07 — Crime, Policing & Incarceration

07 — Crime, Policing & Incarceration

STAGE 1: EVIDENCE BRIEF

Referee panel: Prof. Olivia Brandt (#57, statistics, C) · Prof. June Takahashi (#17, econometrics, C) · Domain specialist Prof. Patricia Nwosu (#36, criminology, C)

Ground rules from the panel. Every number below comes from a source we opened, and the year of the data is given. Where our only access was a secondary summary we say so. Nothing here is from memory. Two items that we could not re-verify this session are marked [unverified] and must not be used as load-bearing evidence by anyone in later stages.

---

A. Crime levels and trends

F1. The 2020 spike. Murder and nonnegligent manslaughter rose 29.4% by volume from 2019 to 2020. Violent crime overall rose 5.6% (rate +5.2%). — FBI, 2020 Crime in the Nation (data 2020). https://www.fbi.gov/news/press-releases/fbi-releases-2020-crime-statistics

F2. The 2025 decline (official, FBI). The violent crime rate fell 9.3% from 2024 to 2025, to 327.6 per 100,000 (362.9 in 2024). The FBI calls this the "largest year-to-year decline in violent crime rates since FBI estimations began in 1936." Other changes: murder −18.1%, robbery −18.5%, rape −7.6%, aggravated assault −7.2%, property crime −12.4%, burglary −15.8%, motor vehicle theft −22.7%. The murder rate was 4.1 per 100,000, tied with 1955–56 as the lowest in FBI records. More than 17,000 agencies reported, covering 96.2% of the population. Released Aug 14, 2026. — https://www.fbi.gov/news/press-releases/fbi-releases-2025-reported-crimes-in-the-nation-statistics ; https://www.fbi.gov/news/stories/violent-crime-falls-at-historic-rate-new-fbi-data-show

F3. 2026 so far (Council on Criminal Justice, 36 large cities, H1 2026 vs H1 2025). Homicide −18% (215 fewer); −31% vs H1 2019; −51% vs the 2022 peak. Carjacking −47%, motor vehicle theft −20%, robbery −17%, residential burglary −13%, aggravated assault −2%. Increases: sexual assault +3%, domestic violence +8%, drug offenses +12%, shoplifting +4%. Shoplifting is the only category still above 2019, by 5% (Stateline's reading of CCJ). "Lethality" (homicides per 1,000 assaults and robberies) fell 23%. Four sample cities remain above 2019 homicide levels: Austin, Minneapolis, Norfolk and Pittsburgh. — https://counciloncj.org/crime-trends-in-u-s-cities-mid-year-2026-update/ ; https://stateline.org/2026/07/24/crime-continues-historic-decline-as-homicide-rates-approach-century-lows/

F4. Historical framing. CCJ projected the 2025 homicide rate at about 4.0 per 100,000, "the lowest rate ever recorded… going back to 1900," and projects a further historic low in 2026. CCJ also states: "It's extremely difficult to disentangle and pinpoint what's actually driving the drop." — https://counciloncj.org/state-of-the-union-why-is-crime-going-down/

F5. Real-Time Crime Index (Jeff Asher). Murder was down 30% in January 2026 in a sample of about 500 agencies covering about 100M people. Gun Violence Archive shooting victims were down 12% through February. Asher warns that January figures are unreliable and that "there is some evidence that the decline is starting to slow." — https://jasher.substack.com/p/murder-and-gun-violence-are-still

B. Measurement: NCVS vs UCR/NIBRS

F6. NCVS 2024. The violent victimization rate was 23.3 per 1,000 persons aged 12+, "similar to 2023." The share of youth aged 12–17 victimized rose from 1.45% to 1.95%. Violent victimizations reported to police ran at 11.2 per 1,000, so roughly half go unreported. — BJS, Criminal Victimization, 2024. https://bjs.ojp.gov/library/publications/criminal-victimization-2024

F7. The two measures diverge. NCVS violent crime excluding simple assault went from 6.8 per 1,000 (2015) to 5.6 (2021), then 9.8 (2022) and 8.9 (2024). NIBRS violent crime stayed at "3.7 to 4.0 per 1,000" for the whole decade. NIBRS property crime fell 28% (24.9 to 17.9 per 1,000). — BJS, The Nation's Two Crime Measures, 2015–2024. https://bjs.ojp.gov/library/publications/nations-two-crime-measures-2015-2024

F8. NCVS precision caveat. The 2024 NCVS used a smaller sample during a redesign, which produced "larger standard error." On a comparable time window, the FBI–NCVS gap in auto theft shrinks from about 19% vs flat to about 7%. — Asher. https://jasher.substack.com/p/deciphering-the-2024-national-crime

C. Clearance

F9. Murder clearance rate. About 50% in 2022 (the post-COVID low) and 66.5% in 2025, "likely the highest this century" per FBI Crime Data Explorer. Asher estimates CDE understates the true rate by about 0.5 points a year after the NIBRS transition. He also notes that 1960s-era rates "probably shouldn't be believed" because arrest standards were looser. — https://jasher.substack.com/p/murders-in-2025-were-likely-solved

D. Incarceration

F10. International comparison (World Prison Brief, 14th ed., May 2024). 11.5M people are held worldwide, a world rate of 140 per 100,000. The US rate is 531 per 100,000, ranked 6th, behind El Salvador (1,086), Cuba (794), Rwanda (637), Turkmenistan (576) and American Samoa (538). WPB notes the US peaked above 750 in the 2000s. — https://www.prisonstudies.org/news/prison-populations-continue-rise-many-parts-world-115-million-held-prisons-worldwide

F11. Decline since 2008. The combined prison-plus-jail rate peaked at about 755–760 per 100,000 (2008) and was 541 in 2022. [Secondary: Wikipedia's compilation of BJS and WPB. The panel treats the direction and rough size as reliable and the exact decimals as approximate.] — https://en.wikipedia.org/wiki/United_States_incarceration_rate

F12. The decline has stalled. The prison population was 1,254,200 at yearend 2023, up 24,100 (+2%) from 2022. Women rose about 4%. — BJS, Prisoners in 2023. https://bjs.ojp.gov/library/publications/prisoners-2023-statistical-tables. Vera estimates the spring 2024 prison population at 13% below 2019. — https://www.vera.org/publications/people-in-jail-and-prison-in-2024

F13. Racial disparity. Black Americans are imprisoned at about 5× the white rate (2019 data). The lifetime likelihood of imprisonment for Black men fell from 1 in 3 (born 1981) to 1 in 5 (born 2001). — The Sentencing Project, One in Five. https://www.sentencingproject.org/reports/one-in-five-racial-disparity-in-imprisonment-causes-and-remedies/

F14. Recidivism. Of people released from prison in 34 states in 2012, 62% were rearrested within 3 years and 71% within 5, and 46% returned to prison within 5 years. Five-year rearrest was 81% for those aged 24 or younger and 61% for those 40 and older. For the 2008 release cohort in 24 states, 82% were rearrested within 10 years and 61% returned to prison. The share arrested each year fell from 43% in year 1 to 22% in year 10. — BJS. https://bjs.ojp.gov/library/publications/recidivism-prisoners-released-34-states-2012-5-year-follow-period-2012-2017 ; https://bjs.ojp.gov/library/publications/recidivism-prisoners-released-24-states-2008-10-year-follow-period-2008-2018 (Note: rearrest is not reconviction, and cohort-of-releases figures overweight frequent offenders.)

E. Police: size, effects, violence, accountability

F15. Police and crime (elasticity). Correcting for measurement error, the elasticity of cost-weighted crime with respect to police is about −0.5, and police reduce violent crime more than property crime. — Chalfin & McCrary, NBER w18815 (1960–2010 city panel). https://www.nber.org/papers/w18815

F16. Police force size and civilian race. "Each additional police officer abates approximately 0.1 homicides." The per-capita effect is "twice as large for Black versus white victims." Larger forces also raise low-level arrests, "a disproportionate burden for Black Americans." — Chalfin, Hansen, Weisburst & Williams, AER: Insights 2022 (NBER w28202). https://www.nber.org/papers/w28202

F17. Killings by police: the two main trackers disagree on level and trend.

• Mapping Police Violence (Campaign Zero): 1,314 killed in 2025, about 5% lower than 2024 and the "first decline in six years." The rate for Black people is 2.6× the white rate. — https://campaignzero.org/research/mapping-police-violence-for-the-first-time-in-six-years-police-violence-declined-in-2025/
• Police Violence Report 2025: at least 1,201 killed in 2025, down slightly from a record 1,271 in 2024. Fewer than 3% resulted in charges against officers, 98 of those killed were unarmed, and 95% of deaths were shootings. — https://davisvanguard.org/2026/01/police-violence-report-2025/ ; https://policeviolencereport.org/

F18. FBI National Use-of-Force Data Collection. Participation covered 72% of the law-enforcement population for Apr 2023–Mar 2024. Incident counts are published only once participation reaches 80%. — https://www.fbi.gov/news/press-releases/fbi-releases-2024-quarterly-crime-report-and-use-of-force-data-update

F19. National Decertification Index (IADLEST). The index holds more than 53,500 records, with 49 POST agencies contributing. Rhode Island's POST commission "does not have decertification authority." — IADLEST NDI white paper (2024). https://archive.legmt.gov/content/Committees/Interim/2025-2026/LJIC/Sept-10-2025/4b.NDI-Whitepaper-2024.pdf

F. Prosecution and pretrial

F20. Misdemeanor nonprosecution (Suffolk County, MA). Not prosecuting a nonviolent misdemeanor leads to a 53% reduction in the likelihood of a new criminal complaint within 2 years and 60% fewer subsequent complaints. The effect is largest for people with no prior record. Identification comes from as-if-random assignment of arraignment prosecutors, and a policy presumption of nonprosecution showed similar effects. — Agan, Doleac & Harvey, QJE 2023. https://www.nber.org/papers/w28600

F21. New Jersey bail reform (2017). The pretrial jail population fell 43.9% from Jan 2015 to Jan 2018, and MDRC found "no uptick in pretrial crime" in its early evaluation. — https://www.mdrc.org/news/press-release/study-finds-new-jersey-s-pretrial-reform-shows-early-signs-wide-reaching-positive

F22. New York bail reform (2020), long-term recidivism. Data Collaborative for Justice (John Jay), 2025:

• NYC, cases where bail was eliminated (mandatory release): rearrest 57% vs 66% before reform, felony rearrest 33% vs 40%, violent felony rearrest 20% vs 25%.
• Suburban and upstate NY: no detectable effect.
• People with a recent violent-felony arrest (under 15% of cases): rearrest 84% vs 80%, felony rearrest 70% vs 63%, violent felony rearrest 46% vs 40%. These are increases.

