05 — Elections, Polarization & the Information Ecosystem

STAGE 1 — EVIDENCE BRIEF

Referee panel: Prof. Olivia Brandt (#57, statistics, C) · Prof. June Takahashi (#17, econometrics, C) · domain specialist Prof. Catherine Doyle (#25, election law, C)

Ground rules from the panel: every number below was checked against the linked source on or before 2026-09-24. Where a primary source could not be retrieved we say so. Where we rely on a secondary compilation (e.g., Wikipedia's running redistricting table) we flag it, because those numbers move.

---

A. Affective polarization — how big, how unusual

1. ANES out-party feeling thermometer (0–100): 48 in 1978 → 31 in 2016 → 26 in 2024 (in-person/phone modes). Self-completed (web) modes run ~6–8 points lower in every year, so mode matters for trend comparisons. In-party ratings have been "pretty stable" since the mid-1970s — the growth is in out-party dislike, not in-party love. Source: Good Authority summary of ANES, 2025 — https://goodauthority.org/news/americans-views-of-the-opposite-party-have-declined-even-further/ (data 1978–2024).
2. Cross-country: Boxell, Gentzkow & Shapiro studied 12 OECD countries over four decades. The US had the largest increase in affective polarization; five countries had smaller increases and six had decreases. Source: Review of Economics and Statistics / NBER w26669 — https://www.nber.org/papers/w26669 (published 2022–24; data ~1980–2020).
3. Who polarized fastest: growth in a nine-measure polarization index was largest among adults 75+ vs. 18–39 — the groups least likely to use the internet/social media. Boxell, Gentzkow & Shapiro, NBER w23258 (2017) — https://www.nber.org/papers/w23258.
4. Moralized hostility (Pew, 2022): 62% of Republicans and 54% of Democrats view the other party very unfavorably (under 25% in both parties in 1994). 72% of Republicans / 63% of Democrats call the other party "more immoral than other Americans," up from 47% / 35% in 2016. https://www.pewresearch.org/politics/2022/08/09/as-partisan-hostility-grows-signs-of-frustration-with-the-two-party-system/ (2022).
5. Perception gaps (More in Common, 2019, n=2,100, YouGov): partisans "imagine almost twice as many of their political opponents as reality hold views they consider 'extreme'." Those who read news "most of the time" had gaps nearly three times those who read "only now and then"; people who post political content averaged a gap of 29 vs. 18 for non-posters; the politically disengaged were ~3× more accurate than the "wings." Among Democrats, gaps grew with education. https://perceptiongap.us/ (2019).

B. Primaries, competitiveness, districts

1. Swing seats have halved: Cook Political Report counts 164 swing seats (PVI R+5 to D+5) in 1999 vs. 82 in 2023. Of the 77 net swing seats lost since 1997, 58% (net −45) came from areas trending redder/bluer ("sorting/realignment") and 42% (net −32) from boundary changes. The median House Democrat's district moved D+8 → D+13 and the median Republican's R+7 → R+13 (1997→2023). https://www.cookpolitical.com/cook-pvi/realignment-more-redistricting-has-decimated-swing-house-seats (2023).
2. 2025 Cook PVI: a slight decline in geographic polarization after 2024 (Trump gains in heavily minority districts), but "voters are no more likely to split their tickets than they used to be." https://www.cookpolitical.com/cook-pvi/introducing-2025-cook-partisan-voting-index-slightly-depolarizing-election (2025; detailed counts paywalled — panel could not verify the 2025 swing-seat total).
3. Gerrymandering vs. roll-call polarization: McCarty, Poole & Rosenthal (AJPS 2009) found "very little evidence" that districting drives congressional polarization; simulated neutral maps produce comparable polarization; polarization is "primarily a function of the differences in how Democrats and Republicans represent the same districts." https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1154054.
4. Extremist nominees lose: Hall (APSR 2015), RD design on House primaries 1980–2010: when an extremist narrowly wins the primary, the party's general-election vote share falls ~9–13 points and its win probability falls 35–54 points. https://stanforddpl.org/papers/hall_extremists_2015/.
5. Primaries in the long run: Hirano & Snyder, Primary Elections in the United States (Cambridge, 2019), ~a century of data: primaries "inject electoral competition" into one-party areas and help select relatively qualified candidates; their link to modern polarization is real but not determinative. https://www.cambridge.org/core/books/primary-elections-in-the-united-states/C15A9969FFE0323F1751A79B917D8B81.
6. Primary turnout and openness: Ferrer (2026 working paper; 2.6 billion voter-file observations, DiD across AK, CO, ID, ME, OK, WA): primary electorates average ~21% of eligible voters vs. 53% in generals; opening primaries to unaffiliated voters raises turnout ~5 points; switching to nonpartisan primaries raises it ~11 points; electorates become younger and more unaffiliated but remain unrepresentative. https://www.joshuaferrer.com/publication/partisan_primaries/partisan_primaries.pdf (2026).
7. California's reforms and polarization: Kousser, Phillips & Shor (2016) found neither the Citizens Redistricting Commission nor the top-two primary "halted the continuing partisan polarization of California's elected lawmakers"; Ahler, Citrin & Lenz (2015) found too little crossover voting to reliably help moderates. (Via literature summary: https://en.wikipedia.org/wiki/Nonpartisan_blanket_primary — panel flags this as a secondary summary.)
8. California commission design: 14 members — 5 from each major party, 4 unaffiliated; Prop 11 (2008) for legislative maps, Prop 20 (2010) for Congress. In the prior (legislature-drawn) decade, only 1 of 255 California congressional elections changed party hands. Prop 50 (Nov 2025) replaced the commission's congressional map with a legislature-drawn map through the 2030 elections. https://en.wikipedia.org/wiki/California_Citizens_Redistricting_Commission.

C. Redistricting law and the 2025–26 map war

1. Rucho v. Common Cause (June 27, 2019, 5–4): partisan gerrymandering claims "present political questions beyond the reach of the federal courts." https://en.wikipedia.org/wiki/Rucho_v._Common_Cause.
2. Efficiency gap (Stephanopoulos & McGhee): wasted-vote difference ÷ total votes; proposed presumptive threshold ±7%. A district court found Wisconsin's 2011 map at 11.69–13% pro-GOP; SCOTUS dismissed Gill v. Whitford on standing. Critics (e.g., Bernstein) note the metric penalizes proportional outcomes in some cases. https://en.wikipedia.org/wiki/Efficiency_gap.
3. Louisiana v. Callais (Apr 29, 2026, 6–3, Alito, J.): Louisiana's second majority-Black district was an unconstitutional racial gerrymander; §2 vote-dilution claims now require substantially more (the opinion requires plaintiffs to show the state could meet its goals with alternative maps; the dissent says §2 becomes "all but a dead letter"). https://en.wikipedia.org/wiki/Louisiana_v._Callais ; opinion: https://www.supremecourt.gov/opinions/25pdf/24-109_21o3.pdf.
4. Mid-decade redistricting 2025–26 (secondary compilation, as of Sept 2026 — numbers are projected seat shifts, not results): Texas +5 R (Aug 2025), Missouri +1 R, North Carolina +1 R, Ohio +2 R, California +5 D (Prop 50, 64.42% yes), Utah +1 D (court-ordered), Florida +4 R, Tennessee +1 R, Louisiana +1 R, Alabama +1 R; Virginia's +4 D referendum (51.61% yes, Apr 2026) was voided by the Supreme Court of Virginia on May 8, 2026. Compiled totals: R +14, D +10, net R +4; ~14 competitive seats eliminated. https://en.wikipedia.org/wiki/2025%E2%80%932026_United_States_redistricting ; Virginia ruling: https://virginiamercury.com/2026/05/08/supreme-court-of-virginia-strikes-down-redistricting-amendment-keeps-current-maps-in-place/.

