: 10% of wages for 5–10 years to survivor‑services funds; not victim‑paid unless requested.
  • Long‑term, audited treatment (sexual behaviour/IPV) with external clinical oversight; substance‑use controls if relevant.
  • Curfews, reporting, chaperone requirements where needed; travel and passport restrictions as risk dictates.
  • Swift‑certain consequences for breaches; short, reset custody only where necessary to restore safety, not as default “just deserts.”

4. Integrating reparative obligations (Qureshi) without re‑traumatising

• Ethically permissible forms

  • Direct encounters only with survivor’s sustained, informed consent and control; survivors may opt for indirect acknowledgment (video/written) vetted by facilitators; or surrogate/community panels.
  • No compelled apologies delivered to survivors. No “therapeutic ambush.”
  • Obligations index O calibrated by band: service in violence‑interruption/consent education; income‑based contributions to survivor funds; completion of change plans; tight liberty limits. Convert to custody equivalents transparently (e.g., Band 6: 1,500–2,000 O‑points over 4–6 years).

• Equivalency principles

  • Hours/payment/supervision reflect gendered harms without monetising trauma: the “currency” is community‑benefiting service, verified behaviour change, and long‑term constraints, not buying off pain.
  • Victim preference shapes form, not magnitude; refusal to engage does not raise the burden.

5. Structural critique and safeguards (with Nolan)

• Data sources beyond police

  • Feed S with hospital/clinic records; civil protection order findings; anonymised hotline/service‑provider data; DV fatality reviews; not just police narratives.
  • For R items (weapons, stalking), corroborate beyond police reports when possible; discount uncorroborated claims that track policing intensity rather than risk.

• DIF and impact audits

  • Mandatory DIF tests for Hauto, R, P items; suspend items that behave differently across race/SES without principled reason.
  • Quarterly disparate impact reports disaggregated by race/immigration; if persistent unexplained disparities, automatic weight review or structural adjustment (e.g., down‑weight place‑based “public‑space risk” absent independent evidence).

• Upgrade currently underweighted harms

  • Create specific items and high T aggravators for wage‑theft‑for‑sex, immigration‑based sexual coercion, landlord exploitation (“sex for rent”), and digital sexual coercion (non‑consensual image distribution, cyberstalking).
  • Add an “Aggregate Harm” module within S for population‑level harms (e.g., systemic sexual exploitation in workplaces), so that these can enter Bands 6–7 when capability losses are large.

6. Evidence and audit metrics

• Evidence base

  • Protective orders: where enforced, associated with significant reductions in re‑abuse and fear (e.g., Logan & Walker; Campbell’s DA‑informed interventions).
  • GPS in DV: evaluations (e.g., Spain; selected US jurisdictions) show reduced violations/re‑assault when GPS is monitored with swift response.
  • Advocacy + housing: strong evidence that survivor‑centred advocacy plus housing supports reduce revictimisation and improve autonomy indices.
  • Sexual offender treatment: modest but real reductions when programmes are risk‑need‑responsivity aligned; indiscriminate programmes can harm.
  • Victim‑offender dialogue in serious violence: improved victim well‑being when voluntary and well‑facilitated; evidence in sexual violence is thin—any use must be survivor‑led.

• Guideline audit metrics (beyond recidivism)

  • Autonomy restoration index (validated composite: mobility, work/school continuation, social connectedness).
  • Fear reduction (IES‑R/PCL‑5 changes; self‑reported safety).
  • Time‑to‑violation and seriousness of violations under EM/exclusion orders.
  • Verified weapons relinquishment and non‑contact compliance.
  • Survivor housing and income stability over 12–24 months.
  • Disparate‑impact metrics for S bands and modalities; corrective action triggers.

Bottom line: concrete revisions to S, PIS, and the grid

• S

  • Add latent P (Coercive Control) with αP guardrail 0.15–0.30.
  • Within H, enforce Hauto ≥ 40% weight in SV/IPV; publish subcomponent scores.
  • Expand R items to include strangulation, stalking, tech‑facilitated abuse; import Danger Assessment/ODARA/SARA items with open parameters.
  • Add Structural Leverage Aggravators to T; define functionally and tie to offender conduct.
  • Mandate DIF testing, non‑police data, and structural adjustment protocols.

• PIS

  • Add “survivor‑defined safety benefit” as an explicit term; privilege certainty/celerity of enforcement over severity in SV/IPV.
  • Penalise legitimacy harms from over‑policing/surveillance by requiring community oversight for EM/geofencing and strict data minimisation.
  • Treat custody as preventive only where validated imminent risk remains high despite controls.

• Grid (Judge Villanueva)

  • Create SV/IPV‑specific modality menus for Bands 5–6 that allow full non‑carceral substitution when O‑points and risk controls are satisfied; codify Band‑level O targets.
  • Automatic firearm relinquishment and verification as a pre‑condition for any community‑based sentence.
  • Upward departures across bands cannot be justified by public‑space/weapon factors alone; must show independent evidence of elevated risk not explained by policing saturation.
  • Require a “structural leverage” finding (jury for ceiling effects; judge C&C within band) where immigration/work/housing leverage was exploited; treat as T.

If we want a proportionality scheme that answers to gendered harms without sliding into carceral overreach, we must centre autonomy loss and coercive control as first‑order harms; constrain severity with dignity and distributive guardrails; and switch our prevention levers from piling on years to enforcing what actually keeps survivors safe: weapons removal, no‑contact, geofencing, income supports, and sustained, enforceable obligations of repair.

---

Speaker: Host

Dr. Sokolov, this is a natural moment to bring your Fingarette‑style internalism to the table. Bedau’s two objections bite hard: (i) rules—even laws—don’t conceptually require punishment (his chess analogy), and (ii) your view risks bypassing moral justification by treating retributive hard treatment as built into law’s authority. Please respond directly:

1. Law vs. games. What is the precise normative difference that makes authoritative legal directives conceptually call for institutional censure plus hard treatment when breached, whereas chess rules do not? Please give a principled account (authority, standing commitments, assurance conditions), not just an intuition.

2. Asymmetry without “moral alchemy.” On your view, why does breach by the guilty generate a reason for intentional hard treatment that does not apply to innocents—without appealing to desert as moral accounting or to preventive goods? Is it (a) constitutive of practical reason under authority, (b) a fairness commitment agents undertake as part of law‑membership, or (c) something else? Spell out the normative logic and its limits.

3. Avoiding circularity. How does your internalism avoid the dilemma Bedau sketches—either reduce to consequentialism (appeal to goods like catharsis/denunciation) or collapse into circularity (“we punish because law says so”)? What anchors the requirement in reasons that agents share, and what would defeat it (e.g., gross illegitimacy of the order)?

4. Non‑ideal conditions. Engage Dr. Baptiste’s critique: under background injustice, does law retain the authority to conceptually warrant hard treatment? Do you accept a distributive audit as a defeater for the presumption of penal authority? If so, what threshold failures of baseline provision or structural domination suspend the internalist warrant and force parsimony or substitution (reparative/administrative controls)?

5. Interfaces with the emerging architecture. Be concrete about institutionalization:

   • A1/A2: Do you treat hard treatment as analytically part of penal sanctions (A1) and guilt as gatekeeper (A2)? If yes, what counts as minimally sufficient “hard treatment”—can Qureshi‑style reparative packages satisfy A1 when they include non‑optional burdens and liberty limits?
   • S‑bands and grid: Can Chen’s S serve as the non‑arbitrary way to instantiate your internalist warrant (i.e., it fixes the form/degree of censure and burdens)? Which items are conceptually essential to preserve the “breach‑answer” structure (e.g., culpability must have a minimum weight floor)?
   • Departure rules: What internalist limits would you codify (no degrading penalties; no strict‑liability enhancements; reason‑giving as a condition of censure)?