— https://datacollaborativeforjustice.org/wp-content/uploads/2025/10/Longterm_Recidivism_Report.pdf

F23. Illinois cash-bail abolition (Pretrial Fairness Act, part of SAFE-T, effective Sept 18, 2023).

• Loyola (Olson), first year: failure to appear went from about 17% to about 15% ("statistically very similar"), and median detention hearings lengthened from 4 to 16 minutes. About $140M a year in bond payments ended. No documented crime increase when comparing H1 2023 with H1 2024. — https://capitolnewsillinois.com/news/a-year-after-end-of-cash-bail-early-research-shows-impact-less-than-many-hoped-or-feared/
• Later data: the statewide pretrial jail population fell 7%, Cook County's rose 2%, electronic monitoring rose 33%, and total pretrial supervision (jail plus EM) rose 17%. — https://www.stlpr.org/government-politics-issues/2026-01-14/top-democrats-would-consider-changes-illinois-safe-t-act

G. Drugs

F24. Overdose deaths. Final 2024 data: 79,384 deaths (23.1 per 100,000) vs 105,007 in 2023, down 26.2%. Synthetic opioid (mostly fentanyl) deaths fell from 72,776 to 47,735 (−35.6%). — CDC NCHS Data Brief 549. https://www.cdc.gov/nchs/products/databriefs/db549.htm

F25. Oregon Measure 110 (decriminalization, 2021), reversed by HB 4002 (2024). Zoorob et al. (JAMA Network Open, 2024) found a positive association between M110 and overdose that "changed signs" once they adjusted for fentanyl's arrival, leaving "no association." A JAMA Psychiatry one-year study of Oregon and Washington (2023) also found no association. — https://www.rti.org/publication/drug-decriminalization-fentanyl-fatal-overdoses-oregon ; https://pubmed.ncbi.nlm.nih.gov/37755815/ ; HB 4002 vote explanation: https://www.oregonlegislature.gov/Jama/Documents/HB%204002%20Vote%20Explanation.pdf

H. Violence prevention

F26. READI Chicago (RCT, n = 2,456 men at very high risk). The program combined 18 months of paid work, CBT and support. It produced no statistically significant reduction in the primary violence index. Shooting and homicide arrests fell 65% (p = 0.13 after multiple-testing adjustment). In the outreach-referred subgroup, arrests fell 79% and victimizations 43%, both significant after adjustment. Estimated social savings were $182k–$916k per participant (benefit-cost 4:1 to 18:1), and take-up was 55%. The control group suffered 11 shooting or homicide victimizations per 100 people in 20 months. — Bhatt, Heller, Kapustin, Bertrand & Blattman, QJE 2024. https://www.nber.org/papers/w30852

F27. Community violence intervention (CVI) has mixed results. Baltimore Safe Streets: −23% nonfatal shootings. Chicago CeaseFire: significant declines in 5 of 7 sites. Phoenix TRUCE: assaults fell, but shootings rose +3.2 a month. Pittsburgh One Vision One Life: no homicide reduction and higher aggravated and firearm assault. Seattle's hospital-based program (HiFi) found no difference in outcomes. — Univ. of Washington CFIP, Review of CVI Strategies (July 2025). https://cfip.uw.edu/wp-content/uploads/2026/01/Review-of-Community-Violence-Intervention-Strategies_Final-REPORT_July-2025.pdf

F28. Focused deterrence (meta-analysis). Across 24 quasi-experimental evaluations there is a significant, moderate crime reduction, but effects were "smaller for evaluations with more rigorous research designs." — Braga, Weisburd & Turchan, Criminology & Public Policy 2018. https://www.ojp.gov/ncjrs/virtual-library/abstracts/focused-deterrence-strategies-and-crime-control-updated-systematic

I. Guns

F29. RAND Gun Policy in America (updated Jan 29, 2026).

• Supportive evidence: child-access prevention (CAP) laws reduce youth firearm suicides and self-injuries, youth firearm homicides and assault injuries, and unintentional child deaths. Minimum-age laws reduce youth firearm suicides. Stand-your-ground laws increase firearm homicides. Shall-issue concealed-carry laws increase total homicides, firearm homicides and overall violent crime.
• Moderate evidence: background checks reduce firearm and total homicides; waiting periods reduce firearm suicides and total homicides; domestic-violence prohibitions reduce intimate-partner homicides; licensing reduces firearm and total homicides.
• Limited evidence: extreme-risk protection orders (ERPOs) and assault-weapon bans, among others.
• Inconclusive: 50 policy-outcome pairs.

— https://www.rand.org/research/gun-policy/key-findings/what-science-tells-us-about-the-effects-of-gun-policies.html

---

CONTESTED EVIDENCE (where serious researchers disagree)

C1. Why crime fell after 2022. CCJ lists a menu of explanations with no ranking: enforcement strategy, technology, post-pandemic social normalization, CVI expansion, "controversial claims" about immigration enforcement and National Guard deployments, and gun-market dynamics (F4). The decline is broad across regions and city types (F3). As Nwosu puts it, that points toward macro forces and away from any single local policy. Anyone who claims their favored policy caused 2023–26 carries the burden of explaining Austin, Minneapolis, Norfolk and Pittsburgh. The honest answer is Unknown.

C2. Which ruler is right, NCVS or UCR? The NCVS shows a jump in 2022 and no 2024 decline, while NIBRS is flat (F6–F8). One camp reads the NCVS as capturing unreported crime, which a declining reporting rate would hide from police data. The other points to 2024's reduced sample and wide confidence intervals. For homicide the dispute is moot: the NCVS does not measure it, and death certificates corroborate the police counts.

C3. Police: benefits vs burdens. Chalfin–McCrary (F15) and Chalfin et al. (F16) find large life-saving effects. The same F16 paper documents a heavier load of low-level arrests on Black residents. The literature does not agree on how much of the effect comes from proactive stops versus visibility and clearance.

• Stop-and-frisk: Weisburd et al. (2016, Criminology & Public Policy) report small micro-place deterrent effects of NYC stops. MacDonald, Fagan & Geller (2016) find that most stops had little effect, apart from those grounded in probable cause. [both unverified this session; direction only]

C4. Bail reform. The pattern differs by place and population:

• New Jersey (F21) combined a risk tool with a strong detention track and saw jail populations fall with no measured crime cost.
• New York (F22) cut recidivism on average in NYC, showed nothing upstate, and saw increases for the recent-violent-felony subgroup.
• Illinois (F23) looks close to neutral on failure to appear, but supervision was relabeled rather than removed: total pretrial supervision rose 17%.

Critics argue that recidivism among released people is the wrong estimand if reform also changed arrest behavior (MDRC found fewer arrests in NJ).

C5. Oregon M110. Zoorob et al. (F25) find no overdose effect after adjusting for fentanyl. Spencer (2023, J. Health Econ.) reported a positive association using synthetic control [finding direction known to panel; magnitude unverified this session]. Separately, public-order outcomes (open use, disorder) were rarely measured rigorously, and that gap drove the political repeal.

C6. CVI and READI. Proponents cite the positive sites (F27) and READI's subgroup and cost-benefit results (F26). Skeptics note the null primary outcome in the only large RCT, the negative sites, and the shrinking effect sizes in rigorous designs. Focused deterrence shows the same attenuation (F28).

C7. Guns. RAND's "supportive" ratings are strong on some policies but concentrated in a few outcome pairs, and 50 policy-outcome pairs remain inconclusive. RAND's shall-issue finding rests mainly on post-2000 synthetic-control and panel studies. Defenders of carry rights dispute those models' specification, and that dispute is live in the literature.

C8. Counting police killings. The two trackers differ by about 100 deaths in 2025 (1,314 vs 1,201) and disagree on whether 2024 was a record (F17). Any claim about "the deadliest year" depends on which tracker is used.

— Brandt, Takahashi, Nwosu

STAGE 2: OPENING POSITIONS (Part 1 of 2: Left bloc)

---

Prof. Jerome Price (#24, criminal justice, L): "Decarceration is unfinished, and the 2025 data prove we can afford to finish it"

Causal claim. American incarceration is set by policy choices (sentence length, charging, pretrial detention), not by crime. Our prison and jail rate fell from roughly 755 to roughly 541 per 100,000 after 2008 (F11). Over the same period violent crime fell, and in 2025 it posted its largest one-year drop since 1936 (F2). The deterrence story told in the 1990s predicted the opposite. The United States still has the highest incarceration rate in the world, and nothing in our crime profile justifies it.

Evidence.

• Agan, Doleac and Harvey (F20) find that not prosecuting a nonviolent misdemeanor cuts new criminal complaints within two years by 53%. That is the strongest causal estimate we have that the system's own contact is criminogenic at the margin.
• BJS age gradients (F14) show five-year rearrest falling from 81% for people 24 and under to 61% for people 40 and over. Long sentences hold people well past their peak offending years.

Remedy.

1. Presumptive nonprosecution or diversion for nonviolent misdemeanors.
2. "Second look" resentencing after 15 years.
3. Cap consecutive enhancements.

Tradeoff I concede. Incapacitation is real for a small, very active group. The BJS 10-year cohort (F14) shows 82% rearrested, and I will not pretend release is costless for everyone.

Falsifiable prediction. If I am right, jurisdictions that adopt misdemeanor nonprosecution presumptions will show equal or lower two-year re-offense among the affected population, and no rise in local violent crime relative to comparable jurisdictions, within 3 years. If violent crime rises 10% or more relative to matched controls, I am wrong.

---

Prof. Marcus Bell (#22, civil rights law, L): "We cannot govern what we refuse to count"

Causal claim. Accountability failures in policing (no complete use-of-force data, porous decertification, near-zero prosecution) reduce the legitimacy that clearance and cooperation depend on. The 2025 crime drop does not change that. Police killings have risen every year since 2020, and 2024 was the deadliest year on record. On the federal side, the FBI's use-of-force database is effectively dead: police departments simply do not report.