D. Alternative systems: Alaska, RCV, NYC

1. Alaska 2024 repeal vote: Ballot Measure 2 (repeal top-four + RCV) failed 160,230 yes (49.88%) vs. 160,973 no (50.12%), a 743-vote margin after recount. https://en.wikipedia.org/wiki/2024_Alaska_Ballot_Measure_2 ; https://alaskapublic.org/elections/2024-11-20/alaskas-ranked-choice-repeal-measure-fails-by-664-votes (pre-recount margin 664).
2. Alaska 2026: a new Ballot Measure 2 on the November 2026 ballot would repeal RCV, restore closed partisan primaries, and repeal the 2020 "true source" disclosure rules for donors over $2,000; APOC estimates $2.6M implementation cost; sponsors gathered 34,000+ signatures. https://alaskapublic.org/news/politics/elections/2026-07-29/alaskans-to-vote-on-ballot-measure-that-would-overhaul-state-election-system.
3. Alaska 2022 special (the center-squeeze case): Peltola beat Palin 51.48–48.52 in the final round; Begich was eliminated first although 53% of voters ranked him above Peltola (and a majority preferred him to Palin) — a Condorcet failure; ~11,243 Begich ballots ranked neither finalist, more than the final margin. https://en.wikipedia.org/wiki/2022_Alaska%27s_at-large_congressional_district_special_election.
4. Alaska legislature after top-four: the 2025 House majority is a 21-member bipartisan coalition (14 D, 5 independents, 2 R), with 19 Republicans in the minority. https://en.wikipedia.org/wiki/2024_Alaska_House_of_Representatives_election.
5. NYC 2021 (first RCV mayoral primary): Adams 404,513 (50.4%) vs. Garcia 397,316 (49.6%) in the final round; the Board of Elections erroneously added ~135,000 test ballots to a preliminary tabulation. https://en.wikipedia.org/wiki/2021_New_York_City_Democratic_mayoral_primary.

E. Money

1. Outside spending (OpenSecrets, all federal races): 2008 $338.4M; 2010 $309.8M; 2012 $1.038B; 2016 $1.411B; 2020 $2.914B; 2022 $2.028B; 2024 $4.218B (nominal dollars). https://www.opensecrets.org/outside-spending/by_cycle. Note: Citizens United was decided Jan 2010; SpeechNow followed in March 2010. 2008 is the last fully pre-CU cycle.

F. Social media, local news, trust

1. Meta/academic 2020 election studies (Science/Nature, July 2023): (a) switching users to a chronological feed sharply cut time on platform but produced no significant changes in affective or issue polarization, knowledge, or turnout; (b) removing reshares cut exposure to untrustworthy sources (6.2% → 2.5% of political content) with no distinguishable change in polarization; (c) only 20.6% of users got >75% of exposures from like-minded sources, and cutting like-minded content by one-third had no measurable effect on 8 preregistered attitudes; (d) ideological segregation in news exposure was substantial and asymmetric — conservative audiences were more homogeneous. https://www.techpolicy.press/a-primer-on-the-meta-2020-us-election-research-studies/ ; https://www.science.org/doi/10.1126/science.abp9364 (2023). Independence caveat: Meta shaped the agenda (M. Wagner's rapporteur assessment).
2. Deactivation, 2020 (Allcott et al., PNAS 2024): 19,857 Facebook and 15,585 Instagram users deactivated six weeks before the election. No detectable effect on affective or issue polarization, turnout or legitimacy beliefs; Facebook deactivation reduced Trump support ~1 point, falling "just short" of the preregistered significance threshold. https://www.niemanlab.org/2024/05/people-who-got-off-facebook-for-6-pivotal-weeks-in-2020-may-have-been-less-likely-to-vote-for-trump/ ; paper https://www.pnas.org/doi/10.1073/pnas.2321584121. A 2026 correction was published (https://pubmed.ncbi.nlm.nih.gov/41849406/); the panel could not retrieve its text and does not know whether it alters estimates.
3. Deactivation, 2018 (Allcott, Braghieri, Eichmeyer & Gentzkow, AER 2020): four-week deactivation "reduced both factual news knowledge and political polarization" and increased subjective well-being. https://www.aeaweb.org/articles?id=10.1257%2Faer.20190658.
4. Local news (Medill State of Local News 2025): 213 counties with no local news source (up from 206); 1,524 more with only one; ~50 million Americans with limited or no access; nearly 40% of local newspapers (~3,500) gone since 2005; 136 disappeared in the past year; 270,000+ newspaper jobs (>75%) lost since 2005. https://www.niemanlab.org/2025/10/in-medills-latest-state-of-local-news-report-a-festering-20-year-old-problem-looms-larger-than-ever/ ; report: https://localnewsinitiative.northwestern.edu/projects/state-of-local-news/2025/report/.
5. Newspaper closures → straight-ticket voting: Darr, Hitt & Dunaway (J. Communication 2018), matched counties: split-ticket voting fell 1.9% after a local paper closed. https://academic.oup.com/joc/article-abstract/68/6/1007/5160090.
6. Trust in media (Gallup, Sept 2025): 28% trust mass media "a great deal/fair amount" — a record low; Republicans 8%, independents 27%, Democrats 51%; 68–72% in the 1970s. https://news.gallup.com/poll/695762/trust-media-new-low.aspx.
7. Confidence in institutions (Gallup, 2026): Congress 9%, Supreme Court 27%, presidency 27%, military 61% ("great deal/quite a lot"). https://news.gallup.com/poll/1597/confidence-institutions.aspx.
8. Trust in federal government (Pew): 77% in 1964; 22% in 2024 (D 35%, R 11%); 17% in 2025 (R 26%, D 9%) — trust flips with the White House. https://www.pewresearch.org/politics/2025/12/04/public-trust-in-government-1958-2025/ ; https://www.pewresearch.org/politics/2024/06/24/public-trust-in-government-1958-2024/.