6. Contrast cases. Take our recurring pairings and show how your view guides outcomes:

   • Provoked killing vs felony murder: If hard treatment is conceptually required for breach, what distinguishes the warranted burdens here? Do you reject strict felony‑murder liability and grade by risk emplaced and reasons‑responsiveness? What custodial parsimony follows from internalism rather than prevention or desert metrics?
   • Sexual violence: Does patterned coercive control (O’Rourke’s P‑dimension) alter the conceptual warrant’s intensity even with low acute injury? How do you integrate non‑carceral controls while preserving the censure‑plus‑burden core?

7. Capital punishment. From an internalist standpoint, is death ever conceptually eligible as the law’s hard treatment, or do dignity and reversibility constraints make it categorically or practically off‑limits? If your answer is “in principle but not in our systems,” specify the authority/epistemic thresholds that would have to be met.

Please be precise about where your view converges with and diverges from: (a) Strauss’s moral accounting (desert), (b) Holm’s preventive legitimacy, and (c) Qureshi’s reparative substitution. Conclude with two concrete guideline provisions you would draft to embed internalism without sliding into Bedau’s ‘moral alchemy’—for example, a culpability weight floor and a dignity constraint that invalidates degrading sanctions across all bands.

---

Speaker: Dr. Petra K. Sokolov

Thank you, Chair. Let me take the objections straight on and then show how an internalist account can be institutionalized without “moral alchemy.”

1. Law versus games: why breach conceptually calls for censure plus hard treatment in law, but not in chess

• Authority and exclusionary reasons. Legal directives purport to give content‑independent, exclusionary reasons for action: they settle what we shall do together despite private calculations. That claim is not made by game rules except qua voluntary participation in a practice you can exit without wronging anyone. You cannot simply “walk off the field” of criminal law without imposing costs on others’ protected interests.

• Standing commitments and assurance. A legal order is not just a set of primary rules; it includes standing commitments—by officials and citizens—to uphold them and to assure one another that breaches will be taken up. The point is not merely coordination, but securing equal status claims (rights) against defection. Secondary rules assign officials an institutional duty to censure breach; that duty is part of what it is to have law’s authority at all.

• Directed duties, not pastime conventions. Breaching a criminal norm wrongs particular persons and the polity’s order of mutual accountability. Breaking a chess rule wrongs no one, because the “ought” of a game is internal to a voluntary activity whose only standing sanction is exclusion from the game.

So: with law there is a built‑in answerability structure—an assurance practice backed by institutional duties to rebuke breach. That is why censure plus burdensome response is conceptually internal to legal authority, not to games.

2. The guilty/innocent asymmetry without “moral alchemy”

• Constitutive reasons of membership. In accepting a legal order as authoritative (as officials or as citizens), we undertake a fairness‑like commitment: if we defect, we will submit to institutional censure and to a non‑optional burden that marks that we, not our victims or fellow citizens, bear the cost of our defiance. That practical commitment is part of what it is to be a member of an authority that claims our obedience.

• Second‑personal standing. Breach makes the offender the addressee of directed claims from the victim and polity: “Answer for this, as the responsible author.” Those claims ground a reason to impose intentional hard treatment on the guilty that does not exist toward innocents, who have not authored a defiance. This is not payback for its own sake; it is the constitutive form of answering under authority.

• Limits. The burden must (a) be linked to the breach and its public meaning, (b) be the least burdensome sufficient to register censure and restore assurance, and (c) respect dignity. That excludes scapegoating and instrumentalizing innocents.

3. Avoiding circularity and consequentialist collapse

• Not “because the law says so.” The warrant flows from agents’ shared reason to participate in an authoritative scheme that secures equal standing through assurance. Censure‑plus‑burden is part of that scheme’s constitutive point: it is how breaches are taken up as breaches, independent of net preventive gain in a given case.

• Not reducible to consequences. Even if a particular sanction does not deter, the polity owes the victim and itself a taking‑up of the wrong as wrong. That is a reason of right, not of outcome. At the same time, it is bounded by dignity and parsimony; where burdens can be minimized consistent with that taking‑up, they must be.

• Defeaters. Gross illegitimacy defeats the claim: if the order systematically denies equal standing, or if basic conditions for fair compliance are missing, the internalist warrant for hard treatment is pro tanto suspended and may be displaced by non‑penal controls and repair.

4. Non‑ideal conditions and distributive audits

I accept Dr. Baptiste’s challenge in this form: authority requires a baseline of fair compliance conditions.

• Distributive audit as defeater. Where the state has failed to secure minimally adequate capabilities for lawful compliance (housing, subsistence, basic health care), or where a rule functions to entrench domination, the claim to impose more than minimal censure‑burdens is weakened or suspended. In such cases the response should default to reparative/administrative controls and targeted safety measures, with custody reserved for imminent serious risk.

• Parsimony by design. Even where the audit passes, internalism is parsimony‑forcing: because the point is to mark and mend the breach of equal standing, we have reason to choose the least burdensome measure that accomplishes that—often a structured reparative package rather than prolonged warehousing.

5. Interfaces: making internalism work in the proposed architecture

• A1/A2. Yes: A2 is the eligibility gate; A1 is analytic of penal sanctions—punishment is censure plus hard treatment. “Hard treatment” is minimally: a non‑optional, burdensome demand or liberty restriction imposed as censure. Robust reparative packages that impose significant, structured burdens and constraints qualify.

• Using S‑bands to instantiate the warrant. Prof. Chen’s S can serve as the non‑arbitrary map from breach gravity to permissible burden, provided the model embeds internalist essentials:

  • A culpability floor: mental state and reasons‑responsiveness must have a minimum weight; no strict liability.
  • A separate risked‑harm dimension so we do not collapse everything into outcomes.
  • A coercive‑control dimension (as Prof. O’Rourke urged) for patterned domination.
  • Dignity constraints: certain penalties excluded categorically.

• Departure rules under internalism. No degrading or sexualized penalties. No strict‑liability enhancements (e.g., felony murder) without proof and grading of risk embraced. Reason‑giving is intrinsic to censure: the sentence must say how the burden answers the breach and restores assurance. Lenity and uncertainty triggers protect against overreach.

6. Contrast cases under internalism

• Provoked killing vs felony murder.

  • Provoked killing: Intentional killing, but partial excuse (impaired self‑control without loss of agency) reduces breach gravity. The warranted burden is high but not maximal; custody may be shorter, with strong reparative obligations and a formal censure process.
  • Felony murder: Reject strict liability. Absent purpose to kill, censure tracks the degree of lethal risk knowingly embraced in predation. If extreme, the breach is grave; if risk was modest and the death freakish, the breach is materially lower than purposeful killing. Custodial parsimony follows: we do not add years to buy deterrence; we match the burden to the quality of the breach.

• Sexual violence and coercive control.

  • Patterned coercive control intensifies the breach even with low acute injury: it is sustained defiance of the victim’s equal standing. That warrants a heavier censure‑burden than a one‑off assault because the answer owed is deeper.
  • Non‑carceral controls can satisfy the censure‑plus‑burden core when they are significant and public (weapons relinquishment, geofencing, long‑term no‑contact, income redirection, verified treatment, and formal censure hearings). Minimal custody may still be justified to secure compliance or mark gravity in some cases, but the architecture should lean toward burdens that directly rectify the relational wrong.

7. Capital punishment

• Conceptual eligibility. Internalism does not specify modes; it requires censure‑plus‑burden consistent with respecting the addressee as a responsible agent. Death sits uneasily with that: it forecloses continued address and rectification, is irreversible, and in practice degrades. I therefore take it to be, at minimum, practically off‑limits, and quite plausibly categorically incompatible with dignity under authority.

• Thresholds no system meets. Even an “in‑principle” space would require near‑perfect legitimacy, vanishing error, non‑arbitrariness across persons and places, and procedures that preserve the person’s status through to the end. Our systems fall far short. Abolition follows.