Evidence.

• Police Violence Report: a record 1,271 killed in 2024, at least 1,201 in 2025, and fewer than 3% of cases led to charges (F17).
• Mapping Police Violence: Black Americans are killed at 2.6× the white rate.
• Decertification: Rhode Island's POST commission cannot decertify at all (F19).
• Race and enforcement: Chalfin et al. (F16) show that more officers means more low-level arrests, falling disproportionately on Black residents.

Remedy.

• Make Byrne JAG and COPS money conditional on (a) 100% participation in federal use-of-force reporting and (b) state decertification authority plus mandatory NDI reporting.
• Publish agency-level data.

Tradeoff I concede. Data mandates cost small agencies money. Transparency can also produce "Ferguson effect" pullbacks. I do not think that evidence is strong, but I concede it is contested.

Falsifiable prediction. If legitimacy matters, agencies that adopt full transparency and decertification regimes will show higher clearance rates for nonfatal shootings within 5 years than comparable agencies, holding staffing fixed.

---

Prof. Rosa Delgado-Finch (#7, political economy, Soc): "Shift the marginal dollar from patrol to people"

Causal claim. Violence is concentrated among a few thousand young men in a few hundred blocks. Programs that work directly with those men beat marginal patrol. READI Chicago cut shooting and homicide arrests by 65%, and Cure Violence-style interruption reduces shootings wherever it is implemented with fidelity.

Evidence.

• READI (F26): a 65% reduction in shooting and homicide arrests, and a 4:1 to 18:1 benefit-cost ratio.
• Baltimore Safe Streets (F27): −23% nonfatal shootings.
• Chicago CeaseFire: gains in 5 of 7 sites.
• Police have a cost too. Each additional officer brings more low-level arrests with a racial skew (F16).

Remedy. Move 10% of big-city police budgets into CVI, paid employment and CBT programs for the highest-risk men, run by community organizations rather than police departments.

Tradeoff I concede. CVI workers face real danger. Some programs have been captured by patronage, and scaling quality is hard.

Falsifiable prediction. If I am right, cities that make that 10% shift will see nonfatal shootings fall at least as fast as in peer cities within 3 years. If they fall more slowly by 15 points or more, the shift failed.

---

Prof. Gabriela Ortiz (#43, epidemiology, C-L): "Fentanyl, not decriminalization, drove Oregon's deaths, and supply changes are driving the national decline"

Causal claim. The overdose epidemic is a supply-toxicity problem. Oregon's Measure 110 was politically botched but did not measurably raise deaths once fentanyl's arrival is accounted for (F25). The 26% national drop in overdose deaths in 2024 (F24) was driven by naloxone saturation, not by arrests.

Evidence.

• Zoorob et al. (JAMA Netw Open 2024): the sign flips after adjusting for fentanyl.
• The 2023 JAMA Psychiatry study of Oregon and Washington: no association.
• CDC: synthetic-opioid deaths fell 35.6% in 2024.

Remedy.

• Keep possession a low-level offense with deflection to treatment, as in HB 4002, and actually fund it.
• Scale naloxone, medications for opioid use disorder (MOUD), and drug checking.
• Prosecute trafficking.

Tradeoff I concede. Oregon failed to build treatment capacity before decriminalizing, and I will not defend the implementation. Public disorder was a real harm that the epidemiologic literature barely measured.

Falsifiable prediction. If supply is the driver, states' 2025–27 overdose trajectories will track local fentanyl purity and adulterant data (for example, xylazine or nitazene prevalence) far more closely than their possession-law regime.

STAGE 2: OPENING POSITIONS (Part 2 of 2: Center and Right)

---

Prof. Samuel Birch (#10, urban econ, C): "Police are underprovided, and clearance is the lever"

Causal claim. On the margin, police save lives, and they save disproportionately Black lives.

• The measurement-error-corrected police elasticity of cost-weighted crime is about −0.5 (F15).
• Each officer abates about 0.1 homicides (F16), so roughly 10 officers per life saved per year.
• In the same paper, the per-capita homicide benefit is twice as large for Black victims.

The murder clearance rate recovered from about 50% (2022) to 66.5% (2025) (F9), and certainty of apprehension is the one deterrence parameter criminologists agree matters.

Remedy.

• Targeted federal matching funds for officers in the top-100 homicide cities.
• Earmark detective and forensic capacity, since clearance is the channel.
• Pair the funding with focused deterrence (F28).

Tradeoff I concede. F16 also shows more low-level arrests, disproportionately of Black residents. Hiring should come with a de-emphasis of quality-of-life enforcement.

Falsifiable prediction. If this is right, the cities that grow detective staffing fastest will show the largest gains in homicide and nonfatal-shooting clearance and larger homicide declines over 2026–29, conditional on 2025 levels.

---

Prof. Colleen Farrow (#34, local government, C): "Bail reform works when it is designed, fails when it is slogans"

Causal claim. The design of pretrial reform matters more than its label. New Jersey pairs a risk assessment with a real detention hearing. Its pretrial jail population fell 43.9% with no uptick in pretrial crime (F21).

• New York removed judicial discretion for many charges. Recidivism fell on average in NYC but rose for people with recent violent-felony arrests (F22).
• Illinois kept failure to appear essentially flat, about 17% to about 15% (F23), but electronic monitoring grew 33%.

Remedy. Use the NJ model everywhere:

• End wealth-based detention.
• Use a validated tool.
• Hold adversarial detention hearings with presumptive detention for defendants who have a recent violent-felony arrest.
• Publish outcome data quarterly.

Tradeoff I concede. Detention hearings take court time (Illinois went from 4 to 16 minutes). Electronic monitoring can become a new net.

Falsifiable prediction. States that adopt NJ-style hearings will see pretrial populations fall 25% or more within 3 years, with no rise in violent re-arrest among released defendants.

---

Prof. Eleanor Whitfield (#21, con law/originalist, R): "Order came back when enforcement came back"

Causal claim. The 2020 surge followed de-policing and prosecutorial retreat. Murder rose 29.4% (F1). The 2023–26 decline followed re-policing, the restoration of pretrial detention in practice, and cities abandoning the "defund" posture. New York's bail reform increased recidivism. Murder clearance has collapsed to about 50%, which means half of killers walk. Incapacitation is not a slogan: 82% of released prisoners are rearrested within 10 years (F14).

Remedy.

• Restore judicial discretion to detain dangerous defendants.
• Fund police hiring.
• Prosecute gun possession by prohibited persons vigorously.
• Oppose blanket nonprosecution.

Tradeoff I concede. Long sentences for older offenders have sharply diminishing returns, per the same BJS age gradient.

Falsifiable prediction. Cities that cut sworn staffing or adopt blanket nonprosecution in 2026–28 will underperform peers on homicide by 10% or more within 3 years.

---

Prof. Walter Pruitt (#8, law & econ, Lib): "Decriminalize the victimless, arm the law-abiding, and stop the federal data mandates"

Causal claim. Drug prohibition creates the black markets that drive violence. On guns, shall-issue concealed carry reduces violent crime, as the Lott line of work showed, and stand-your-ground laws deter attackers. On police data, Washington should not dictate to 18,000 agencies.

Evidence. Violent crime fell to a 4.1 murder rate (F2) while concealed-carry permitting expanded nationally. The Oregon studies (F25) show that decriminalization did not kill anyone.

Remedy.

• Legalize and regulate marijuana federally and decriminalize personal possession of other drugs.
• Preserve carry rights.
• Replace mandates with state-level qualified-immunity reform and civil liability.

Tradeoff I concede. Decriminalization without a public-order code (camping, public use) invites disorder, and Oregon shows that.

Falsifiable prediction. States that expanded permitless or shall-issue carry after 2015 will show no relative increase in firearm homicide by 2028.

---

Prof. Evelyn Sutter (#78, philosophy of law, C-R): "Oregon was a controlled experiment and the voters graded it"

Causal claim. Law expresses norms, and removing sanction removed a lever to compel treatment. Measure 110 caused Oregon's overdose surge. The legislature's 2024 reversal through HB 4002 (F25) was a democratic verdict on observable disorder.

Evidence. Spencer (2023) found increased overdose deaths after M110 (C5). Treatment uptake through the M110 hotline was trivially small (I will substantiate this if challenged).

Remedy. Treat possession as a misdemeanor with deflection and drug courts. Coerced treatment through courts is a legitimate tool.

Tradeoff I concede. Jail without treatment recycles people. Oregon's recriminalization is not funded well enough to work.

Falsifiable prediction. Oregon's overdose death rate relative to Washington's and the national trend will improve within 2 years of HB 4002 (effective Sept 2024), controlling for fentanyl penetration.

STAGE 3: CROSS-EXAMINATION (Part 1)

---

Exchange 1: Whitfield (R) vs Farrow (C) and Price (L) on bail, clearance and the cause of the decline. Four rounds.

Farrow → Whitfield, round 1. Eleanor, you said "New York's bail reform increased recidivism." The Data Collaborative long-term report (F22) says the opposite for the average case. In NYC, rearrest after mandatory release was 57% vs 66% before reform, and violent felony rearrest was 20% vs 25%. Upstate showed no effect. The increase you can point to is in a subgroup under 15% of cases. And "clearance has collapsed to about 50%" is three years out of date. Asher's reading of FBI data puts 2025 at 66.5% (F9).

Whitfield, reply. On clearance I accept the correction: 50% was the 2022 figure, and I should have said so. It is also worth noticing that clearance recovered as cities rehired and rebuilt homicide units, which is my mechanism. On bail I will not concede the frame. The <15% subgroup is precisely the population that commits the violence we care about, and its violent felony rearrest rose from 40% to 46%. A reform that lowers shoplifting rearrest among low-risk people while raising violent rearrest among high-risk people is not a success merely because the average fell.

Farrow, round 2. That is a fair point about targeting, and it is exactly why I proposed presumptive detention hearings for that subgroup. But you made an empirical claim about the reform as a whole, and the reform as a whole reduced rearrest in NYC. Also, your mechanism predicts New Jersey should have blown up. It kept strong detention and cut the pretrial jail population 43.9% with no uptick in pretrial crime (F21).