G. Access and integrity

1. Strict voter ID (Cantoni & Pons, QJE 2021): DiD on a 1.6-billion-observation panel, 2008–2018: "no negative effect on registration or turnout, overall or for any group defined by race, gender, age, or party affiliation," and no effect on actual or perceived fraud. https://www.nber.org/papers/w25522.
2. Newer ID evidence (Lipkovitz, Research & Politics 2025, synthetic DiD, 1984–2020): aggregate effect null, but heterogeneous: −2.7 pts in presidential elections for late-adopting states; +2.9 pts in midterms. https://journals.sagepub.com/doi/full/10.1177/20531680251401762.
3. Fraud incidence: Brennan Center (2007): proven individual voter fraud occurs "about as often as death by lightning strike"; in Washington's 2004 governor's race (2M+ votes) one ballot was suspected of double-voting. https://www.brennancenter.org/our-work/research-reports/myth-voter-fraud. Heritage's Election Fraud Map (updated Dec 12, 2025) catalogs "a sampling of proven instances," including absentee-ballot and duplicate-voting convictions. https://electionfraud.heritage.org/. Both sources agree fraud exists and is prosecuted; they disagree about whether the documented cases are the tip of an iceberg.
4. Election security, 2020: the CISA/Election Infrastructure GCC joint statement calling 2020 "the most secure in American history" is widely cited; the panel's fetch of the CISA page failed (403), so we list it as not independently verified here.

---

Contested evidence (where the literature genuinely disagrees)

• Is affective polarization as bad as thermometers say? Tyler & Iyengar (APSR 2024, "Testing the Robustness of the ANES Feeling Thermometer Indicators") defend the measure; Druckman, Klar, Krupnikov, Levendusky & Ryan (JOP 2022) argue respondents picture activists/elites when rating "the party," inflating hostility toward ordinary opponents. Mode effects (item 1) add ~6–8 points of noise.
• Do primaries cause polarization? Hall (2015) shows general electorates punish extremists — a moderating filter. Hirano & Snyder find primaries' contribution modest. But the threat channel — incumbents voting to avoid a challenge — is harder to test; Anderson, Butler & Harbridge-Yong (Rejecting Compromise, 2020) find legislators believe primary voters punish compromise. Candidate-supply arguments (Hall, Who Wants to Run?, 2019) suggest the damage is in who runs, not who wins.
• Districts vs. sorting: MPR (2009) → little roll-call polarization from maps; Cook (2023) → 42% of the lost swing seats from boundaries. These are not contradictory: maps matter for competitiveness and seat shares more than for ideology given party.
• Social media: the 2020 Meta experiments (short, large, platform-cooperative) find near-null attitude effects; the 2018 AER deactivation study found a polarization reduction; Boxell et al. find polarization rose fastest among low-internet users. Unresolved: long-run, equilibrium, and elite-level effects that individual-level randomization can't capture.
• Voter ID: Cantoni & Pons (null everywhere) vs. Lipkovitz (null on average, heterogeneous by context) vs. earlier Hajnal-Lajevardi-Nielson (2017) estimates of racial turnout gaps that were later challenged on data-coding grounds (Grimmer et al. 2018).
• RCV/top-four: Alaska's bipartisan legislative coalitions vs. the 2022 Condorcet failure and the near-repeal. NYC 2021 shows RCV can be administered but was marred by a tabulation error.
• Money: outside spending rose ~12× in nominal terms 2008→2024, but whether it changes outcomes (vs. following competitive races) is disputed; the literature on spending effects in general elections finds small marginal effects.

STAGE 2 — OPENING POSITIONS (Part 1 of 2)

Format per professor: causal claim · evidence · remedy · concession · falsifiable prediction.

---

1. Prof. Amara Johnson (#38, voting rights, L)

Causal claim. The central threat to representation in 2026 is not polarized feelings; it is the collapse of legal limits on map-drawing combined with barriers to the ballot. Rucho (2019) closed federal courts to partisan claims; Callais (April 2026) has gutted §2 as a check on racial vote dilution; and the result is the 2025–26 mid-decade arms race (Brief #17) in which roughly a dozen states redrew maps outside the census cycle, erasing ~14 competitive seats.
Evidence. Brief #14, #16, #17. Cook's own decomposition (#6) attributes 42% of the lost swing seats to boundaries. And strict voter-ID laws, which states have multiplied, reduce minority turnout by several points — that is what the ID-law literature shows.
Remedy. (i) Federal statute banning mid-decade congressional redistricting absent a court order; (ii) mandatory independent commissions for congressional maps; (iii) restore §2 via statute (John Lewis VRAA framework); (iv) automatic voter registration.
Concession. Commissions don't fix polarization in the chamber — MPR (#8) and Kousser et al. (#12) are persuasive on that. My argument is about fairness and competitiveness, not ideology.
Prediction. If I'm right, the 2026 House seat-share-to-vote-share gap in states that redrew mid-decade will exceed that of states that didn't, and the number of House races decided by <5 points will fall below its 2022 level.

---

2. Prof. Rosa Delgado-Finch (#7, political economy, Soc)

Causal claim. Polarization is downstream of oligarchic financing. When a few thousand donors and a few hundred super PACs set the agenda, parties compete for donors rather than median voters, and donors are more ideological than voters.
Evidence. Outside spending went from $338M (2008) to $4.2B (2024) — more than twelvefold (#23). The post-*Citizens United* regime turned elections into auctions: the bigger spender wins roughly nine races in ten.
Remedy. Constitutional amendment overturning Citizens United; small-donor public matching (6:1, NYC-style); full donor disclosure for any group spending on elections.
Concession. Nominal dollars overstate the rise somewhat (inflation, population); and Bernie Sanders-style small-donor campaigns show money doesn't only flow rightward — "big money" on the left (Bloomberg 2020) also exists.
Prediction. Within two cycles of a matching-funds program in a state, the share of state legislative candidates' money from donors under $200 should rise by >15 points, and roll-call extremity of those legislators should not rise faster than the national trend.

---

3. Prof. Naomi Fischer (#76, media studies, L)

Causal claim. Engagement-optimizing platforms are the accelerant. The algorithm rewards outrage, the reshare button multiplies misinformation, and the 2020 Meta studies prove how much power these systems have over what people see.
Evidence. Removing reshares cut untrustworthy-source exposure from 6.2% to 2.5% (#24b); segregation is extreme and asymmetric (#24d); the 2018 deactivation study found Facebook causes polarization (#26); the 2024 PNAS deactivation found Facebook moved Trump support (#25). Posters have perception gaps of 29 vs. 18 (#5).
Remedy. Platform duty-of-care statute; mandatory algorithmic audits and researcher data access (EU DSA Art. 40 model); default non-engagement ranking for political content.
Concession. The chronological-feed and like-minded-content experiments found null attitude effects over ~3 months.
Prediction. If platforms matter, a country that imposes default chronological/non-engagement ranking (EU via DSA) will show slower growth in affective polarization over 2026–2030 than the US.