Where I converge and diverge

• With Strauss. Convergence: breach generates answerability; censure‑plus‑burden is owed to agency equality. Divergence: I ground this in the constitutive reasons of authority and membership rather than in moral accounting by itself; I am more parsimony‑driven about form and magnitude.

• With Holm. Convergence: legitimacy side‑constraints are non‑negotiable; structured transparency and due process matter. Divergence: even when prevention gains are marginal, there remains an internal reason to censure and burden the guilty; prevention then chooses within the band, not whether there should be any band.

• With Qureshi. Convergence: reparative, dialogic obligations can be punishment when censure‑backed and burdensome. Divergence: some burdensome element is conceptually required for penal censure; pure repair that imposes no non‑optional burden is not punishment (though it may be valuable).

Two concrete guideline provisions to embed internalism without “moral alchemy”

• Culpability weight floor and anti‑strict‑liability rule. “In all seriousness computations (S), the culpability component must carry a weight of at least 0.40. No person may be sentenced above Band 4 for a death outcome absent proof, by clear and convincing evidence, of at least recklessness regarding lethal risk; felony‑murder liability shall be graded solely by the degree of risk knowingly embraced.”

• Dignity and assurance clause with parsimony. “No sanction may be imposed that degrades or instrumentalizes the person (including sexualized, mutilative, or terrorizing penalties). Every sentencing statement shall explain how the sanction (a) communicates authoritative censure of the breach, (b) restores assurance to victims and the polity, and (c) is the least burdensome measure sufficient to accomplish (a) and (b). Where uncertainty spans bands, the court shall apply lenity and prefer reversible, reparative burdens.”

That is how we keep punishment conceptually internal to law’s authority—anchored in reasons agents can share—while avoiding both the chess fallacy and the alchemy worry.

---

Speaker: Host

Prof. Strauss, Dr. Baptiste has pushed hard on a distributive-justice audit as a precondition for imposing hard treatment under non‑ideal conditions. From within your deontic “moral accounting” view, can you make that conditionality precise without collapsing into consequentialism? Please be concrete:

1. A fair‑opportunity‑to‑comply test. What baseline capabilities or state duties must be in place (e.g., access to housing, subsistence, basic health care, legal work authorization) before the polity may legitimately impose custodial hard treatment as the mode of answering for breach? Specify:

   • The threshold you would codify (a checklist or a composite index?),
   • Who bears the burden of proof and at what standard,
   • Whether failing the audit eliminates custody altogether or creates a strong presumption for reparative/non‑carceral burdens.

2. Operationalizing within the guideline system. Would you endorse a structural‑deprivation adjustment (automatic one‑band downward shift absent proof of fair opportunity), as Baptiste proposed? How would that interact with A2 and with your moral equivalence metric? Please draft language you would be willing to see in the guidelines that preserves a culpability floor while recognizing distributive defeaters.

3. Case applications.

   • Survival theft vs. wage theft/tax fraud: Using your metric, and assuming an audit finds the survival‑theft defendant lacked baseline capabilities while the wage‑theft contractor and tax cheat did not, how do the bands and burdens differ? Are there deontic reasons to elevate aggregate economic harms (contractor/tax cheat) into higher bands to avoid the asymmetry Baptiste highlights?
   • Drug markets: Street‑corner seller with a gun in a saturated, poor neighborhood vs. higher‑volume private‑network seller without a gun. How do you keep the moral‑equivalence assessment from laundering policing differentials into higher bands for the former? Should “public‑space risk” be down‑weighted unless independently corroborated, or does your deontic framework require an equalization move here?

4. Felony murder and strict liability. You and Dr. Sokolov both resist strict liability. State explicitly how your moral‑equivalence metric would grade felony‑murder homicides purely by risk knowingly embraced and reasons‑responsiveness, and whether you would write a categorical rule barring elevation above a certain band absent proof of at least extreme recklessness regarding death.

5. Capital punishment under non‑ideal conditions. Given your earlier principled allowance in narrow circumstances, does the distributive‑audit lens push you to practical abolition? If not full abolition, specify the additional distributive/epistemic thresholds that must be met beyond what you already listed (e.g., proof that capital charging and sentencing are free of race/county effects to a defined statistical tolerance).

6. Integration with the emerging architecture. Identify two concrete amendments you would add to Judge Villanueva’s grid and Prof. Chen’s S so your deontic side‑constraints are explicit: for example, (a) a culpability weight floor/anti‑strict‑liability clause, and (b) a distributive‑defeater provision with reason‑giving requirements.

Please aim to give draftable rule text where possible, and show how your deontic account can honor moral accounting while acknowledging Baptiste’s non‑ideal cautions about background injustice and policing asymmetries.

---

Speaker: Prof. Miriam K. Strauss

Thank you. I can accommodate Dr. Baptiste’s non-ideal cautions without abandoning the deontic core. The key is reciprocity: the polity’s standing to impose custodial hard treatment as the mode of moral accounting presupposes that it has discharged baseline duties that make lawful compliance a fair demand. When those duties are not met, accountability remains, but the permissible mode shifts toward reparative, non-carceral burdens, save where immediate safety requires incapacitation. That is not consequentialism; it is a condition on authority grounded in equal agency.

1. A fair‑opportunity‑to‑comply test (deontic “reciprocity” audit)

• Baseline duties (core capabilities the state must have made reasonably accessible during the relevant period, e.g., the 90 days prior to the offense):

  1. Lawful subsistence: access either to minimally adequate housing or shelter, and a subsistence income floor via employment or income support.
  2. Access to basic care: primary health, mental health, and, where clinically indicated, timely addiction treatment; language access for enrollment.
  3. Legal enablement: government ID; no civil-disability barriers that make ordinary compliance impossible (e.g., license suspensions solely for debt where driving is necessary for lawful work); lawful work authorization where the state controls the barrier.
  4. Normative notice: reasonable notice of the criminal norm (including translation where relevant).

• Form: a checklist aggregated to a composite “Baseline Compliance Capability (BCC) Index.” The threshold is conjunctive: the state must show all core items were in place or reasonably offered.

• Burden and standard: the state bears the burden to prove fair opportunity by clear and convincing evidence. The defendant may raise specific deficits; the state must then rebut.

• Effect of failure: failure of the audit does not erase liability (A2 holds). It creates a strong presumption against custodial hard treatment. The court must impose censure-backed reparative and supervisory burdens instead, unless the state further proves imminent, serious risk that cannot be managed with non-carceral controls.

Draftable rule text “Fair‑Opportunity‑to‑Comply. Prior to imposing any custodial sentence, the court shall find, by clear and convincing evidence, that during the 90 days preceding the offense the defendant had reasonable access to (a) minimally adequate housing or shelter and subsistence income; (b) basic health care, including clinically indicated substance-use treatment; (c) legal enablement for lawful work and movement; and (d) reasonable notice of the criminal prohibition. Absent such findings, the sentence shall consist of non-carceral, censure‑backed reparative and supervisory obligations, unless the court also finds, by clear and convincing evidence, that custodial incapacitation is necessary to avert imminent serious harm.”

2. Operationalizing in the guideline system

• Yes, I would endorse a structural‑deprivation adjustment with a culpability floor.

Draftable guideline language “Structural‑Deprivation Adjustment. Where the Fair‑Opportunity‑to‑Comply finding is not made, the court shall apply an automatic one‑band downward adjustment from the S‑band indicated by the offense seriousness score. In offenses in Bands 6–7 involving purposeful or knowing violence against persons, the adjustment shall not reduce below Band 4. Any departure from the one‑band adjustment requires written reasons identifying imminent risk unmanageable by non-carceral means.”

• Interaction with A2 and the moral‑equivalence metric: A2 remains the gate; the moral‑equivalence assessment (culpability + harm, with independent culpability weight) fixes the gravity and the desert ceiling. The distributive audit does not shrink culpability; it restricts the custodial modality used to discharge the deontic debt where reciprocity is missing.