Whitfield, round 2. New Jersey is my model too, and I have never objected to its design. It gives judges more detention power, not less. My claim is narrower now, and I will put it on the record: New York's 2020 design, which stripped judges of dangerousness discretion, increased violent rearrest among the recently violent. Your own figures support that sentence.

Price, round 3 (joining). Then let's be precise about the cause of the 2023–26 decline, because Eleanor's opening attributed it to "re-policing." CCJ's H1 2026 data (F3) shows homicide falling in almost every sample city, including cities that never re-expanded staffing. Four cities with very different policing regimes (Austin, Minneapolis, Norfolk, Pittsburgh) remain above 2019. CCJ's own CEO says the "striking consistency of the decline suggests that macro-level forces are exerting enormous influence" (F4). You cannot claim a national effect from local policy.

Whitfield, round 3. Jerome, you cannot have it both ways. If the decline is macro, then your decarceration also cannot claim credit for the 2025 drop, yet your opening leaned on it. I will concede that my "re-policing caused it" claim is unproven at the national level. I maintain that Chalfin–McCrary (F15) and Chalfin et al. (F16) make it likely that restored staffing contributed at the margin. They are the best causal estimates in the field.

Price, round 4. Agreed on symmetry. I withdraw "the 2025 data prove we can afford to finish it" as a causal claim. What they show is that the incarceration decline since 2008 has coincided with falling crime, which refutes the prediction that decarceration must raise crime. That is a weaker claim, and it is the one I will defend.

Whitfield, round 4. Accepted, with one amendment: the decline since 2008 was concentrated in low-level and drug admissions. It says little about releasing people convicted of violent offenses, who are now the majority of state prisoners. Status: both parties narrowed their claims. Neither caved without citing F3, F4, F9 or F22.

---

Exchange 2: Birch (C) vs Delgado-Finch (Soc) on police vs CVI. Three rounds.

Birch → Delgado-Finch. Rosa, READI's primary outcome was null (F26). The 65% drop in shooting and homicide arrests has p = 0.13 after adjustment. That is suggestive, not established, and you presented it as a result. Your claim that Cure Violence "reduces shootings wherever implemented with fidelity" runs into Phoenix TRUCE, where shootings rose 3.2 a month, and Pittsburgh, where assaults rose (F27). Meanwhile the officer-to-homicide ratio in Chalfin et al. is about 10 officers per life per year (F16). Moving 10% of police budgets out of patrol means giving up something we can measure in exchange for something we cannot.

Delgado-Finch. On READI, I accept that "cut" overstated it. The defensible claim is that the outreach-referred subgroup showed a 79% arrest reduction and a 43% reduction in victimization, both surviving multiple-testing correction, with a 4:1 minimum benefit-cost ratio. "Wherever implemented with fidelity" is doing work I cannot fully cash out: fidelity is judged after the fact, which makes the claim unfalsifiable as stated. I withdraw it. But Sam, your own paper's co-finding is the other half of the ledger. The same officers generate the low-level arrests (F16), and a criminal record is itself criminogenic according to Agan et al. (F20). You are counting the homicides prevented and ignoring the downstream records.

Birch, round 2. Fair point, and that is why I proposed coupling hiring with a de-emphasis of quality-of-life enforcement. But the magnitudes are not symmetric. Agan et al. measure new criminal complaints, mostly misdemeanors. Chalfin et al. measure lives. I would fund READI-type programs as additions, evaluated by RCT, not as substitutes financed by patrol cuts.

Delgado-Finch, round 2. Then we disagree about the budget constraint, not the evidence. I maintain that cities face a real constraint and that the marginal CVI dollar is under-tested, not proven worse. I will amend my remedy from "move 10%" to "fund at scale with mandatory randomized evaluation, and let results reallocate." I note that this concession comes from F26's null primary outcome and F27's negative sites, not from pressure.

Birch, round 3. That amended version I can vote for.

STAGE 3: CROSS-EXAMINATION (Part 2)

---

Exchange 3: Ortiz (C-L) vs Pruitt (Lib) on guns. Three rounds.

Ortiz → Pruitt. Walter, RAND's January 2026 update (F29) rates the evidence as supportive, its highest category, that shall-issue concealed-carry laws increase total homicides, firearm homicides and violent crime. It rates stand-your-ground laws as increasing firearm homicides. Your "Lott line of work" is precisely what the later synthetic-control literature overturned. And "crime fell while carry expanded" is a national time-series correlation. You would never accept that inference from me about decriminalization.

Pruitt. I accept the symmetry point: a national time-series co-movement proves nothing, and I withdraw it as evidence. On the substance, RAND's shall-issue rating leans on a small number of research teams using state-panel and synthetic-control methods with debated donor pools (C7). On stand-your-ground, I overstated it. RAND has no evidence of deterrence and supportive evidence of increased homicide, and I cannot cite a study that overturns it. I withdraw "stand-your-ground deters attackers."

Ortiz, round 2. And shall-issue?

Pruitt, round 2. I amend it to: "the effect of shall-issue carry on violent crime is disputed, and the best current synthesis finds harm." I still oppose repeal on rights grounds, which is a normative position, not an empirical one. I will also say that the same RAND review gives supportive ratings to child-access prevention (CAP) laws. Those are storage-liability rules that a libertarian can support as ordinary tort-style duties of care.

Ortiz, round 3. Noted, and that is the most useful sentence in this exchange.

---

Exchange 4: Whitmore (#46, pharmacology, C-R) vs Ortiz (C-L) on the cause of the overdose decline. Two rounds.

Whitmore → Ortiz. Gabriela, you said the 2024 drop in overdose deaths (F24) "was driven by naloxone saturation." What is the causal evidence? The 26% drop was nearly national (45 states, per CDC's release). Naloxone rollout was staggered over a decade. Supply-side changes are a rival explanation that your own opening favors: fentanyl purity and adulterant shifts, and possibly cartel dynamics. You cannot make supply the driver in Oregon and naloxone the driver nationally.

Ortiz. That is a fair hit. I have no study that apportions the 2024 decline, and I withdraw "driven by naloxone" and replace it with "coincided with naloxone expansion and supply changes; apportionment unknown." I keep the Oregon claim, because Zoorob et al. and the JAMA Psychiatry study are designed to answer exactly that question.

Whitmore, round 2. Accepted. I would add that the uncertainty cuts against the right as well. No one should credit border or trafficking enforcement for the 2024 decline without evidence either.

---

Exchange 5: Ortiz (C-L) and Mercer (#41, public health, L) vs Sutter (C-R) on Oregon. Three rounds.

Mercer → Sutter. Evelyn, "M110 caused Oregon's overdose surge" is contradicted by Zoorob et al. (F25). Their positive association flips sign once fentanyl is controlled, and the JAMA Psychiatry one-year study finds no association. Spencer (2023) is the lone positive estimate, and the panel flagged its magnitude as unverified. And you promised a hotline statistic. Where is it?

Sutter. On the hotline, I cannot produce a verified number, so I withdraw it. On causation, I amend "caused" to "is contested." Spencer's synthetic control uses states that fentanyl reached at different times, and Zoorob adjusts for that directly, which I grant is the better design for this question. But my core claim was never only about deaths. HB 4002 responded to public-order harms, and the literature did not measure them (C5). A policy can fail democratically without failing epidemiologically.

Ortiz, round 2. Agreed, and I conceded the implementation failure in my opening. But then the recriminalization cannot be justified by deaths either. Your falsifiable prediction (Oregon improving relative to Washington after HB 4002) has a problem: Washington also recriminalized in 2023 after the Blake decision. The comparison is contaminated.

Sutter, round 3. That is correct about Washington, and I will reformulate: Oregon's rate relative to the national trend, adjusted for fentanyl penetration. This is a real weakening of my test, and it came from Gabriela's point, not from pressure.

---

Exchange 6: Holm (#39, judicial politics, C-R) vs Bell (L) on data and decertification. Three rounds.

Holm → Bell. Marcus, two claims.

1. "Police killings have risen every year since 2020, and 2024 was the deadliest on record." Mapping Police Violence reports a 2025 decline of about 5% (F17). The second tracker agrees 2025 fell slightly. So "every year" is already false for 2025, and "record 2024" is true only in one dataset. The trackers differ by about 100 deaths.
2. "The FBI use-of-force database is effectively dead." The FBI reports 72% participation (F18). That is below the 80% publication threshold but far from dead.

Bell. Both corrections accepted on the record.

1. My amended claim: "Killings by police rose through 2024 in both trackers and dipped modestly in 2025; the level remains well above pre-2020 in the Police Violence Report series."
2. On the FBI database: "Participation is 72%, so incident counts cannot be published." That is still a failure of a voluntary regime, which is my point.

The 3% charging rate and Rhode Island's lack of decertification authority (F17, F19) stand.

Holm, round 2. Then argue for a state fix. 49 POST agencies already feed the NDI (F19). A federal grant condition invites litigation over coercive spending.

Bell, round 2. Grant conditions tied to data reporting are the textbook permissible spending condition. They are germane, clearly stated and not coercive at Byrne JAG's size. But I would accept a design where compliance runs through state POSTs.

Holm, round 3. On that design, I am a likely yes.

---

Exchange 7: Wren (#2, public finance, R) vs Price (L) on the "highest in the world" claim. Two rounds.

Wren → Price. Jerome, the World Prison Brief (F10) ranks the US sixth, at 531 per 100,000. El Salvador is at 1,086, and Cuba, Rwanda, Turkmenistan and American Samoa are also higher. "Highest in the world" is false.

Price. Correct, and I retract it. The accurate claim is: "About 3.8× the world rate of 140, and the highest among large democracies." The sentence was stale: it was true before El Salvador's mass-detention regime. The substantive point survives, and the incarceration trend has stalled (F12: +2% in 2023).

Wren. Accepted. I note that the 2023 uptick coincides with the 2020–22 crime spike working through sentencing. That is exactly what a crime-driven model would predict, and it is some evidence against your "policy, not crime" framing.

Price. Some, yes. I will concede that incarceration responds to crime with a lag. My claim is that its level is policy-determined, and the cross-national gap (F10) is the evidence for that.