---

4. Prof. Fiona McAllister (#69, journalism, C-L)

Causal claim. The disappearance of local news nationalizes politics: voters who once judged a sheriff or congressman on local performance now judge them as avatars of Team Red/Blue.
Evidence. Medill 2025: 213 news-desert counties, 1,524 single-source counties, ~50M people with limited access, 3,500 papers and 270,000 jobs gone (#27). Darr-Hitt-Dunaway: split-ticket voting fell 1.9% after closures (#28). Cook 2025: ticket-splitting hasn't recovered (#7). News deserts are a primary driver of the nationalization we see in the Cook PVI.
Remedy. Refundable payroll tax credit for local journalists; state civic-information consortia (New Jersey model, est. 2018); tax-exempt status pathways for nonprofit newsrooms.
Concession. Subsidy creates dependence and editorial capture risks; and the Darr effect, while causal-ish, is small.
Prediction. Counties that gain a new nonprofit newsroom should show higher split-ticket voting and higher local-race roll-off completion within two cycles relative to matched controls.

STAGE 2 — OPENING POSITIONS (Part 2 of 2)

5. Prof. Catherine Doyle (#25, election law, C)

Causal claim. The binding constraint is the partisan primary in a safe seat. With 82 swing seats (#6), the only election that matters for ~80% of members is a primary with ~21% turnout (#11). That electorate is older and more partisan, and the threat of it disciplines incumbents even when no challenger materializes.
Evidence. Ferrer 2026 (#11); Cook swing-seat collapse (#6); Alaska's post-reform legislature is governed by a bipartisan 21-member coalition (#21). California's top-two moderated its legislature after 2012.
Remedy. Nonpartisan top-four or top-five primaries with RCV generals (Alaska model); at minimum, open primaries to unaffiliated voters.
Concession. Hall (#9) shows general electorates already punish extremists who get nominated, which caps how much primaries can matter at the nominee level; the channel has to be the threat, which is harder to measure.
Prediction. If I'm right, legislators elected under top-four/RCV should show (a) higher cross-party co-sponsorship and (b) roll-call positions closer to their district median than same-state predecessors, within two cycles.

---

6. Prof. Gregory Stahl (#29, congressional studies, C)

Causal claim. Reformers are treating symptoms. Polarization is overwhelmingly sorting: ideology, religiosity, education, and geography have aligned with party. Institutional tweaks — commissions, RCV, open primaries — work at the margin of a margin.
Evidence. MPR 2009: maps explain little roll-call polarization (#8). Kousser et al.: California's reforms didn't halt polarization (#12). Hall: extremist nominees are punished (#9). The US is an outlier cross-nationally (#2), but Boxell et al. find the fastest growth among the oldest, least online (#3) — consistent with sorting and cable news, not platforms. Gerrymandering explains essentially none of the decline in competitive seats; it's geography.
Remedy. Modest: ban mid-decade redistricting to stop the arms race; restore regular order in Congress (committee power, open amendment process); otherwise temper expectations.
Concession. If Alaska's coalition governance proves durable across multiple cycles, it is real evidence institutions can bend behavior.
Prediction. States adopting open or top-four primaries will show no statistically detectable change in DW-NOMINATE extremity of their delegations over 10 years relative to synthetic controls.

---

7. Prof. Eleanor Whitfield (#21, constitutional law, originalist, R)

Causal claim. Distrust in elections — the thing that most threatens the system — is rational given lax integrity rules and a media class that 92% of Republicans no longer trust (#29: 8%). Integrity measures build confidence; federalizing election administration destroys it.
Evidence. Cantoni & Pons (#32): ID laws do not reduce turnout — the suppression narrative was wrong. Heritage's database documents real, prosecuted fraud (#34). Rucho correctly left political questions to legislatures and Congress's Elections Clause power (#14). Voter ID laws have measurably increased public confidence in elections. Fraud is common enough to swing close races — Alaska 2024 was decided by 743 votes (#18).
Remedy. Photo ID with free IDs; citizenship verification at registration; cleaner voter rolls; oppose any federal commission mandate as unconstitutional commandeering of state legislative functions.
Concession. Callais will produce maps that many minority voters experience as dilution, and mid-decade redistricting by my own side (Texas first) has damaged trust in the process.
Prediction. States with photo ID + free ID will show higher confidence-in-count survey measures among the losing party's voters than non-ID states after 2026.

---

8. Prof. Victor Lindqvist (#37, libertarian legal theory, Lib)

Causal claim. Speech restrictions — whether on money or platforms — entrench incumbents and the information gatekeepers whose collapse in trust (#29, #30) is itself a response to their failures. Outside spending is speech that follows competitiveness; it doesn't create outcomes.
Evidence. The Meta experiments show algorithms don't move attitudes (#24, #25). Trust flips with the White House (#31) — it's partisan, not informational. Outside spending has no measurable effect on who wins; the spenders simply pile into already-close races.
Remedy. Deregulate: raise or abolish contribution limits to candidates and parties (so money flows through accountable parties rather than super PACs); oppose platform "duty of care" laws; protect anonymous political speech (NAACP v. Alabama).
Concession. Disclosure above a high threshold is defensible; anonymous donors can also be foreign.
Prediction. If raising party contribution limits works, the share of spending by super PACs vs. parties will fall within two cycles in states that raise limits, without an increase in legislative extremity.

---

9. Prof. Leila Farouk (#40, political psychology, C)

Causal claim. Much of affective polarization is misperception: we hate a caricature. Partisans think opponents are twice as extreme as they are (#5). This perception gap, fed by heavy news and social posting, drives moralized hostility (#4).
Evidence. More in Common (#5); Pew moralization trend (#4); the Strengthening Democracy Challenge megastudy (Science 2024; https://www.strengtheningdemocracychallenge.org/) tested 25 interventions. Correcting perception gaps durably reduces partisan animosity.
Remedy. Scale "correct-the-meta-perception" interventions via civics curricula, local media, and bridging organizations; reduce the salience of extremes in news selection.
Concession. Reducing animosity doesn't automatically reduce support for antidemocratic practices — those are separable outcomes.
Prediction. A national-scale perception-correction campaign should move ANES out-party thermometers by ≥3 points within 4 years in exposed groups.

STAGE 3 — CROSS-EXAMINATION (Part 1: access, money, districts)

---

3.1 Holm (#39, C-R) → Johnson (#38, L): voter ID

Holm: Amara, you said strict ID laws "reduce minority turnout by several points — that is what the literature shows." The best-identified study we have says the opposite. Cantoni & Pons (QJE 2021; Brief #32) used a 1.6-billion-observation panel and found no negative effect on registration or turnout for any group defined by race, gender, age, or party. The earlier Hajnal–Lajevardi–Nielson estimates you're presumably relying on were challenged by Grimmer et al. (2018) because the survey-based turnout measures were unstable across states. You can't cite "the literature" for a result its best study rejects.

Johnson: Two responses. First, Cantoni & Pons themselves suggest parties counter-mobilized — the null is an equilibrium outcome in which Democratic groups spent resources to overcome the burden. That's not costless; it's a tax on one side's organizing budget. Second, Lipkovitz (2025; #33) finds a −2.7 point effect in presidential elections for late-adopting states. So "no effect anywhere" is also too strong.