3. Case applications

• Survival theft vs wage theft/tax fraud
  • Survival theft: If the audit fails (no baseline subsistence/housing), impose censure‑backed, non‑carceral obligations: restitution where apt, income‑based payments to a victims’ fund, verified community service, and structured support engagement; no custody absent imminent risk. Moral accounting is satisfied by burdens that acknowledge and repair the wrong without compounding the state’s prior failure.
  • Wage theft/tax fraud: Audit will typically pass. Our metric should explicitly elevate aggregate economic harms as breaches of reciprocity. I would add an Aggregate Economic Harm module to S so that systematic appropriation of labor or public funds can reach higher bands when harm is large and culpability is predatory or knowing. Deontically, these are classic cases of shifting one’s costs onto others while free‑riding on the legal order—squarely within what moral accounting condemns.

Draft addition to S “Aggregate Harm Module (AEH). For wage theft, tax fraud, and analogous aggregate harms, S shall include an AEH score that reflects the scope, duration, and distributive impact of the appropriation. Where AEH exceeds published thresholds, offenses are eligible for Bands 5–6.”

• Drug markets: street seller with gun vs private‑network seller
  • Moral‑equivalence requires we grade by the risk knowingly imposed and reasons‑responsiveness, not by policing saturation. “Public-space” as such is a poor proxy for risk; I would require concrete, corroborated indicators (brandishing, discharge, presence of bystanders, crowd density) and independent corroboration beyond mere location. For the private seller, we must not ignore harms simply because they are diffuse; risked harm should reflect potency and foreseeable overdose risks when the actor knows of adulteration, though this must be tightly constrained to avoid re‑inflating the failed “war on drugs.”

Draft equalization provision “Public‑Risk Corroboration. Elevation for ‘public-space risk’ shall require independent corroboration of concrete endangerment (e.g., weapon brandishing or discharge; presence of third parties in the line of fire). Mere location in a high‑policing area shall not, without more, justify a higher risk score. The Commission shall audit place‑based aggravators for disparate impact and adjust weights accordingly.”

4. Felony murder and strict liability

I would write felony‑murder grading to track only the risk knowingly embraced and the agent’s reasons‑responsiveness. No elevation to the intentional‑murder band absent proof of at least extreme recklessness as to death.

Draft categorical rule “Felony‑Murder Grading. No sentence above Band 5 shall be imposed for a homicide predicated solely on felony‑murder unless the State proves beyond a reasonable doubt that the defendant acted with at least extreme recklessness regarding the risk of death—defined as conscious disregard of a very high probability of death under the circumstances. Strict‑liability elevation based solely on commission of the predicate felony is prohibited.”

5. Capital punishment under non‑ideal conditions

The reciprocity audit pushes my already narrow in‑principle allowance toward practical abolition. If a jurisdiction refuses abolition, additional thresholds beyond my earlier epistemic and dignity conditions are necessary and, I think, unattainable in practice:

• Demonstrated non‑disparity: pre‑adoption and annual audits showing no statistically significant race‑of‑victim, race‑of‑defendant, or county effects in capital charging or sentencing after covariate adjustment (e.g., odds ratios within 1.05; p < .01).
• Centralized charging with blind review; statewide uniform aggravator practice.
• Heightened proof: beyond reasonable doubt on all aggravators; a residual‑doubt option.
• Error rate: credible independent estimate of wrongful conviction risk below 1 in 10,000 with narrow confidence intervals across subgroups.

On present evidence, these conditions cannot be met; a moratorium/abolition follows.

6. Two explicit deontic amendments to embed in S and the grid

(a) Culpability floor / anti‑strict‑liability clause “In all seriousness computations (S), culpability shall carry a weight of no less than 0.40. No person shall be sentenced above Band 4 on account of a death outcome absent proof of at least recklessness as to death. Strict‑liability enhancements are prohibited.”

(b) Fair‑opportunity defeater with reason‑giving “Before imposing custodial punishment, the court shall make Fair‑Opportunity‑to‑Comply findings as set out above. If the findings are not made, the court shall (i) apply the Structural‑Deprivation Adjustment; (ii) impose censure‑backed reparative obligations and supervisory controls commensurate with S; and (iii) state on the record why any residual custody is necessary to avert imminent serious harm and why lesser burdens are insufficient.”

A final word on laundering policing asymmetries. The moral‑equivalence metric must measure the wrong the agent chose—mental state, the concrete risk he set in motion, the relational wrong—not background surveillance. That requires corroboration rules, DIF audits, and, where necessary, structural adjustments of weights. It also requires, as Tamsin urged, that we give protected, non‑derivative weight to autonomy loss and coercive control in gendered violence; and as Nolan urged, that we bring aggregate harms into the same moral frame. None of that abandons moral accounting. It clarifies the conditions under which the polity may demand that account through custody, and when, because its reciprocal duties have not been met, it must answer the wrong through reparative and supervisory burdens instead.

---

Speaker: Host

Prof. Chen, several powerful proposals now sit on your plate. I’d like you to commit to concrete revisions to S and its governance, and to show numerically how they change case rankings, while addressing feasibility and de‑biasing.

1. Coercive control as a latent dimension. Prof. O’Rourke proposes adding a patterned coercive control dimension P with αP guardrails (0.15–0.30) and validated instruments (CAS‑R, WEB, DA/SARA items, tech‑facilitated abuse). Please specify:

   • The exact items you would include for P (duration/frequency, isolation/surveillance, financial control, stalking/monitoring, threats to children/pets/immigration, tech coercion), their scoring scales, and priors on item discrimination/difficulty.
   • Revised α‑weight priors across C, H, R, T, P for SV/IPV categories, including the Hauto ≥ 40% internal weight rule O’Rourke wants, and a rationale for those guardrails.
   • How you will ensure DIF testing and bias correction so P does not encode victim demographic differences; specify decision rules for suspending or re‑estimating items when DIF is detected.

2. Aggregate harm module. Dr. Baptiste presses for an AEH module so wage theft, tax fraud, and environmental crimes can enter higher bands when population‑level capability losses are large. Please propose:

   • A concrete AEH construct with indicators (dollar magnitude, duration, number of victims, severity of capability loss proxies such as missed rent/food/housing instability, estimated population exposure in environmental cases), measurement model (e.g., a separate latent dimension with αAEH guardrails 0.10–0.25 for relevant offense classes), and anchoring offenses.
   • A worked numerical example showing how a contractor stealing $20,000 in wages across 10 workers compares on S+AEH to a non‑injury armed robbery, and where bands should land under your revised grid.

3. Structural‑deprivation adjustment. Both Strauss and Baptiste would add a Fair‑Opportunity‑to‑Comply audit and a one‑band downward shift absent proof of baseline capabilities. From a metrics standpoint:

   • Do you incorporate a Baseline Compliance Capability (BCC) index into S (risk: importing SES proxies) or keep it as a post‑S adjustment? Recommend the architecture and justify it.
   • Define BCC items (access to housing/shelter; subsistence income; basic health/mental health/addiction care; legal ID and work authorization; notice), their verification sources, and a scoring/threshold rule. Address how to audit for disparate invocation without gaming.

4. Public‑risk corroboration and non‑police data. Draft the “Public‑Risk Corroboration” measurement rule to avoid policing‑saturation bias: what qualifies as corroboration (video/forensic evidence; third‑party affidavits; hospital/clinic records), and how to discount location alone. Explain how non‑police feeds (medical, service‑provider, civil PO findings) will be integrated technically (data standards, privacy) into S scoring.

5. Governance and feasibility. Courts can’t fit bespoke Bayesian models in real time. Give us:

   • The precomputed scoring instrument (item list for common offense classes; look‑up tables for posterior means/intervals; a digital tool architecture), and a 5‑year re‑estimation cycle plan with open code.
   • The exact uncertainty/lenity triggers you’d codify and how appellate courts should review S computations (legal vs factual standards).