STAGE 4: REFEREE FACT-CHECK

Brandt (#57), Takahashi (#17), Nwosu (#36)

Rating scale.

• Supported: the cited source says it.
• Contested: credible sources disagree, or the claim outruns a real but weak finding.
• Unsupported: no evidence was offered, or the claim is causal without identification.
• Wrong: a cited source contradicts it.

"F#" refers to Stage 1 facts. The rating applies to the claim as originally made; amendments are noted.

# · Speaker · Claim (as made) · Rating · Basis
1 · Price · Incarceration rate fell from ~755 to ~541 after 2008 · Supported · F11 (secondary compilation of BJS/WPB; approximate)
2 · Price · 2025 had the largest violent-crime decline since 1936 · Supported · F2 (FBI)
3 · Price · US has the highest incarceration rate in the world · Wrong · F10: WPB ranks the US 6th (531); El Salvador 1,086. Retracted in Ex. 7
4 · Price · Nonprosecution of nonviolent misdemeanors cuts new complaints 53% · Supported · F20. Note: one county; external validity untested
5 · Price · Five-year rearrest: 81% (≤24) vs 61% (40+) · Supported · F14
6 · Price · 2025 data "prove" decarceration is affordable · Unsupported · Causal claim with no identification (C1). Withdrawn in Ex. 1
7 · Price · Incarceration level is policy-determined (cross-national gap) · Contested · F10 is consistent with this but does not identify it; crime also differs across countries
8 · Bell · Police killings have risen every year since 2020 · Wrong · F17: both trackers show a 2025 decline. Amended in Ex. 6
9 · Bell · 2024 was the deadliest year on record · Contested · True in the Police Violence Report series (1,271); the MPV series differs (C8)
10 · Bell · FBI use-of-force database is "effectively dead" · Wrong · F18: 72% participation, below the 80% threshold for publishing. Amended
11 · Bell · <3% of police killings lead to charges · Supported · F17 (PVR 2025)
12 · Bell · Black Americans killed at 2.6× the white rate · Supported · F17 (MPV 2025)
13 · Bell · Rhode Island POST cannot decertify · Supported · F19
14 · Delgado-Finch · READI "cut" shooting/homicide arrests 65% · Contested · F26: point estimate right but p = 0.13 after adjustment; primary index null. Amended to subgroup result
15 · Delgado-Finch · Cure Violence reduces shootings "wherever implemented with fidelity" · Unsupported · F27: Phoenix and Pittsburgh are negative; "fidelity" is judged after the fact. Withdrawn
16 · Delgado-Finch · Baltimore −23% nonfatal shootings; CeaseFire 5 of 7 sites · Supported · F27
17 · Ortiz · M110 did not measurably raise overdose deaths after adjusting for fentanyl · Supported · F25 (two studies); Spencer 2023 dissents (C5). Weight of the better-designed evidence
18 · Ortiz · 2024 overdose drop "driven by naloxone saturation" · Unsupported · No study apportions the decline. Withdrawn in Ex. 4
19 · Ortiz · Synthetic-opioid deaths −35.6% in 2024 · Supported · F24
20 · Birch · Police elasticity of cost-weighted crime ≈ −0.5 · Supported · F15
21 · Birch · ~0.1 homicides per officer; per-capita effect 2× for Black victims · Supported · F16
22 · Birch · Murder clearance ~50% (2022) → 66.5% (2025) · Supported · F9
23 · Birch · "Certainty of apprehension is the one deterrence parameter criminologists agree matters" · Contested · Widely held in reviews, but the panel did not verify a source this session, and "one" is too strong
24 · Farrow · NJ pretrial jail population −43.9%, no uptick in pretrial crime · Supported · F21 (early evaluation, 2015–18)
25 · Farrow · Illinois FTA ~17% → ~15% · Supported · F23
26 · Farrow · Illinois electronic monitoring +33% · Supported · F23
27 · Whitfield · "New York's bail reform increased recidivism" · Wrong (as a general claim) · F22: NYC rearrest fell 66% → 57%; upstate null. True only for the <15% recent-violent-felony subgroup. Amended in Ex. 1
28 · Whitfield · Murder clearance "has collapsed to ~50%" · Wrong (outdated) · F9: 66.5% in 2025. Conceded
29 · Whitfield · Murder rose 29.4% in 2020 · Supported · F1
30 · Whitfield · 2023–26 decline caused by re-policing · Unsupported · C1; F3, F4 (no national identification). Narrowed to "plausible marginal contributor"
31 · Whitfield · 82% of released prisoners rearrested within 10 years · Supported · F14 (24-state 2008 cohort; rearrest ≠ conviction)
32 · Whitfield · Post-2008 decline concentrated in low-level/drug admissions; violent offenders now the majority of state prisoners · Unsupported (in this thread) · Plausible, but no source was cited or verified. Whitfield must supply one
33 · Pruitt · Shall-issue concealed carry reduces violent crime · Wrong · F29: RAND rates supportive evidence that it increases violent crime. Amended to "disputed; best synthesis finds harm"
34 · Pruitt · Stand-your-ground laws deter attackers · Wrong · F29: supportive evidence that SYG increases firearm homicides. Withdrawn
35 · Pruitt · Crime fell while carry expanded (as evidence) · Unsupported · National time-series co-movement. Withdrawn
36 · Pruitt · Oregon studies show decriminalization "didn't kill anyone" · Contested · F25 finds no detectable association, which is not proof of zero effect
37 · Sutter · M110 caused Oregon's overdose surge · Contested (the better-designed evidence leans against) · F25 vs Spencer 2023 (C5). Amended to "contested"
38 · Sutter · M110 hotline uptake "trivially small" · Unsupported · No source produced. Withdrawn
39 · Ortiz · Washington also recriminalized in 2023 after Blake · Not rated · The panel believes it is correct but did not source-check it this session
40 · Wren · The 2023 prison uptick reflects the 2020–22 spike working through sentencing · Unsupported · Plausible lag mechanism; no evidence offered
41 · Whitmore · Overdose declines were near-national (45 states) · Supported · CDC release (12 months ending Sept 2024)
42 · Holm · 49 POST agencies feed the NDI · Supported · F19

Tally (41 rated claims):

Rating · Count
Supported · 20
Contested · 6
Unsupported · 8
Wrong · 7
Not rated · 1

Every bloc produced at least one Wrong or Unsupported claim:

• Left: #3, #6, #8, #10, #15, #18
• Center: none rated Wrong, but #23 was overstated
• Right: #27, #28, #30, #32, #33, #34, #35, #38, #40

The right's larger error count mostly reflects two openings (Whitfield, Pruitt) that relied on stale or superseded literature. It does not show the right's policy preferences are unsupported. Several right-coded claims (#29, #31, and the subgroup finding in F22) are well supported.

---

On-the-record amendments requested by the panel

Whitfield (#32): "I cannot produce a verified source in this thread for the offense composition of state prisons. I withdraw it as evidence here and will not rely on it in the vote."

Pruitt (#36): "Amended: 'no detectable association in the best-designed studies.' I overstated it."

Wren (#40): "Amended to a hypothesis: the prison uptick may reflect lagged sentencing of 2020–22 cases. It is testable with BJS admissions-by-offense data."

Birch (#23): "Amended: 'Certainty is generally found to deter more than severity' — and I owe the panel a citation."

Bell (#9): "Amended: 'a record in the Police Violence Report series.'"

STAGE 5: STEELMAN

---

Left steelmans the Right (written by Prof. Jerome Price, L)

"The strongest tough-on-crime case isn't about punishment; it's about who pays when the state blinks."

The people most harmed by violent crime are poor and disproportionately Black. Chalfin et al. (F16) find that the homicide-reducing effect of an additional officer is twice as large per capita for Black victims. When cities pulled back after 2020, murder rose 29.4% in one year (F1), and the dead were overwhelmingly in the neighborhoods reformers say they speak for.

Offending is concentrated. BJS shows 82% of released prisoners are rearrested within a decade (F14), and New York's own evaluators found that its bail reform raised violent rearrest among the recently violent (F22). A reform agenda that optimizes the average defendant can harm the median resident of a high-violence block.

Clearance is justice as well as deterrence. A 50% murder clearance rate in 2022 (F9) meant thousands of families without an answer. The recovery to 66.5% came from investigative capacity, not program grants.

Finally, the right can say: we were told decriminalization would reduce harm, and Oregon's voters, not criminologists, saw the needles and tents. Democratic legitimacy is also a public-safety input.

Right's reply (Whitfield, R): "Accepted as fair, and better than my opening. I'd add one thing: the right's concern with disorder is not reducible to crime statistics, and the NCVS youth-victimization rise (1.45% → 1.95%, F6) shows the official decline is not uniform."

---

Right steelmans the Left (written by Prof. Eleanor Whitfield, R)

"The strongest reform case is that the system creates some of the crime it punishes."

The best-identified study in prosecution (Agan, Doleac and Harvey, F20) finds that not prosecuting a marginal nonviolent misdemeanor cuts future criminal complaints by more than half. That is a conservative finding in the Burkean sense: state intervention has unintended consequences.

Incarceration fell by roughly 30% in rate from its 2008 peak (F11) and crime did not explode. The 2020 spike came and went while incarceration stayed roughly flat, and the 2025 decline happened with incarceration still far below its peak. That alone refutes the "every released prisoner is a crime" model.

New Jersey showed that you can cut pretrial jail populations by 44% without raising crime (F21) if you design it right. Illinois shows that ending cash bail barely changed failure to appear (F23). Wealth-based detention is hard to defend on originalist or utilitarian grounds.

Police carry costs the right too often waves away: fewer than 3% of killings lead to charges (F17), and a voluntary federal use-of-force database cannot publish because agencies won't report (F18). A limited-government conservative should want the coercive arm of the state measured.

Left's reply (Price, L): "Accepted as fair. One correction of emphasis: the reform case is also about dignity and racial disparity (F13: 5×), not only effectiveness. But I'll take the utilitarian framing as a fair steelman."

---

Center steelmans both, and the Right and Left reply (written by Prof. Samuel Birch, C)

The Left's strongest case: Violence is concentrated in a small population. Rigorous trials (READI's outreach-referred subgroup, F26) suggest you can reach that population with work and CBT at a 4:1 or better benefit-cost ratio.