Holm: Lipkovitz's aggregate estimate is null, and the same paper finds +2.9 points in midterms. If you take the −2.7 as real you must take the +2.9 as real, and then your "suppression" story nets to roughly zero with sign flips — which looks more like noise across specifications than a mechanism. On counter-mobilization: that's a hypothesis in the paper's discussion, not an estimate.

Johnson (amends on the record): Fair. I withdraw "reduces minority turnout by several points" — Cantoni & Pons's group-level nulls are the strongest evidence and I can't square that phrasing with them. My revised claim: the best evidence finds no average turnout effect; the heterogeneous estimates (Lipkovitz) are unresolved; my objection to strict ID is now primarily about cost-shifting onto organizers and the lack of any measured fraud or confidence benefit (#32) — the policy buys nothing either side claims for it.

---

3.2 Pruitt (#8, Lib) → Delgado-Finch (#7, Soc): money

Pruitt: "The bigger spender wins nine in ten." That statistic is almost entirely incumbency and selection: incumbents raise more and win, and money flows to candidates expected to win. It tells you nothing about the causal effect of a marginal dollar. Your own table (#23) also shows 2022 outside spending ($2.03B) below 2020 ($2.91B) — hardly a monotonic "auction."

Delgado-Finch: I concede the 90% figure is a correlation, not a causal effect. But the mechanism I care about isn't the marginal general-election dollar; it's agenda-setting before the ballot — who can run, whose issues get airtime, what a party refuses to propose. Money's influence on the menu is where I'd point you.

Pruitt: Then name the evidence. Outside spending rose ~12× nominally since 2008 (#23); affective polarization has been rising since the late 1970s (#1), three decades before Citizens United. The timing doesn't fit a CU-driven story.

Delgado-Finch: The timing fits a money story, not a CU story: the donor class's share of contributions was climbing long before 2010. But yes — I'll amend: I can't show from the evidence in this Brief that CU caused polarization. I stand by the claim that outside spending grew roughly twelvefold in nominal terms and that its opacity is a legitimate public concern.

---

3.3 Johnson (#38, L) → Stahl (#29, C): "gerrymandering explains essentially none"

Johnson: Gregory, you wrote that gerrymandering explains "essentially none" of the decline in competitive seats. Cook's own decomposition (#6) — not a left-wing source — attributes 42% (net 32 of 77) of lost swing seats to boundary changes. And that was before the 2025–26 mid-decade redraws, which by the running compilation eliminated ~14 more competitive seats (#17).

Stahl: I conflated two things and you're right to catch it. MPR (#8) is about roll-call polarization — whether members from gerrymandered seats vote more extremely than members from neutral seats representing similar voters. The answer is: barely. Cook is about competitiveness. I retract "essentially none" for competitiveness; the correct statement is "sorting explains most (58%) of the swing-seat decline; boundaries explain a large minority (42%)."

Johnson: And competitiveness matters for polarization via the primary channel Catherine is pushing — a safe seat makes the primary the only election.

Stahl: That's the hinge, and I don't think it holds as tightly as you want. Hall (#9) shows general electorates punish extremists even in competitive environments, and McCarty et al.'s simulations show neutral maps don't create meaningfully more moderate delegations — because sorting has already made most neutral-drawn districts safe too. The 82 remaining swing seats would not become 164 under any commission.

Johnson: No, but under California's commission the state went from 1 party change in 255 congressional races in the prior decade (#13) to real turnover. That's a competitiveness effect even if Kousser finds no ideological one.

Stahl: Accepted — for competitiveness. On polarization-in-the-chamber I don't move: Kousser et al. (#12) is the most direct test we have and it's null.

---

3.4 Ashford (#31, R) → Doyle (#25, C): primaries and California

Ashford: Catherine, you wrote that "California's top-two moderated its legislature after 2012." The Brief's literature summary (#12) says Kousser, Phillips & Shor found the reforms did not halt polarization, and Ahler-Citrin-Lenz found insufficient crossover. And Alaska's system produced a textbook Condorcet failure in 2022 (#20): the candidate a majority preferred to each rival was eliminated first. You're selling a system whose flagship produced the "wrong" winner by its own logic and nearly got repealed (#18).

Doyle: On California: I overstated. There is some work finding modestly more moderate candidates in certain open-seat races, but I can't cite it from the verified record here, and Kousser et al. is the direct test. I withdraw "moderated its legislature" and replace with "did not measurably moderate roll-call behavior." On Alaska, I'll hold: the top-four primary and the RCV general are separable. The Condorcet failure came from RCV's elimination order, which a top-four-plus-Condorcet or approval general could fix. The 21-member bipartisan majority (#21) is the most interesting data point any state has produced.

Ashford: A coalition of 14 Democrats, 5 independents and 2 Republicans in a state Trump won — that's not moderation, that's a minority party governing via a handful of defectors. Republicans see that as the mechanism of the reform: diluting the majority party's primary.

Doyle: Or it's evidence that legislators freed from closed primaries will cross lines — which is exactly the behavior reform intends. Which description is right is empirical: look at roll-call extremity of those 2 Republicans vs. their pre-reform predecessors. I'll concede that we don't yet have a clean study.

STAGE 3 — CROSS-EXAMINATION (Part 2: information ecosystem, integrity, speech, psychology)

---

3.5 Tran (#58, computer science, Lib) → Fischer (#76, L): platforms — four rounds

Tran (R1): Naomi, you listed four findings as proof platforms "accelerate" polarization. Three of the four 2020 experiments you cite found nulls on polarization: chronological feed (no significant change in affective or issue polarization despite a massive drop in usage), reshare removal (no distinguishable change), like-minded reduction (no effect on 8 preregistered attitudes) (#24). The 2020 deactivation of ~35,000 users found no detectable effect on affective or issue polarization (#25). And Boxell et al. find polarization grew fastest among people over 75 (#3). Your causal claim is contradicted by the largest experiments ever run on the question.

Fischer (R1): Three things. (1) Those experiments manipulated individuals for ~3 months in the most saturated information environment in history — an election. You can't detect a stock variable built over 15 years by nudging a flow for 12 weeks. (2) They were conducted with Meta's cooperation and under Meta's constraints (#24 caveat). (3) The 2018 AER study (#26) did find deactivation reduced polarization. So the evidence is mixed, not null.

Tran (R2): (1) is unfalsifiable as stated — any null can be attributed to "long-run effects." What would you accept as disconfirming? (3) The 2018 study's polarization effect was on an index that was driven heavily by issue-position items, and the 2020 replication with ~10× the sample found no effect. When a larger, preregistered replication fails, the prior study gets down-weighted, not averaged in.