6. Updated worked examples (numbers welcome):

   • (a) Stranger rape with weapon/confinement vs (b) acquaintance grooming/repetition: show S with P added (using your proposed α). When, numerically, does (b) exceed (a)?
   • Wage theft/tax fraud vs non‑injury armed robbery: compute S+AEH and map to bands.

7. PIS interface and legitimacy costs. Dr. Holm warns about racially‑coded “community sentiment” in legitimacy terms. State explicitly how PIS will estimate legitimacy costs using blinded vignette compliance/cooperation studies rather than raw attitudes, and how S’s ceiling blocks desert creep inside PIS.

Bottom line: present a concrete revision memo—items, weights, examples, governance—that (i) adds P and AEH; (ii) places BCC as a post‑S adjustment; (iii) codifies corroboration and non‑police data; and (iv) is implementable on Judge Villanueva’s grid. Where does this leave Bedau’s arbitrariness objection—have we reduced it as far as is institutionally possible?

---

Speaker: Prof. Wei-Lun Chen

Thank you, Chair. Here is the revision memo you asked for—items, weights, examples, and governance—so the scheme is implementable on Judge Villanueva’s grid and responsive to Prof. O’Rourke’s and Dr. Baptiste’s critiques.

1. Add a patterned coercive control latent dimension P (SV/IPV)

Items and scoring (graded-response items; 0–3 scale unless noted)

• Duration/frequency of control (CAS‑R; WEB):
  • 0 none; 1 ❤️ months episodic; 2 3–12 months intermittent; 3 >12 months sustained/often.
• Isolation/surveillance:
  • 0 none; 1 social isolation without monitoring; 2 monitoring of movements/communications; 3 GPS/spyware/keystroke logging or coerced shared credentials.
• Financial control:
  • 0 none; 1 occasional coercion re: spending; 2 control of bank/credit; 3 employment sabotage or forced debt/identity theft.
• Stalking/monitoring:
  • 0 none; 1 unwanted following/contact; 2 repeated stalking with threats; 3 stalking with breach of protective orders or tactics that induced relocation.
• Threats exploiting structural leverage (children/pets/immigration/employment):
  • 0 none; 1 implicit threats; 2 explicit threats; 3 threats plus steps toward execution (e.g., contacting employer/ICE).
• Tech‑facilitated coercion (image‑based abuse/cyberstalking):
  • 0 none; 1 threats to share intimate images; 2 dissemination to limited audience; 3 broad dissemination or revenge‑porn site use.

Bayesian IRT priors (per item)

• Discrimination a ~ LogNormal(μ=0, σ=0.5) (median ≈ 1; heavy tails allow high‑info items like stalking‑with‑PO breach).
• Thresholds b1,b2,b3 ~ N(0,1) with ordering constraint (b1<b2<b3); weakly informative to let the data place cutpoints.

Revised α‑weight guardrails (SV/IPV offense classes)

• αC ∈ [0.40, 0.55]; culpability floor.
• αH ∈ [0.25, 0.40], with Hauto ≥ 40% of the internal H weight (so autonomy/capability loss cannot be swamped by Hphys).
• αR ∈ [0.15, 0.25]; credible‑risk environment includes strangulation/stalking.
• αT ∈ [0.05, 0.15]; relational/structural leverage aggravators.
• αP ∈ [0.15, 0.30]; coercive control has protected bite. Rationale: this enshrines independent roles for reasons‑responsiveness (C), lived harms (Hauto), ex ante endangerment (R), abuse of position (T), and patterned domination (P). It directly fixes the Card/Sellin–Wolfgang problem: neither outcome harm nor “full measure” vagueness can dominate.

DIF testing and bias correction

• Multiple‑group MIRT with alignment: test invariance across victim race/ethnicity, immigration status, and neighborhood deprivation.
• Decision rule: if an item shows DIF > 0.3 logits in thresholds or > 0.25 in discrimination across ≥2 protected subgroups, and the effect persists after adjusting for C/H/R profiles, suspend the item and substitute functionally equivalent indicators; re‑estimate the model. All DIF tests and item‑level decisions are published.
• Code offender exploitation, not victim traits: e.g., “threatened to call ICE” counts; “victim is undocumented” does not.

2. Add an Aggregate Economic/Environmental Harm module (AEH)

Construct and indicators (latent AEH for white‑collar/environmental classes)

• Dollar magnitude (log‑scaled): ln(total illicit gain or avoided cost).
• Duration: months of conduct.
• Number of direct victims: count or log(count).
• Capability‑loss proxies:
  • Wage theft: documented missed rent/utility shutoffs/food insecurity events attributable to non‑payment; days of work lost seeking redress.
  • Tax fraud: public‑service shortfall proxies (e.g., school district deficits) or standardized fiscal harm weights by jurisdiction.
  • Environmental: estimated population exposure; toxicity‑weighted exposure (e.g., µg/dL BLL>5 in N persons; EPA toxicity equivalence factors); duration of exceedances.
• Intent: binary/ordinal indicator of willful vs negligent (feeds C, not AEH).

Measurement model

• Separate latent AEH (0–10), calibrated on vignettes and empirical case data; αAEH ∈ [0.10, 0.25] for relevant offense classes; AEH = 0 for street offenses. Anchors:
  • Wage theft $5k/3 workers/1 month = AEH ≈ 3.
  • Tax fraud $50k/1 year = AEH ≈ 4.
  • Environmental release causing 100 residents BLL>5 for 6 months = AEH ≈ 8.

Worked comparison

• Contractor steals $20,000 wages across 10 workers over 6 months; 30 documented missed‑rent/utility events.
  • C ≈ 7.5 (knowing appropriation; predatory reasons).
  • H: Hauto (economic capability loss) ≈ 6.0; Hphys 0; H ≈ 6.0.
  • R ≈ 1.0 (no acute bodily risk).
  • T ≈ 2.0 (abuse of employer power).
  • AEH ≈ 7.0–7.5 (ln(20k)≈10; 10 victims; 6 months; 30 missed rent).
  • Using αC=0.45, αH=0.25, αR=0.10, αT=0.05, αAEH=0.15:
    • S_base = 0.457.5 + 0.256.0 + 0.101.0 + 0.052.0 ≈ 5.38 (on 0–10).
    • Add AEH: 0.15*7.25 ≈ 1.09.
    • S_total ≈ 6.47 → 64.7 (Band 4–5 boundary; with uncertainty ~ [58, 71]).
• Non‑injury armed robbery (brandishing a loaded firearm in public; no injury):
  • C ≈ 7.5; H ≈ 6.5 (Hauto high; Hphys 0); R ≈ 7.0; T=0; AEH=0.
  • With αC=0.45, αH=0.35, αR=0.20:
    • S ≈ 0.457.5 + 0.356.5 + 0.20*7.0 ≈ 7.05 → 70.5 (Band 5). Interpretation: with AEH, serious wage theft no longer sits in Band 2–3 by default; it can enter Band 4–5, reflecting aggregate capability losses. Armed robbery remains Band 5 because acute autonomy invasion and lethal risk matter.

3. Structural‑deprivation (Fair‑Opportunity‑to‑Comply) adjustment (post‑S)

Architecture

• Keep BCC post‑S. Do not bake SES into S. Use BCC only to adjust modality/band after seriousness is measured.

BCC items (binary, conjunctive threshold; 90‑day window)

• Housing/shelter access: documented access (shelter placement or voucher) offered and not unreasonably refused.
• Subsistence income: eligibility assessed and reasonable access provided (employment services or income support).
• Basic health/mental health/addiction care: appointment offered within clinically appropriate timelines; language access provided.
• Legal enablement: possession of state ID or provision facilitated; no state‑imposed civil barriers (e.g., debt‑based license suspension) blocking lawful work; if immigration work authorization is state‑controlled, reasonable steps taken.
• Normative notice: statute and warning notices reasonably accessible (language/reading level commensurate).

Verification sources

• Agency records (housing, social services, clinics); court records; sworn affidavits from service providers; defense can rebut with affidavits and contemporaneous proof.