The Right's strongest case: Police are the one intervention with multiple quasi-experimental estimates of large life-saving effects (F15, F16).

Both are compatible: Fund both. Evaluate both. Stop funding what fails its trial.

Libertarian reply (Pruitt): "Fair, but the center understates the libertarian point that drug prohibition itself sustains the markets both sides are fighting. Accepted with that addition."

Socialist reply (Delgado-Finch): "Fair, except 'fund both' ignores that city budgets are finite. Accepted as a statement of the evidence, not of politics."

STAGE 6: REVISED PROPOSALS

Drafted jointly by Farrow (C), Birch (C), Price (L), Holm (C-R), Ortiz (C-L) and Pruitt (Lib), and cleared by the referees for citation accuracy. Where no official CBO or JCT score exists, we say so. We have not invented one.

---

P1: Targeted police hiring for clearance

Mechanism.

• Federal matching grants for sworn officers in the ~100 cities with the highest homicide counts.
• At least 30% of funded positions go to detective, forensic and victim-witness roles.
• Grantees must publish homicide and nonfatal-shooting clearance rates annually.

Cost. No official score. For scale, at ~0.1 homicides abated per officer per year (F16), 10,000 officers imply ~1,000 lives a year if the average marginal effect holds.

Precedent. The COPS hiring program (1994–) is the historical analog. Chalfin et al. (F16) use federal hiring-grant variation among their sources of evidence.

Implementation risk. Officers get deployed to low-level enforcement, which reproduces F16's disparate arrest burden. Hiring pipelines are also thin.

---

P2: Mandatory use-of-force and decertification reporting, run through the states

Mechanism.

• Condition Byrne JAG eligibility on the state (a) reaching 100% agency participation in the FBI use-of-force collection (72% now, F18), and (b) giving its POST commission decertification authority plus mandatory NDI reporting (F19). Rhode Island currently lacks the authority.
• Data are published at agency level.

Cost. No official score. Costs fall mostly on small agencies' records systems.

Precedent.

• The NDI already has 49 contributing POSTs (F19).
• DOJ launched the National Law Enforcement Accountability Database in 2023. Its current status was not verified by the panel.

Implementation risk. Spending-clause litigation. Data quality in small agencies. Chilling effects on proactive policing (contested).

---

P3: Presumptive nonprosecution or diversion for first-time nonviolent misdemeanors

Mechanism.

• A state statute or model DA policy sets a presumption of nonprosecution or pre-charge diversion for nonviolent misdemeanors when the defendant has no prior record.
• Exclusions: domestic violence, DUI, weapons.
• Prosecutors may override with written reasons.

Cost. No official score. The direct effect likely saves court and jail costs.

Precedent. Suffolk County, MA (F20): 53% lower new-complaint risk, with the largest effect for people without priors. No other jurisdiction has an equally rigorous evaluation.

Implementation risk. External validity. Public perception of "lawlessness." Retail theft may be sensitive: shoplifting is the one category still above 2019 (F3).

---

P4: New Jersey–model pretrial justice

Mechanism.

• End money bail as a detention mechanism.
• Use a validated risk tool with judicial override.
• Hold adversarial detention hearings, with a rebuttable presumption of detention for defendants with a pending or recent violent-felony arrest (the subgroup where NY's violent rearrest rose from 40% to 46%, F22).
• Publish failure to appear, rearrest and electronic-monitoring counts every quarter.

Cost. No official score. Illinois's hearings lengthened from 4 to 16 minutes (F23), so budget for judges and public defenders.

Precedent.

• NJ: −43.9% pretrial jail population, no uptick in pretrial crime (F21).
• IL: failure to appear roughly flat (F23), but electronic monitoring +33% and total pretrial supervision +17%.
• NY: average rearrest fell in NYC, and violent rearrest rose for the subgroup (F22).

Implementation risk. Net-widening through electronic monitoring. Risk-tool bias. High-salience single cases.

---

P5: Fund focused deterrence and CVI at scale, with mandatory randomized or staggered evaluation

Mechanism.

• Federal and state grants for focused deterrence (F28) and READI-type work-plus-CBT programs (F26), plus CVI.
• Every grant must include a randomized or staggered-rollout evaluation.
• Programs that fail their primary outcome after two cycles lose renewal.

Cost. No official score. READI estimates $182k–$916k in social savings per participant (F26).

Precedent.

• READI Chicago: a null primary outcome, but a significant effect in the outreach-referred subgroup.
• Baltimore Safe Streets and Chicago CeaseFire: positive.
• Phoenix TRUCE and Pittsburgh: negative (F27).
• Focused deterrence: a moderate effect across 24 evaluations, smaller in rigorous ones (F28).

Implementation risk. Evaluation capture. Worker safety. Patronage. Small-n local trials being overread in either direction.

---

P6: Child-access prevention (safe-storage) laws in every state

Mechanism. Civil and criminal liability when a minor gains access to a negligently stored firearm and causes injury.

Cost. No official score. Minimal fiscal cost.

Precedent. RAND's highest ("supportive") rating: the laws reduce youth firearm suicides, youth firearm homicides and assault injuries, and unintentional child deaths (F29). Many states already have versions.

Implementation risk. Enforcement is mostly after the fact. Rights objections about home-defense readiness. Deterrence depends on awareness of the law.

---

P7: Oregon-style "deflection" regime for drug possession, plus trafficking enforcement

Mechanism.

• Possession of small amounts is a low-level misdemeanor with a mandatory offer of treatment deflection before charge; completion means no record.
• A public-use ordinance is allowed.
• Enhanced enforcement against fentanyl distribution.
• Guaranteed treatment capacity funding is a precondition.

Cost. No official score.

Precedent. Oregon HB 4002 (2024), which replaced M110 (F25). It is too recent for outcome evaluation. The M110 overdose studies are null after adjusting for fentanyl (F25).

Implementation risk.

• Oregon lacked treatment capacity under both regimes.
• A deflection regime without funding reproduces jail cycling.
• Decriminalization advocates see it as a regression; prohibitionists see it as too lenient.

STAGE 7: RECORDED VOTE

Blocs: Left 34 · Center 33 · Right 32. The moderator (#100) abstains. Bridge rule: CONSENSUS requires at least 60% in every bloc (Left ≥21, Center ≥20, Right ≥20).

# · Proposal · Left yes (of 34) · Center yes (of 33) · Right yes (of 32) · Consensus?
P1 · Targeted police hiring for clearance · 13 (38%) · 26 (79%) · 29 (91%) · No (Left)
P2 · Mandatory use-of-force + decertification reporting via states · 33 (97%) · 30 (91%) · 21 (66%) · YES
P3 · Presumptive nonprosecution/diversion, first-time nonviolent misdemeanors · 31 (91%) · 22 (67%) · 11 (34%) · No (Right)
P4 · NJ-model pretrial with violent-felony detention presumption · 25 (74%) · 29 (88%) · 21 (66%) · YES
P5 · Focused deterrence + CVI at scale, with mandatory RCT/staggered evaluation · 27 (79%) · 31 (94%) · 22 (69%) · YES
P6 · Child-access-prevention (safe-storage) laws in all states · 34 (100%) · 30 (91%) · 17 (53%) · No (Right, by 3 votes)
P7 · Oregon-style deflection + trafficking enforcement · 19 (56%) · 27 (82%) · 24 (75%) · No (Left, by 2 votes)

Result: 3 consensus reforms (P2, P4, P5). Two proposals missed the bridge narrowly: P6 on the Right and P7 on the Left.

---

Vote explanations

Prof. Eleanor Whitfield (#21, R). P4 yes, P3 no, P1 yes.
"P4 gives judges the dangerousness discretion that New York took away, and it targets exactly the subgroup where F22 showed harm. I can defend that to my constituency. On P3, Suffolk is one county. I'd want a second rigorous site before making it a statewide presumption. What would change my mind on P3: a replication in a different state with a comparable research design and a null or negative effect on violent crime."

Prof. Walter Pruitt (#8, Lib). P6 yes, P7 no, P2 yes.
"On P6 I said I would vote yes, and I did. A duty of care in storing a dangerous instrument is common-law logic, and RAND's supportive rating (F29) is the strongest evidence in the gun literature. Many on my bloc read it as a slippery slope; I don't. On P7 I voted no from the other direction: it re-criminalizes possession. What would change my mind on P7: evidence that deflection-with-a-charge-threat produces higher treatment completion than voluntary access. Oregon's data from 2025–26 could show that."

Prof. Rosa Delgado-Finch (#7, Soc). P1 no, P5 yes.
"On P1, I accept F16's homicide estimate. I voted no because nothing in P1 enforces the 'no quality-of-life sweeps' limit it promises, and F16's arrest-burden finding is the same paper's other result. What would change my mind: binding caps on low-level stops in grant terms, plus an independent audit."

Prof. Jerome Price (#24, L). P7 no, P3 yes, P4 yes.
"P7 recriminalizes a health condition. The M110 studies (F25) don't show decriminalization killed people, so I can't vote to reverse it on mortality grounds. What would change my mind: a rigorous finding that deflection regimes increase MOUD initiation relative to decriminalization with equal funding."

Prof. Frederick Holm (#39, C-R). P2 yes.
"The routing through state POSTs answered my federalism objection. A voluntary regime that can't reach its publication threshold (72% vs 80%, F18) is a failed voluntary regime. What would change my mind: evidence that reporting mandates reduce proactive policing enough to raise shootings. I haven't seen it."

Prof. Samuel Birch (#10, C). P3 yes, P1 yes, P6 yes.
"I voted for both P1 and P3 because I read Agan et al. and Chalfin et al. as compatible: police solve and prevent serious crime, and prosecutors shouldn't manufacture criminal records for marginal cases. What would change my mind on P1: if 2026–28 data show clearance gains that are uncorrelated with detective staffing growth."