Fischer (R2): Disconfirming evidence: if cross-national trends in affective polarization show no relationship to social media penetration, I'd drop the "accelerant" claim. And Boxell-Gentzkow-Shapiro (#2) actually help me a little — six countries saw affective polarization decline over the same decades in which social media spread globally, which means platforms can't be sufficient. Fine. But the 2020 deactivation study found Facebook moved Trump support ~1 point (#25). In an election decided in three states by tens of thousands of votes, that is not nothing.

Tran (R3): You just cited Boxell et al. against yourself — platforms spread everywhere and polarization fell in half the sample. That's disconfirming by your own test. And the Trump estimate fell short of the preregistered significance threshold, per the authors. Also — there's a 2026 correction to that PNAS paper (#25) whose content none of us has read. Citing a not-significant point estimate from a paper under correction to claim electoral effects is exactly the overreach the referees will flag.

Fischer (R3): Your point on the cross-national data lands. I'll amend: the evidence does not support "platforms are the accelerant of affective polarization" at the individual-attitude level. I retract "the 2018 study proves Facebook causes polarization" — it found a reduction in a single short experiment that the larger 2020 study didn't replicate. I keep two claims: (a) platform design demonstrably shapes exposure — reshare removal cut untrustworthy content by more than half (#24b) — which matters for the information environment even if attitudes are sticky; (b) the conservative-audience segregation asymmetry (#24d) is a real structural feature.

Tran (R4): Accepted on (a) and (b) as descriptions. But (a) cuts against your remedy: if exposure changes don't move attitudes, a "duty of care" statute imposes speech costs for an unmeasured benefit. I'd support researcher data access — that's the thing that would actually resolve our disagreement.

Fischer (R4): Then we agree on the one remedy that generates evidence. I'm not dropping duty-of-care, but I'll move it to "contingent on what access reveals."

---

3.6 Ward (#68, rural sociology, R) → McAllister (#69, C-L): local news

Ward: Fiona, "news deserts are a primary driver of nationalization." Darr et al. (#28) find a 1.9% drop in split-ticket voting after closures. Meanwhile Pew shows moralized hostility nearly doubled between 2016 and 2022 (#4) — a period when the national media, not local, dominated. And rural Republicans stopped trusting the press long before their papers closed; Republican trust is now 8% (#29).

McAllister: 1.9% is small per closure, but it's per county and there have been ~3,500 closures (#27). Aggregate them and you get a meaningful fraction of the lost ticket-splitting.

Ward: You can't just multiply — closures cluster in the counties that were already the most sorted, and the Darr design matches on observables only. Show me the aggregate estimate in a published source.

McAllister: I can't. Amend: local-news loss is a contributor to nationalization with credible but small causal estimates; "primary driver" is not supported. I maintain the civic case (coverage of local government, school boards) independent of polarization.

---

3.7 Bell (#22, civil rights law, L) → Whitfield (#21, R): integrity claims

Bell: Eleanor, two claims. (1) "Voter ID laws have measurably increased public confidence." The same paper you cited for turnout — Cantoni & Pons (#32) — found no effect on actual or perceived fraud. You can't use half the paper. (2) "Fraud is common enough to swing close races — Alaska 2024 was decided by 743 votes." Nothing in the record connects any fraud to Alaska's margin. Heritage's own map calls itself "a sampling" of cases (#34); the Brennan Center finds fraud about as common as lightning strikes.

Whitfield: On (1): I'll concede the paper finds no effect on perceived fraud; I was relying on survey toplines showing ID is popular, which is not the same as increasing confidence. Withdrawn. On (2): the claim is that with margins of hundreds of votes, the tolerance for any irregularity is near zero; Heritage documents duplicate voting and absentee fraud that do occur. I did not claim fraud decided Alaska.

Bell: You wrote "fraud is common enough to swing close races" and then cited Alaska. That's an implication, and the evidence doesn't support "common."

Whitfield: Amended: "Documented fraud is rare but non-zero; in extremely close races, rare events can matter, which justifies low-cost integrity measures." I retain my view that Rucho was correctly decided and that federal commission mandates would raise serious Elections-Clause and anti-commandeering questions — Congress may regulate congressional elections, but it cannot direct state legislatures to create state agencies of a particular design.

Bell: On Elections Clause power over congressional maps, Congress plainly can set standards — it mandated single-member districts in 1967. We'll disagree on mandating commissions specifically.

---

3.8 Kerr (#30, public policy, C-L) → Lindqvist (#37, Lib): money and trust

Kerr: Victor, "outside spending has no measurable effect on who wins" is stronger than any estimate in the record. The honest statement is that the marginal general-election effect is hard to identify and probably small; nobody has shown "no effect." And your trust argument cuts both ways: if trust is purely partisan (#31), then "gatekeeper failure" can't explain it either.

Lindqvist: Fair on the first. Amend: "Marginal effects of outside spending on general-election outcomes are small and poorly identified; correlation with winning mostly reflects targeting of close races." On trust: the White-House flip (2024: D 35/R 11; 2025: R 26/D 9) is about government; media trust (8% among Republicans) hasn't flipped with power — so media trust has a non-partisan-cycle component that looks like a verdict on performance.

Kerr: Or a verdict produced by a generation of elite cues telling Republicans not to trust the press. The data can't distinguish those.

Lindqvist: Agreed that they can't. That's the honest place to leave it.

---

3.9 Hendricks (#99, game theory, Lib) → Farouk (#40, C): durability

Hendricks: Leila, "correcting perception gaps durably reduces animosity." The megastudy you cite tested 25 interventions and — as the project reports — many reduced partisan animosity in the short run, but the headline lesson was that effects on antidemocratic attitudes were weaker and that animosity reductions faded. I see nothing in the record establishing durability.

Farouk: The project page I linked doesn't report effect sizes or follow-up durations, and I was relying on memory. Amend: perception-correction interventions reduce measured animosity in the short run; durability is not established; effects on antidemocratic attitudes are weaker and distinct. I'd add that More in Common's gradient — the disengaged being ~3× more accurate (#5) — suggests the gap is manufactured by engagement, which is a structural input, not a one-shot lesson.

STAGE 4 — REFEREE FACT-CHECK

Brandt (#57) · Takahashi (#17) · Doyle (#25, recused on her own claims — Brandt & Takahashi rated those)

Ratings: S = Supported · C = Contested (credible evidence on both sides, or supported only in part) · U = Unsupported (no evidence in the record / not verifiable) · W = Wrong (contradicted by the best available evidence). "As amended" is noted where a professor corrected on the record in Stage 3; we rate the original claim and note the amendment.