Threshold and adjustment

• State bears clear‑and‑convincing burden; conjunctive standard (all cores must be met).
• If BCC fails: automatic one‑band downward shift (floor at Band 4 for purposeful/knowing violence against persons), custody presumptively replaced by censure‑backed reparative/supervisory obligations unless imminent serious risk is proven (clear and convincing). Audit
• Publish invocation rates by race/SES; random file audits; sanction agencies that game records.

4. Public‑risk corroboration and non‑police data integration

Measurement rule (codified)

• “Public‑risk elevation (R) requires corroboration of concrete endangerment beyond location:
  • Qualifying corroboration: video or body‑cam; forensic evidence (e.g., shell casings); hospital/clinic records (injuries consistent with brandishing/discharge); third‑party affidavits; civil protection orders; or contemporaneous 911 calls.
  • Location alone (e.g., ‘school zone,’ ‘high‑crime area’) does not justify R > 4 absent corroboration.
  • Where only location is present, apply a 50% discount to any place‑based R enhancer.” Non‑police data feeds
• Technical: HL7 FHIR for medical encounters; standardized XML/JSON from courts for civil PO findings; secure APIs from certified victim‑service providers (aggregate/flag only; strict privacy governance).
• Privacy: statutory data‑sharing gateways; de‑identification where possible; role‑based access in the digital tool; audit logs; independent privacy officer oversight.

5. Governance and feasibility

Precomputed instrument

• Offense‑class worksheets: a fixed item set per class (e.g., homicide, robbery, SV/IPV, white‑collar/environmental), with clear coding guides.
• Look‑up tables: posterior means and 95% credible intervals for typical item patterns; a simple additive scoring sheet maps to precomputed posterior summaries (no courtroom MCMC).
• Digital tool: open‑source web application with on‑device caching; inputs entered; outputs include Ŝ, [S5,S95], item diagnostics, and warnings (e.g., DIF‑flagged items). Re‑estimation plan
• Annual Bayesian updates; 5‑year full re‑estimation; open code (GitHub‑like repository); independent replication by an academic partner; public notice‑and‑comment.

Uncertainty and lenity triggers (codify)

• If [S5,S95] crosses a band threshold: default to the lower band; upward movement requires clear and convincing proof of the specific facts resolving uncertainty and written reasons; automatic review.
• If [S5,S95] spans >2 bands: treat as underdeterminate; prefer reversible sanctions; permit remand for additional fact‑finding. Appellate review
• Legal adherence to the tool and rules: de novo.
• Item codings (fact findings): clear error.
• Compliance with lenity/trigger rules: de novo.

6. Updated worked examples with P and AEH

SV pair (αC=0.45, αH=0.30 with Hauto≥40%, αR=0.15, αT=0.05, αP=0.20)

• (a) Stranger rape (weapon, confinement, threats):

  • C 8.5; H 8.8 (Hauto weight ≥40%); R 7.5; T 0; P 3.0.
  • S_raw = 0.458.5 + 0.308.8 + 0.157.5 + 0.050 + 0.20*3.0 ≈ 3.825 + 2.640 + 1.125 + 0 + 0.600 = 8.190 → S ≈ 81.9 (95% ~ [75, 88]) → Band 6.

• (b) Acquaintance grooming/repetition:

  • C 8.5; H 8.2 (Hauto heavy via repetition); R 6.0; T 2.0; P 7.5.
  • S_raw = 0.458.5 + 0.308.2 + 0.156.0 + 0.052.0 + 0.20*7.5 ≈ 3.825 + 2.460 + 0.900 + 0.100 + 1.500 = 8.785 → S ≈ 87.9 (95% ~ [81, 93]) → Band 6/7 boundary. Threshold insight: (b) exceeds (a) whenever P ≥ ~6 given these weights; the tool will show a “P‑driven elevation” flag to force reasons and prevent rote privileging of “stranger + weapon” cases.

White‑collar vs street robbery (with AEH)

• Wage theft case above: S_total ≈ 64.7 (Band 4–5 boundary).
• Non‑injury armed robbery: S ≈ 70.5 (Band 5). Result: both are treated as serious; the contractor can now sit in high Band 4/low 5 rather than being relegated to lower bands.

7. PIS interface and legitimacy costs

Estimating legitimacy costs (without racialized sentiment)

• Use blinded vignette experiments (mask race/class cues) to estimate how different sanction mixes affect willingness to report/cooperate and perceived fairness among diverse samples. Validate against behavioral proxies (reporting rates, clearance rates) where policies change (policy trials).
• Feed those elasticities into PIS as “cooperation impacts,” not “public anger.” Publish models; forbid social‑media proxies or media salience.
• S as ceiling/floor: PIS cannot recommend a sanction above the S‑band; within band, it optimizes expected prevention and cooperation given the published elasticities. No “desert creep” via “legitimacy”—if the only reason to go higher is expressive satisfaction, the ceiling stops it.

Where this leaves Bedau’s arbitrariness objection

We will never eliminate judgment. But we can bound it.

• Content transparency and guardrails: separate latent pillars; protected weights (culpability floor; Hauto minimum in SV/IPV; αP and αAEH floors for their domains); dignity constraints external to S.
• Quantified uncertainty and lenity: we always show [S5,S95]; we default down when uncertainty matters; we require reasons and automatic review to go up.
• De‑biasing: corroboration rules; non‑police data; DIF tests with public suspension/adjustment; structural‑deprivation adjustment as a post‑S rule with a conjunctive, state‑burden standard; aggregate‑harm module to correct offense‑definition asymmetries.
• Democratic revision: open code, five‑year re‑estimation with hearings; sunset unless reauthorized; disparate‑impact dashboards with binding corrective action.

Two closing notes of challenge and convergence

• To Prof. O’Rourke: P is now first‑class with protected weight and validated instruments; Hauto cannot be minimized. That is how we encode gendered harms without carceral overreach—by upgrading measurement, not reflexively adding years.
• To Dr. Baptiste: we keep S free of SES, add AEH to capture diffuse harms, codify corroboration to blunt location bias, and put BCC post‑S with a strict state burden. If the polity hasn’t kept faith on baselines, custody recedes by rule. That’s an institutional answer, not a promise.

Judge Villanueva can drop this into the grid tomorrow: the worksheet reflects these items; the digital tool returns Ŝ and bands with flags; BCC is a checkbox audit with a clear standard; and appellate courts know exactly what to review. Under non‑ideal conditions, that is as far as we can push arbitrariness down—and far enough, I believe, to meet Bedau’s challenge in practice.

---

Speaker: Host

Judge Villanueva, we now have a substantially revised architecture on the table: (i) Prof. Chen’s updated S with two major modules (P for patterned coercive control in SV/IPV with protected weights and validated instruments; AEH for aggregate economic/environmental harms), (ii) a Fair‑Opportunity‑to‑Comply (BCC) audit as a post‑S, one‑band downward adjustment absent clear‑and‑convincing proof of baseline capabilities, and (iii) a Public‑Risk Corroboration rule plus non‑police data feeds (medical, service‑provider, civil PO findings) to reduce policing‑saturation bias. Please take the implementer’s chair and get concrete about how a jurisdiction would roll this out.

1. Operational plan. Draft a 12–18 month implementation roadmap that covers: (a) rulemaking to adopt the revised S (adding P and AEH guardrails), the BCC audit, and corroboration rules; (b) data‑sharing MOUs and technical standards (HL7 FHIR for clinical; court XML for POs; provider APIs), with privacy governance; (c) building the precomputed scoring tool (paper worksheets + web app) and bench/Bar training; and (d) a pilot + phased statewide rollout (which case types first, e.g., SV/IPV, robbery, wage theft/environmental), with an independent evaluation plan.