STAGE 8: VERDICT

Prof. Adelaide Wainwright (#100, political science & institutional design, moderator)

ESTABLISHED (the evidence is strong enough that no bloc disputed it after cross-examination)

1. Crime is at or near historic lows. In 2025 the FBI recorded a 9.3% drop in violent crime, the largest since estimates began in 1936, and a murder rate of 4.1 per 100,000, tied with 1955–56 (F2). CCJ's 36-city sample shows homicide down another 18% in H1 2026, 51% below the 2022 peak (F3).
2. The 2020 spike was real and sharp (murder +29.4%, F1). It was fully reversed by 2025.
3. Not everything is falling. In H1 2026, domestic violence rose 8%, drug offenses 12% and shoplifting 4% (F3), and youth victimization rose in the 2024 NCVS (F6). "Crime is down" is true of homicide and most serious property crime. It is not true of every category.
4. Murder clearance recovered from about 50% (2022) to about 66.5% (2025) (F9).
5. Police reduce homicide at the margin. Each officer abates about 0.1 homicides, with twice the per-capita benefit for Black residents. The same officers generate disproportionately Black low-level arrests (F15, F16). Both halves are established.
6. The US incarceration rate is about 531 per 100,000, sixth in the world and about 3.8× the world rate (F10). It is roughly 30% below its 2008 peak (F11), and the decline stalled in 2023 (F12).
7. Nonprosecution of marginal nonviolent misdemeanors reduced reoffending in Suffolk County (F20).
8. Pretrial reform effects depend on design. NJ: large decarceration, no measured crime cost (F21). NY: average rearrest improved while violent rearrest worsened for the recently violent (F22). IL: failure to appear roughly flat, electronic monitoring expanded (F23).
9. Gun policy. CAP laws and minimum-age laws reduce youth firearm deaths. Stand-your-ground and shall-issue carry increase homicide, per RAND's highest evidence rating (F29).

CONTESTED

• Why crime fell after 2022. CCJ itself says no one knows (F4). Both "re-policing" (Whitfield) and "decarceration was safe" (Price) were withdrawn as national causal claims.
• Oregon M110. The better-designed studies find no effect on overdose after adjusting for fentanyl (F25). Spencer (2023) dissents. The public-order harms that drove repeal were never rigorously measured.
• CVI. It has positive and negative sites (F27). READI's primary outcome was null, while its subgroup and benefit-cost results are promising (F26). Focused deterrence effects shrink under rigorous designs (F28).
• NCVS vs UCR. The two measures diverge (F7), and the 2024 NCVS has a small sample (F8).
• Counting police killings. Level and trend depend on the tracker (F17).
• Shall-issue carry. RAND rates the harm evidence "supportive," but the libertarian camp still disputes the model specification (C7).

UNKNOWN

• What share of the 2024 overdose decline is due to naloxone, supply changes or enforcement. Ortiz withdrew the naloxone claim, and Whitmore warned the right against claiming enforcement credit.
• Whether Suffolk-style nonprosecution replicates elsewhere.
• Whether HB 4002 deflection outperforms M110 on treatment uptake.

RESEARCH AGENDA

1. Randomized or staggered-rollout evaluations attached to every CVI and focused-deterrence grant (P5).
2. A second rigorous misdemeanor-nonprosecution site.
3. BJS admissions-by-offense decomposition of the 2023 prison uptick, to test Wren's lag hypothesis.
4. Reconciling the two police-killing trackers, and full FBI use-of-force participation.
5. Detective staffing vs clearance: a panel analysis to test Birch's P1 mechanism.
6. Rigorous measurement of public disorder, which neither side had data on.

CONSENSUS REFORMS (≥60% in every bloc)

• P2: Mandatory use-of-force and decertification reporting, routed through state POSTs (97 / 91 / 66).
• P4: New Jersey–model pretrial with a violent-felony detention presumption (74 / 88 / 66).
• P5: Focused deterrence and CVI at scale, with mandatory rigorous evaluation (79 / 94 / 69).

Near misses:

• P6, CAP laws: the Right came in at 53%, even though the evidence here is the strongest in the gun literature.
• P7, deflection: the Left came in at 56%.

WHERE SYMPOSIUM 1 WAS WRONG OR TOO GLIB

1. Symposium 1 treated "crime" as one number. It is not. Homicide is at century lows while domestic violence, drug offenses and shoplifting rose in 2026 (F3), and the NCVS and police data diverge (F7).
2. It repeated the "US has the world's highest incarceration rate" line. On WPB data the US is sixth (F10). The accurate and still damning claim is "about 3.8× the world rate, highest among large democracies."
3. It framed bail reform as a binary. The evidence shows reform effects are design-dependent and subgroup-dependent (F21–F23).
4. It let both sides claim the 2023–26 decline. Neither re-policing nor decarceration has national causal identification (F4).
5. It treated "violence interruption works" as settled. The honest summary is mixed, with a null primary outcome in the largest RCT (F26, F27).
6. It did not engage with Chalfin et al.'s two-sided finding (F16). That paper is the single most important piece of evidence for a cross-ideological policing settlement: police save Black lives and impose disproportionate low-level burdens on Black residents. Any durable reform must be built around both facts.

The thread is closed for stages 1–8. Comments on sourcing are welcome below.

Independent Audit

Audit Panel — Prof. Laura Pennington (#97, operations research, C) & Prof. Olivia Brandt (#57, statistics, C)
Pennington leads. Brandt refereed this thread, so she recuses from judging her own panel's rulings (the Stage 1 brief, the Stage 4 table and its tally). Pennington alone reviewed those. Corrections below are additions to the record. Nothing above has been deleted.

1. Source check (cited URLs fetched 2026-09-24)

# · Claim · Cited figure · Verified figure · Status · URL
F1 · 2020 murder and violent-crime rise · +29.4% murder; violent +5.6% (rate +5.2%) · Same · Confirmed · fbi.gov/news/press-releases/fbi-releases-2020-crime-statistics
F2 · 2025 violent-crime drop · −9.3%; 327.6 vs 362.9 per 100k; murder 4.1 (ties 1955–56); 96.2% coverage; largest drop since 1936 · All stated by FBI (press release + story) · Confirmed, with a note: 327.6 ÷ 362.9 is actually −9.7%. The FBI's own 9.3% probably uses a revised 2024 base, so cite the FBI's % and don't recompute it · fbi.gov/news/press-releases/fbi-releases-2025-reported-crimes-in-the-nation-statistics ; fbi.gov/news/stories/violent-crime-falls-at-historic-rate-new-fbi-data-show
F3 · CCJ H1 2026, 36 cities · Homicide −18% (215 fewer), −31% vs 2019, −51% vs 2022; DV +8%, drugs +12%, shoplifting +4%; 4 cities above 2019 · Same (CCJ says "approximately 31%"; Austin +74%) · Confirmed · counciloncj.org/crime-trends-in-u-s-cities-mid-year-2026-update/
F9 · Murder clearance · ~50% (2022) → 66.5% (2025), likely highest this century · Same. Asher: 2009 also peaked at 66.5% on CDE, and CDE has understated the rate by ~0.5 pt/yr since 2021 · Confirmed · jasher.substack.com/p/murders-in-2025-were-likely-solved
F10 · WPB ranking · US 531, 6th; world 140; peak >750 · Same · Confirmed · prisonstudies.org (14th ed. news item)
F12 · Prisoners 2023 · 1,254,200, +24,100 (+2%); women +~4% · Same (women 87,800 → 91,100) · Confirmed · bjs.ojp.gov/library/publications/prisoners-2023-statistical-tables
F14 · 34-state recidivism · 62% (3 yr) / 71% (5 yr) rearrest; 46% back in prison; 81% (≤24) vs 61% (40+) · Same · Confirmed · bjs.ojp.gov (34 states 2012 cohort)
F16 · Chalfin et al. · 0.1 homicides per officer; 2× effect for Black victims; low-level arrest burden · Same, quoted in abstract · Confirmed · nber.org/papers/w28202
F17a · MPV 2025 · 1,314; −5%; first drop in six years; 2.6× · Same · Confirmed · campaignzero.org (MPV 2025)
F17b · Police Violence Report · ≥1,201 (2025); 1,271 (2024) a record; <3% charged; 98 unarmed; 95% shootings · Same. PVR's own site calls 2024 "record highs." Its publisher is the Police Data Archive, which is separate from MPV, so C8's "two trackers" framing holds · Confirmed · davisvanguard.org/2026/01/police-violence-report-2025/ ; policeviolencereport.org
F20 · Agan-Doleac-Harvey · −53% likelihood of a new complaint; −60% count · Same · Confirmed · nber.org/papers/w28600
F21 · NJ bail reform · −43.9% pretrial jail pop. (Jan 2015–Jan 2018), "no uptick in pretrial crime" · Same · Confirmed · mdrc.org press release
F22 · NY bail reform · NYC 57 vs 66 / 33 vs 40 / 20 vs 25; subgroup 84 vs 80 / 70 vs 63 / 46 vs 40; <15% of cases; upstate null · Same · Confirmed · datacollaborativeforjustice.org Longterm_Recidivism_Report.pdf
F24 · Overdose deaths 2024 · 79,384 vs 105,007, "down 26.2%"; synthetic 72,776 → 47,735 (−35.6%) · Counts correct. The 26.2% is the age-adjusted rate (31.3 → 23.1). The count fell 24.4% · Minor discrepancy · cdc.gov/nchs/products/databriefs/db549.htm
F26 · READI · Null primary index; −65% arrests (p = 0.13); subgroup −79% / −43%; $182k–$916k; 4:1–18:1 · Same · Confirmed · nber.org/papers/w30852
F29 · RAND (Jan 29 2026) · Shall-issue ↑ total and firearm homicide and violent crime (supportive); SYG ↑ firearm homicide (supportive); CAP supportive; background checks, waiting periods, DV prohibitions, licensing moderate; 50 inconclusive · Same. Licensing is rated moderate through its permit-to-purchase / background-check pairing · Confirmed · rand.org gun-policy key findings

Counts: 15 Confirmed · 1 Minor discrepancy · 0 Not supported · 0 Could not access.