# · Professor · Claim (original wording, condensed) · Rating · Basis
1 · Johnson · Rucho closed federal courts to partisan-gerrymandering claims · S · Brief #14
2 · Johnson · Callais has gutted §2 as a check on vote dilution · C · Holding narrows §2 sharply (Brief #16); "gutted" is the dissent's characterization; §2 not struck down
3 · Johnson · ~a dozen states redrew mid-decade, erasing ~14 competitive seats · C · Secondary compilation (Brief #17): 11 enacted maps + VA voided; "14 competitive seats" is a compiler estimate, not an election result
4 · Johnson · Cook attributes 42% of lost swing seats to boundaries · S · Brief #6
5 · Johnson · Strict ID laws reduce minority turnout by several points · W · Cantoni & Pons (Brief #32): no negative effect for any racial group. Retracted in 3.1.
6 · Johnson · Lipkovitz finds −2.7 pts in presidential elections for late adopters · S · Brief #33 (with +2.9 in midterms; aggregate null)
7 · Johnson (3.3) · CA went from 1 party change in 255 races to "real turnover" · C · "1 of 255" is S (Brief #13); magnitude of post-commission turnover not quantified in record
8 · Delgado-Finch · Outside spending $338M (2008) → $4.2B (2024), >12× · S · Brief #23 (nominal; 12.5×)
9 · Delgado-Finch · Bigger spender wins ~9 in 10 races (implying causation) · U · No source in record; as a correlational stat it's plausible but it is not causal evidence. Amended in 3.2.
10 · Delgado-Finch · CU-era "auction" drove polarization · U · Timing mismatch: out-party hostility rising since 1978 (Brief #1). Amended.
11 · Pruitt (3.2) · 2022 outside spending below 2020 · S · Brief #23 ($2.03B vs $2.91B; midterm vs presidential cycle, so partly apples-to-oranges)
12 · Fischer · Reshare removal cut untrustworthy-source exposure 6.2%→2.5% · S · Brief #24b
13 · Fischer · News-exposure segregation is extreme and asymmetric · C · Asymmetry S (Brief #24d); "extreme" contradicted by 20.6% figure for >75% like-minded exposure (#24c)
14 · Fischer · 2018 deactivation study shows Facebook causes polarization · C · Found a reduction (Brief #26) but not replicated in the larger 2020 study (#25). Retracted as "proves."
15 · Fischer · 2020 deactivation shows Facebook moved Trump support · C · ~1 pt point estimate, "just short" of preregistered threshold; 2026 correction content unknown (Brief #25)
16 · Fischer · Posters' perception gap 29 vs 18 · S · Brief #5
17 · Fischer · Platforms are the accelerant of affective polarization · U · Individual-level RCTs null (#24, #25); cross-national declines despite platform spread (#2); growth fastest among least-online (#3). Amended in 3.5.
18 · Tran (3.5) · 2018 study's polarization effect was driven mainly by issue items · U · Not verifiable from sources the panel could retrieve (NBER abstract lacks component detail)
19 · Tran (3.5) · 2020 replication had ~10× the sample · C · 2020 n ≈ 35,000 (Brief #25) is verified; 2018 n not verified by panel
20 · McAllister · Medill figures (213 / 1,524 / 50M / 3,500 / 270,000) · S · Brief #27
21 · McAllister · Split-ticket voting fell 1.9% after closures · S · Brief #28
22 · McAllister · News deserts are a primary driver of nationalization · U · Only small per-closure estimates exist; no aggregate estimate. Amended in 3.6.
23 · Ward (3.6) · Moralized hostility roughly doubled 2016–2022 · C · "Immoral": R 47→72, D 35→63 (Brief #4) — +53% and +80%; "nearly doubled" holds for Democrats only
24 · Doyle · Primaries ~21% of eligible vs 53% general · S · Brief #11
25 · Doyle · Only ~80% of members face a meaningful general (82 swing seats) · C · 82 swing seats of 435 (Brief #6) implies ~81% outside R+5/D+5; but PVI outside ±5 ≠ uncompetitive in every cycle
26 · Doyle · California's top-two moderated its legislature · W · Kousser, Phillips & Shor (Brief #12): did not halt polarization. Retracted in 3.4.
27 · Doyle · Alaska House governed by 21-member bipartisan coalition · S · Brief #21
28 · Ashford (3.4) · Alaska 2022 produced a Condorcet failure · S · Brief #20
29 · Stahl · Maps explain little roll-call polarization · S · McCarty-Poole-Rosenthal (Brief #8)
30 · Stahl · Fastest polarization growth among oldest/least online · S · Brief #3
31 · Stahl · Gerrymandering explains essentially none of the swing-seat decline · W · Cook: 42% (Brief #6). Retracted in 3.3.
32 · Stahl (3.3) · 82 swing seats wouldn't become 164 under any commission · C · Consistent with 58% sorting share (#6) but a counterfactual; not directly tested
33 · Whitfield · Republican media trust 8% · S · Brief #29
34 · Whitfield · ID laws don't reduce turnout · S · Brief #32 (with Lipkovitz heterogeneity, #33)
35 · Whitfield · Voter ID has measurably increased confidence in elections · W · Cantoni & Pons: no effect on perceived fraud (Brief #32). Retracted in 3.7.
36 · Whitfield · Fraud is common enough to swing close races (citing Alaska 743) · U · No evidence connecting fraud to Alaska's margin; Heritage "sampling" + Brennan (Brief #34) both show documented fraud is rare. Amended.
37 · Bell (3.7) · Congress mandated single-member districts in 1967 · S · Uniform Congressional District Act — https://en.wikipedia.org/wiki/Uniform_Congressional_District_Act
38 · Lindqvist · Meta experiments show algorithms don't move attitudes · C · Accurate for 3-month individual-level attitudes (#24); not evidence on long-run/equilibrium effects
39 · Lindqvist · Trust in government flips with the White House · S · Pew (Brief #31)
40 · Lindqvist · Outside spending has no measurable effect on who wins · U · No estimate in record establishes a null; amended in 3.8 to "small and poorly identified"
41 · Lindqvist (3.8) · Media trust hasn't flipped with power, unlike government trust · C · Republican media trust 8% in 2025 under a GOP president (#29) is consistent; panel lacks party-by-year media series to confirm
42 · Farouk · Partisans imagine ~2× as many opponents hold extreme views · S · Brief #5
43 · Farouk · Correcting perception gaps durably reduces animosity · U · Project page gives no durability data; amended in 3.9
44 · Farouk (3.9) · Disengaged ~3× more accurate · S · Brief #5
45 · Delgado-Finch · Small donors / NYC 6:1 matching as model · — · Policy proposal, not an empirical claim; not rated

Tally (44 rated claims): Supported 21 · Contested 13 · Unsupported 7 · Wrong 4 — wait, recount by row: S = 1,4,6,8,11,12,16,20,21,24,27,28,29,30,33,34,37,39,42,44 → 20; C = 2,3,7,13,14,15,19,23,25,32,38,41 → 12; U = 9,10,17,18,22,36,40,43 → 8; W = 5,26,31,35 → 4. Final: S 20 · C 12 · U 8 · W 4 (44 claims).

Referee notes.