2. Due process and triers. Specify which new facts must be submitted to a jury BRD versus found by a judge C&C: e.g., high P scores that would authorize Band 6/7 ceilings, AEH thresholds that unlock higher bands, and BCC audit failures that allow custody. Provide draft language to slot these into existing Apprendi/Blakely frameworks.

3. Resource and workload. Identify the agencies that bear new burdens (sentencing commission, courts, AG/county counsel for MOUs, public defenders, prosecutors, IT, hospitals/service providers). Provide an order‑of‑magnitude budget and the staffing/training plan (bench cards, CLEs, facilitator accreditation) to make this administrable without bogging courts down.

4. Guarding against gaming and inequity. Propose safeguards against (a) BCC bad‑faith documentation, (b) unequal access to non‑police evidence (e.g., hospitals refusing to share; service providers overloaded), and (c) differential surveillance that could inflate P/R for marginalized communities. Include random audits, sanctions, and fallback evidentiary pathways.

5. Appellate and oversight. Update your earlier review standards to accommodate P and AEH and the BCC adjustment. Would you create a specialized appellate docket or a sentencing review panel for the first two years? What public dashboards (disparities; item‑level usage; BCC invocation/outcomes) would you require, and what triggers corrective action?

6. Pilot sentencing template. Provide an updated, succinct sentencing statement template that adds: (i) P and AEH component scores and intervals where applicable; (ii) the BCC finding (met/not met) and sources; (iii) corroboration sources for public‑risk items; and (iv) the within‑band PIS rationale, including any reparative package selections and survivor‑safety controls. Keep it tight—what exactly must a line judge write to make reasons reviewable and non‑arbitrary?

7. Failure modes and course‑correction. Identify two likely failure modes in the first year (e.g., inconsistent P coding; bottlenecks in data feeds; over‑ or under‑invocation of BCC) and your pre‑planned fixes (hotline guidance; emergency rule tweaks; supplemental training; temporary presumptions).

8. Final stance on capital. With BCC and corroboration rules now in the mix, does your practical moratorium recommendation on capital charging change at all? If not, state whether you would issue a formal statewide directive to suspend capital filings pending demonstration of non‑disparity and vanishing error.

Please be concrete—statutory/regulatory text snippets and checklists are welcome. The goal is to show how the refined scheme can leave Bedau’s arbitrariness worries as far behind as our institutions can manage, without blowing up administrability.

---

Speaker: Hon. Elena P. Villanueva

Thank you, Chair. I’ll put my guideline‑designer hat on and lay out a practical rollout plan, with the legal plumbing and operational guardrails you’ll need to keep discretion structured, reasons reviewable, and error costs contained.

1. 12–18 month implementation roadmap

Phase 0 (Month 0–2): Governance and scoping

• Constitute an Implementation Steering Group (ISG): Sentencing Commission (lead), Courts Administrator, AG’s Office, State Health Information Exchange (HIE), hospital association, victim‑services coalition, public defender/DA reps, IT, privacy officer, and an academic evaluation partner.
• Adopt project charter with timelines, deliverables, and a change‑management plan.

Phase 1 (Month 2–6): Rulemaking and data infrastructure

• Rules/adoption package (notice‑and‑comment; Commission vote; legislative ratification where required)
  • Adopt revised S with guardrails:
    • Insert the P (patterned coercive control) dimension and AEH (aggregate economic/environmental harm) module as published by the Commission, with α‑weight floors and DIF testing requirements.
    • Codify the Hauto ≥ 40% rule in SV/IPV.
  • Enact the Fair‑Opportunity‑to‑Comply (BCC) audit as a post‑S, one‑band downward adjustment absent clear‑and‑convincing proof by the state; include the imminent‑risk exception.
  • Adopt Public‑Risk Corroboration rule (location alone insufficient; list of qualifying corroborants).
  • Dignity constraints and culpability weight floor (no strict‑liability elevation; minimum αC).
• Data‑sharing MOUs and standards
  • HIE MOUs with hospital systems/clinics using HL7 FHIR resources (encounters, diagnoses) for limited, event‑tied flags; privacy impact assessment; role‑based access.
  • Court XML schema for civil protection orders and findings; event subscription API.
  • Provider APIs for service‑provider attestations (presence at sessions; coercive‑control indicators), with standardized JSON payloads; de‑identified where possible.
  • AG/HIE to promulgate a Data Governance Policy: data minimization, logging, independent privacy officer oversight, breach response, and audit rights.

Phase 2 (Month 4–10): Tools and training

• Precomputed scoring tool
  • Paper worksheets by offense class (Homicide, Robbery/Burglary, SV/IPV, Wage/Tax/Environmental) with item definitions and coding examples.
  • Web app (open‑source) that implements the Commission’s precomputed posterior look‑ups; outputs Ŝ, 95% interval, band, and flags (e.g., “interval straddles threshold,” “P‑driven elevation,” “AEH anchor exceeded”).
• Bench/Bar training
  • Judicial bench cards and a 4‑hour CLE covering: S items and guardrails, BCC audit, corroboration standards, uncertainty/lenity triggers, and reason‑giving requirements.
  • Prosecutor/defender CLEs; DA/PD “coding clinics” with hypotheticals; facilitator accreditation for restorative packages.

Phase 3 (Month 10–15): Pilot and independent evaluation

• Pilot sites and case types
  • SV/IPV dockets (P active), robbery with weapon (corroboration rule), wage theft/environmental crimes (AEH).
  • Three urban and three rural circuits; 6–9 months operational pilot.
• Evaluation plan
  • Academic partner conducts an interrupted time‑series evaluation: sentence dispersion; disparity measures; time‑to‑sentencing; victim safety metrics in SV/IPV; charging patterns in wage/environment.
  • Concurrent usability study; issue rapid‑cycle improvement memos to the ISG.

Phase 4 (Month 15–18): Statewide rollout

• Adjust rules/tool from pilot lessons.
• Staggered rollout to remaining circuits; maintain hotline support and monthly “brown‑bag” webinars for judges and counsel.
• Launch public dashboards (see §5).

2. Due process and triers (Apprendi/Blakely‑compliant)

Facts requiring jury BRD (if they authorize a higher statutory maximum or S‑band ceiling):

• P‑elevating facts that unlock Band 6/7 ceilings in SV/IPV (e.g., “sustained coercive control exceeding 12 months with stalking in violation of orders” where statute ties degree/maximum to P). Draft: “Any fact that authorizes sentencing in Band 6 or higher for SV/IPV offenses shall be alleged and proved to a jury beyond a reasonable doubt or admitted by the defendant.”
• AEH thresholds that move white‑collar/environmental offenses into higher degree/band ceilings (e.g., “aggregate harm exceeds Level 3”).
• Traditional Apprendi aggravators (weapon discharge, abuse of public office) when they expand the ceiling.

Facts for judge by clear and convincing (within‑band findings and modality):

• Specific item codings for C, H, R, T, and P that do not alter the statutory ceiling (e.g., Hauto magnitude; R corroboration met; grooming present).
• AEH item details once the ceiling is unlocked by a jury finding.
• BCC audit: whether fair opportunity existed (state bears C&C). Draft: “Fair‑Opportunity‑to‑Comply shall be determined by the court by clear and convincing evidence, with the State bearing the burden.”

3. Resource and workload (order‑of‑magnitude)

Agencies and roles

• Sentencing Commission: rules, tool, training, dashboards, audits. Staff: +6 FTE (policy counsel, data scientist, trainer, QA, project manager, privacy analyst).
• Courts (AOC): judicial education; IT integration. Staff: +4 FTE (trainer, analyst, helpdesk, integration dev).
• AG/County Counsel: MOUs; privacy/GDPR‑like compliance. Staff: +3 FTE (counsel, data‑sharing attorney, privacy).
• Public Defenders/Prosecutors: training and coding expertise. Stipends for 2 “sentencing specialists” per large office; shared resources for small counties.
• HIE/Hospitals/Providers: interface development; data liaisons. Small grants for integration costs; +1 FTE liaison per hospital region; provider mini‑grants.