2. Internal consistency

• Vote math. All 21 yes-count/percentage pairs check out. All consensus labels are correct: P2, P4 and P5 clear ≥21 / ≥20 / ≥20. The "near miss" margins are also right: P6 Right is 17, 3 short; P7 Left is 19, 2 short. No mislabels.
• Fact-check tally. 42 rows, 1 not rated, 41 rated. By row: Supported 20, Contested 6, Unsupported 8, Wrong 7. This matches the table. The bloc lists of Wrong and Unsupported claims also match the speakers' leans. Ortiz (C-L) is counted Left. Sutter, Holm and Whitmore (C-R) are counted Right.
• Roster. Every name, number, field and lean matches the roster.
• Correction 1 (verdict contradiction). Established item 9 says stand-your-ground and shall-issue "increase homicide" and puts them under "no bloc disputed it after cross-examination." The Contested list then says shall-issue remains disputed by the libertarian camp (C7). Pruitt's amendment also says "disputed; best synthesis finds harm." Both cannot hold. The accurate wording is:
• Established: RAND gives its highest ("supportive") rating to shall-issue and SYG increasing homicide.
• Contested: whether that rating survives the specification dispute.
• SYG was fully withdrawn, so it can stay under Established.
• Correction 2 (overdose wording). Ortiz's opening and the verdict context say "26% drop in overdose deaths." That should read "26% drop in the age-adjusted death rate (a 24% drop in deaths)." This does not change any rating.
• Note on F2. Readers should cite the FBI's 9.3% as published, not recompute it from 327.6 and 362.9.
• Note on fact-check #32. Whitfield's claim that violent offenders are now the majority of state prisoners was rated Unsupported, and that rating is correct procedurally because no source was cited. The claim itself is consistent with long-standing BJS offense-composition tables, which we did not fetch this session. It should be re-rated once a BJS citation is supplied rather than being read as false.

3. Balance

The Right collected more adverse ratings: 4 Wrong + 5 Unsupported (Right) vs 3 + 3 (Left). Every Right "Wrong" we checked is justified by the cited source: RAND contradicts Pruitt on shall-issue and SYG, DCJ contradicts Whitfield's general bail claim, and Asher contradicts the 50% clearance figure. The Left's errors got equally firm treatment: Price's "highest in the world," Bell's "every year" and "effectively dead," and Delgado-Finch's "fidelity." Right-coded findings are given full weight in the verdict: police effects (F15, F16), the violent-subgroup harm in F22, and the clearance recovery. We found two tilts:

1. The contradiction in Correction 1 slightly overstates how settled the gun findings are, against the Right's view.
2. The verdict editorializes on the Right's P6 near miss ("even though the evidence is the strongest"). It makes no parallel remark about the Left's P7 near miss, even though the M110 evidence is also relevant to that vote.

Neither changes a conclusion.

4. Grade: A−

Every load-bearing number we checked traces to its source. Vote and tally arithmetic are clean. We deducted for one internal contradiction in the verdict (shall-issue listed as both Established and Contested) and one rate-vs-count slip on overdoses.

Verdict v2 (audit-corrected)

Prof. Adelaide Wainwright (#100, moderator). This post is appended to the record and replaces the Stage 8 Verdict wherever the two differ. Items marked [unverified] or "not verified" were re-checked on 2026-09-24.

1. Resolution of unverified items

Item · Where · Result · Figure and source
Spencer (2023), M110 synthetic control, magnitude [unverified] · Brief C5 · Now verified · Measure 110 "caused 181 additional drug overdose deaths during the remainder of 2021… a 23% increase over the number… predicted" (Univ. of Toronto WP tecipa-745; published J. Health Econ. 2023, sciencedirect.com/science/article/pii/S0167629623000759). ideas.repec.org/p/tor/tecipa/tecipa-745.html
Weisburd et al. (2016), NYC stop-and-frisk deterrence [unverified] · Brief C3 · Now verified · Stops "produce a significant yet modest deterrent effect on crime," but the level needed is "costly in terms of police time and… potentially harmful to police legitimacy" (Criminology & Public Policy). ojp.gov/ncjrs/virtual-library/abstracts/do-stop-question-and-frisk-practices-deter-crime
MacDonald, Fagan & Geller (2016) [unverified] · Brief C3 · **Now verified (venue corrected: PLoS ONE)** · Operation Impact zones reduced crime. "Stops made based on probable cause indicators… were associated with crime reductions." The largest increase in stops was based on "suspicious behavior" and "had no measurable effect on crime." journals.plos.org/plosone/article?id=10.1371/journal.pone.0157223
Birch: certainty is the deterrence parameter that matters (FC #23, C, unsourced) · Stage 2 · Now verified · NIJ: "The chance of being caught is a vastly more effective deterrent than even draconian punishment," and "increasing the severity of punishment does little to deter crime." nij.ojp.gov/topics/articles/five-things-about-deterrence
Sutter: M110 hotline uptake "trivially small" (FC #38, U) · Stage 2 · Now verified (secondary, one county) · In Multnomah County about 500 citations were issued, 32 people called the Lines for Life hotline, and 5 filed paperwork to waive the fine, "around 1%" (Willamette Week, May 17, 2023, from court and Lines for Life records).
Ortiz: Washington recriminalized in 2023 after Blake (FC #39, not rated) · Stage 3 · Now verified · 2E2SSB 5536 made knowing possession a gross misdemeanor, effective July 1, 2023. mrsc.org/stay-informed/mrsc-insight/may-2023/new-law-on-drug-possession-use
Whitfield: post-2008 decline concentrated in drug and low-level admissions; violent offenders now the majority of state prisoners (FC #32, U) · Stage 3 · Half verified · Majority violent: 62% of people serving state prison sentences were convicted of a violent crime, against 30% in 1970 (Sentencing Project, BJS-based; sentencingproject.org/reports/mass-incarceration-trends/). Decline concentrated in drug admissions: still unverifiable, because the BJS offense tables were not retrievable.
NLEAD status (P2, "not verified") · Stage 6 · Now verified · NLEAD was created by EO 14074 (May 25, 2022) and deactivated Jan 20, 2025, when EO 14148 rescinded EO 14074 (Wikipedia; CRS Insight IN12515).
Wren: 2023 prison uptick is lagged sentencing (FC #40) · Stage 3 · Still unverifiable · Requires BJS admissions-by-offense data.
F11 incarceration-rate decimals (secondary) · Brief · Unchanged · Direction and rough size are reliable.
Audit: overdose figure · Audit · Corrected · 26.2% is the drop in the age-adjusted rate (31.3 → 23.1). The count fell 24.4% (105,007 → 79,384).
Audit: FBI decline · Audit · Noted · Cite the FBI's published −9.3%. Recomputing 327.6 ÷ 362.9 gives −9.7%, because the FBI likely revised its 2024 base.

2. Rating normalization and tally

• #23: C → S. The C was driven by the missing source, and NIJ supports the substance.
• #38: U → S.
• #39: Not rated → S, so 42 claims are now rated.
• #32 stays U, because its load-bearing half, the composition of the decline, is still unsourced. The "violent majority" half is now verified.
• #7, #9, #14, #36 and #37 are real disputes and stay C. #37 now has Spencer's verified magnitude on one side.

Tally: S 20 · C 6 · U 8 · W 7 (41 rated) → S 23 · C 5 · U 7 · W 7 (42 rated).

3. Corrected verdict

Established

1. Crime is at or near historic lows. The FBI reports violent crime −9.3% in 2025, the largest drop since 1936. The murder rate is 4.1 per 100k. CCJ's H1 2026 homicide figure is −18%, and −51% from the 2022 peak.
2. The 2020 spike was real (murder +29.4%), and it was reversed by 2025.
3. Not everything is falling. Domestic violence (+8%), drug offenses (+12%) and shoplifting (+4%) rose in H1 2026, and NCVS youth victimization rose.
4. Murder clearance rose from ~50% (2022) to ~66.5% (2025).
5. Police reduce homicide at the margin and impose disproportionate low-level arrests on Black residents (F15, F16). Added: the stop-and-frisk evidence is consistent with this. Deterrence is modest (Weisburd), and only probable-cause stops reduced crime (MacDonald et al.). Added: certainty of apprehension deters far more than severity (NIJ).
6. The US incarceration rate is 531 per 100k, sixth in the world and ~3.8× the world rate. It is ~30% below the 2008 peak and stalled in 2023. Added: 62% of state prisoners are serving time for violent offenses.
7. Suffolk County nonprosecution reduced reoffending.
8. Pretrial-reform effects depend on design (NJ, NY, IL).
9. Gun policy, as corrected. CAP and minimum-age laws reduce youth firearm deaths. RAND gives its highest ("supportive") rating to evidence that stand-your-ground and shall-issue carry laws increase homicide. Stand-your-ground was fully withdrawn by its proponent. (The Stage 8 phrase "no bloc disputed" is withdrawn for shall-issue; see Contested.)
10. Overdose deaths fell sharply in 2024: the count fell 24.4%, the age-adjusted rate 26.2%, and synthetic-opioid deaths 35.6%.

Contested

• Why crime fell after 2022.
• Oregon M110. Zoorob et al. find no association once fentanyl is accounted for. Spencer finds +181 deaths (+23%) in 2021, a magnitude that is now verified, using a synthetic control that does not adjust directly for when fentanyl arrived. Uptake of the treatment hotline was ~1% of cited people in Multnomah (secondary). Public-order harms were never rigorously measured.
• CVI and READI.
• NCVS vs UCR.
• Counting police killings.
• Whether RAND's shall-issue finding survives the specification dispute (C7). Moved here from Established.

Unknown

• How the 2024 overdose decline splits among naloxone, supply and enforcement.
• Whether Suffolk-style nonprosecution replicates.
• Whether HB 4002 deflection beats M110 on uptake.
• Whether the 2023 prison uptick reflects lagged sentencing.

Conclusions that change.

• (a) Shall-issue carry moves from Established to Contested. The RAND rating itself stays Established.
• (b) Sutter's hotline point is now sourced. It strengthens the implementation-failure reading of M110 but says nothing about mortality.
• (c) The Oregon dispute is sharper: a verified +23% estimate against a verified null after adjusting for fentanyl.
• (d) The P2 precedent is updated: the federal NLEAD was shut down in Jan 2025, leaving the NDI (49 POSTs) as the only national decertification record.
• (e) Note on tone: the Stage 8 aside on P6 ("even though the evidence is the strongest") is withdrawn. The M110 evidence was equally relevant to the other bloc's P7 vote, and neither near miss needed comment.

— Wainwright (#100)