Budget

• Startup (18 months): $18–25M statewide (tooling $4–6M; training $3–4M; data integrations $6–8M; staffing $5–7M; evaluation $1–2M).
• Ongoing (annual): $6–9M (maintenance, audits, dashboards, training refreshers, privacy compliance).

4. Guarding against gaming and inequity

• BCC bad‑faith documentation
  • Random file audits (10% BCC cases per quarter); sanctions for agencies misrepresenting service availability; judicial contempt powers if parties mislead the court.
  • Defense access to agency records under protective orders; ex parte motions allowed for safety‑sensitive data.
• Unequal access to non‑police evidence
  • Court‑ordered production protocols; standing subpoenas to hospitals with privacy protections; Commission grants to providers for documentation capacity; a centralized “Evidence Liaison” office to coordinate.
  • Fallback affidavits: where records are unattainable, sworn third‑party affidavits (advocates/clinicians) acceptable; caution flags on S output with reason‑giving required.
• Differential surveillance inflating P/R
  • Enforce corroboration rule; discount location‑only R; require non‑police corroborants where feasible.
  • Quarterly disparate‑impact audits of P and R items by race/zip; if disparities persist unexplained, Commission must suspend/adjust weights within 60 days.

5. Appellate and oversight

• Specialized docket
  • Create a temporary Sentencing Review Panel (3‑judge rotating) for two years to hear S/P/AEH/BCC appeals on an expedited basis.
• Standards of review (updated)
  • Legal adherence to the tool, guardrails, corroboration rule, and BCC standard: de novo.
  • Item codings (fact findings): clear error; jury fact sufficiency: substantial evidence.
  • Lenity/uncertainty triggers and band selection: de novo for compliance; abuse of discretion within band.
• Public dashboards (quarterly; county/statewide)
  • Band distributions; within‑band modalities; reversal rates by ground; item‑level usage rates; DIF flags raised and resolved; BCC invocation and outcomes; demographic disparities (race, gender, indigence); SV/IPV safety metrics (time‑to‑violation; weapon relinquishment compliance).
• Corrective triggers
  • If any offense category shows an unexplained disparity (e.g., Black defendants 1.5× odds in higher bands controlling for S) for two quarters, automatic Commission review and interim variance authority.
  • If reversal rates for a judge/county exceed 15% on S/BCC grounds, mandatory training and case sampling.

6. Pilot sentencing statement template (max 2–3 pages)

Caption; Offense; Case No.

A. Seriousness computation (attach worksheet)

• Components: C=, H= (Hauto %), R=, T=, P= (SV/IPV only), AEH=__ (if applicable).
• Ŝ = ; 95% interval [, __]; applicable band per Commission grid: Band __.
• Thresholds: interval crosses [yes/no]. Lenity applied [yes/no]. If “no,” reasons and clear/convincing proof of upward facts: [cite evidence].

B. Fair‑Opportunity‑to‑Comply (BCC)

• Finding: Met / Not met (State’s burden C&C).
• Sources: [agency records; provider affidavits; defense rebuttal].
• If not met: applied one‑band downward adjustment [yes/no]; if “no,” imminent risk findings and reasons.

C. Public‑risk corroboration (if R elevation)

• Corroborants: [video/forensic/medical/affidavits/PO findings]; location‑only enhancement discounted [yes/no].

D. Sanction and rationale

• Selected sentence within Band __: [custody months; supervision years; modality package].
• Prevention (PIS) analysis: incapacitation benefit; deterrence (weapon‑use enhancements/certainty investments); specific change plan; collateral costs.
• Reparative package (if any): [obligation points; service hours; income‑based payments; treatment; acknowledgment form]; victim participation: direct/surrogate/declined; survivor‑safety controls: [no‑contact; geofencing; firearm relinquishment verified].
• Equality/bias check: no forbidden proxies; structural mitigators considered [yes/no].

E. Dignity and reasons

• Statement that sanction communicates censure, restores assurance, and is least burdensome sufficient to do so.

Judge’s signature/date.

7. Likely failure modes and pre‑planned fixes

• Inconsistent P coding across courts
  • Fix: statewide calibration clinics with live vignettes; issue a P Coding Guide with exemplars; hotline for real‑time guidance; require a second‑reader review for P≥7 during the first year; quarterly inter‑rater reliability reports; if IRR < 0.75, emergency clarifications.
• Data‑feed bottlenecks (hospital/provider records)
  • Fix: temporary evidentiary presumption allowing sworn provider/advocate affidavits; Commission micro‑grants to providers; emergency rule extending timelines; phased enforcement of corroboration rule where feeds are not yet live, with mandatory reasons and caution flags.
• Over‑ or under‑invocation of BCC
  • Fix: automated BCC checklist in the tool; training with hypotheticals; quarterly audits; if under‑invoked in high‑deprivation circuits, require sua sponte inquiry; if over‑invoked without basis, sanctions and remedial training.

8. Final stance on capital

The BCC audit and corroboration rules only strengthen my prior view. I would recommend a formal statewide directive:

• “Effective immediately, capital charging is suspended pending Commission verification that (a) race‑of‑victim/defendant and county effects are statistically non‑significant within a 1.05 odds ratio tolerance; (b) an independent innocence audit demonstrates a wrongful‑conviction risk < 1 in 10,000; and (c) uniform charging and review protocols are operational. Absent such demonstration within 24 months, capital punishment shall be abolished by statute.”

Closing thought

We are not going to eliminate judgment from sentencing. But with this architecture—protected culpability floors, validated P and AEH modules, corroboration requirements, a BCC audit with a state‑side burden, narrow bands, lenity triggers, written reasons, and appellate policing—we can do what Bedau demanded: push arbitrariness down and make discretion transparent, reviewable, and correctable. I’m comfortable standing up a pilot in 12 months and a statewide rollout by 18, if we resource it and commit to iterating in public.

---

Speaker: Host

We set out with Bedau’s challenge: retributivism’s justificatory core (R3) risks “moral alchemy,” and proportionality (R2) drifts into arbitrariness once talion is abandoned. The panel converged on a practical architecture that narrows arbitrariness and renders reasons reviewable, while candidly disagreeing on ultimate foundations.

On justification, three nonconsequentialist routes were articulated: Strauss’s deontic moral accounting (answerability to past agency under dignity constraints), Sokolov’s internalist authority (censure-plus-burden as constitutive of law’s assurance practice), and Qureshi’s reparative accountability (desert as a claim to answer and repair). Holm defended a preventive regime bounded by legitimacy side-constraints that explain the guilty/innocent asymmetry without desert. Baptiste pressed the non-ideal condition: under background injustice, penal authority requires distributive preconditions and anti-domination guardrails.

On proportionality, Chen’s multidimensional “S” gained broad uptake, upgraded to include: (1) a coercive-control dimension (P) with protected weight; (2) autonomy-loss weight floors in SV/IPV; (3) an Aggregate Economic/Environmental Harm module (AEH) to elevate diffuse harms; (4) quantified uncertainty with lenity triggers; (5) corroboration rules and non-police data to blunt policing saturation; and (6) open code, DIF testing, and democratic re-estimation. Villanueva translated this into a workable grid with narrow bands, written reasons, appellate oversight, and a Fair‑Opportunity‑to‑Comply audit that presumptively shifts away from custody where baseline state duties were unmet.

Policy upshots: strict liability rejected; culpability floors and dignity constraints codified; reparative packages authorized as punitive when censure‑backed and burdensome; prevention (PIS) optimizes within bands. There was near-consensus on practical abolition of the death penalty under current error and disparity conditions.

Enduring debates remain over the ultimate moral basis of punishment, the scope of custody for grave wrongs where robust repair is feasible, the stringency of distributive defeaters, and how to model legitimacy without importing racialized sentiment. But the session showed a credible path from theory to institutions: transparent metrics, calibrated burdens, distributive audits, and structured discretion with reasons and review.