• Brandt: The four "Wrong" ratings are distributed across blocs — one Left (Johnson on ID), two Center (Doyle on California, Stahl on gerrymandering), one Right (Whitfield on confidence). All four were retracted on the record before this table, which is how it should work.
• Takahashi: The most common failure mode was not fabrication but promotion of a short-run or correlational estimate to a causal law (spending-wins, deactivation-causes, news-deserts-drive, ID-increases-confidence). Also note the pattern that the same paper (Cantoni & Pons) was used selectively by both flanks — Whitfield took the turnout null and ignored the confidence null; Johnson ignored both.
• Doyle (on the Brief itself): Brief #17 relies on a live secondary compilation, and #35 (CISA 2020 statement) could not be verified by fetch. We have not relied on #35 for any rating.

STAGE 5 — STEELMAN EXCHANGE

Each bloc writes the strongest version of an opposing bloc's case; the target bloc replies.

---

5.1 Right steelmans the Left — by Prof. Frederick Holm (#39, C-R)

"The Left's best case is not about voter ID and not about 'misinformation.' It's this: the legal guardrails on how power converts votes into seats have been removed in six years, and the incentives now reward whoever redraws most aggressively. Rucho (2019) took federal courts out of partisan claims; Callais (2026) raised the §2 bar so high that majority-minority districts in the South are now vulnerable; and the 2025–26 cycle showed the equilibrium — 11 enacted mid-decade maps, ~14 competitive seats erased, one Democratic countermove (Virginia) voided by a state court while the Republican moves largely stood. Even if the net seat effect is modest (compiled R +4), the process has shifted from a decennial constraint to a continuous weapon. In that world, the defense of minority representation and competitive seats can't rely on courts; it requires statute. The Left is also right that outside spending rose ~12× nominally since 2008 while the share that is undisclosed makes accountability impossible — and Alaska's own voters wrote 'true source' disclosure into law in 2020."

Left reply (Johnson #38): Accepted as fair — more precise than my opening. I'd add only that the Right's own federalism logic supports banning mid-decade redraws, since it was Texas's 2025 move, not any Democratic state, that started the cascade.

---

5.2 Left steelmans the Right — by Prof. Tamsin Kerr (#30, C-L)

"The Right's best case: a system half the country does not trust cannot govern, and the Left has spent a decade predicting harms from integrity rules that the best evidence says didn't happen. Strict ID laws did not reduce turnout for any group (Cantoni & Pons); the 'Jim Crow 2.0' framing was empirically wrong. Meanwhile, trust in media among Republicans is at 8%, and confidence in Congress is 9%. When the losing side's trust collapses, low-cost integrity measures that are popular across parties are a legitimate way to rebuild it — even if the fraud they prevent is rare. On speech: the 2020 platform experiments found that algorithm changes don't move attitudes, which means 'duty of care' statutes would impose real speech costs for benefits nobody has measured. And federal mandates on how states structure redistricting bodies raise real constitutional questions — Congress has Elections Clause power over the rules, but not an obvious power to dictate the internal organization of state lawmaking."

Right reply (Whitfield #21): Accepted as fair, with one correction: I no longer claim ID increases confidence — Cantoni & Pons found no effect on perceived fraud, and I retracted. The honest Right case is that ID is cheap and costless to turnout, not that it's a confidence cure.

---

5.3 Center steelmans the Flanks' shared skepticism of structural reform — by Prof. Colleen Farrow (#34, C)

"Both flanks have a good argument against the Center's favorite toys. Left version: RCV and top-four are procedural fixes that let reformers avoid the substantive fights over money and access, and Alaska's 2022 center-squeeze shows RCV can fail its own majoritarian logic. Right version: nonpartisan primaries strip parties of their First Amendment association right to choose nominees, and in Alaska they produced a legislature governed by a coalition that the majority party's voters didn't choose. Both: California's commission + top-two didn't stop polarization (Kousser et al.), and the fastest-polarizing Americans are the oldest and least online — nothing about primaries or platforms explains that."

Left reply (Johnson): Fair.
Right reply (Ashford #31): Fair; I'd add the 2026 Alaska repeal measure is itself evidence that the reform lacks durable consent.

---

5.4 Right & Left jointly steelman the Center — by Prof. Victor Lindqvist (#37, Lib) and Prof. Marcus Bell (#22, L)

"The Center's best case is competitiveness as a public good independent of ideology. Even if maps don't change how members vote (McCarty et al.), they change whether members can lose — and 42% of lost swing seats came from maps. Primary electorates are ~21% of eligible voters; opening them raises turnout 5–11 points (Ferrer 2026). These are modest, cheap, evidence-backed changes to who participates, and they don't require anyone to win the substantive argument about money or speech."

Center reply (Doyle #25): Accepted — and better stated than my opening, which overclaimed on moderation.

STAGE 6 — REVISED PROPOSALS

Drafted by Doyle (#25) with Johnson (#38), Whitfield (#21), Stahl (#29), Fischer (#76) and Tran (#58); costs vetted by Takahashi (#17). No proposal below has a CBO or JCT score that the panel could locate; where we say "no official score," we mean it.

---

P1. Federal ban on mid-decade congressional redistricting

Mechanism. Amend 2 U.S.C. §2c (the 1967 single-member-district statute — Congress's existing Elections Clause footprint) so each state draws congressional districts once per decennial census, with redraws permitted only under court order (VRA/constitutional violation) or if the enacted map is invalidated. Effective for the 2032 cycle (after the 2030 census) to avoid rewarding or punishing either side's 2025–26 maps.
Cost. No official score; administrative cost negligible — it reduces redistricting activity.
Precedent. Colorado's Supreme Court held in 2003 (People ex rel. Salazar v. Davidson) that the state constitution permits one congressional redistricting per census; Texas's 2003 mid-decade redraw was upheld federally in LULAC v. Perry (2006) because nothing federal forbade it. The 2025–26 cascade (Brief #17: 11 enacted maps; one voided) is the natural experiment in what happens without a rule.
Key risk. A court-order exception can be gamed via friendly state courts; and the Right will argue (Whitfield) that locking in the 2025–26 maps until 2032 is itself partisan — hence the 2032 effective date.

P2. Require independent/bipartisan commissions for congressional maps

Mechanism. Federal statute requiring each multi-district state to use a commission (CA/MI-style balanced membership) for congressional maps.
Cost. No official score; CA-style commissions cost a few million dollars per cycle per state (panel did not verify a figure).
Precedent. California (Prop 11/20): competitiveness rose from 1 party change in 255 races in the prior decade (Brief #13) but polarization did not halt (#12); in 2025 California voters suspended their own commission's congressional map (Prop 50, 64.42%) to counter Texas — demonstrating that unilateral commissions are vulnerable to arms races.
Key risk. Constitutional challenge on anti-commandeering grounds (Whitfield); commissions can be captured by "independents" who are partisans.

P3. Open partisan primaries to unaffiliated voters (state-level model law)

Mechanism. States let unaffiliated registrants choose one party's primary ballot; parties keep their own nominations.
Cost. No official score; ballot-design and voter-education costs only.
Precedent. Colorado (2016 Prop 108), plus AK, ID, ME, OK, WA in Ferrer's sample: opening primaries to unaffiliated voters raised primary turnout ~5 points and made electorates younger and more unaffiliated (Brief #11). No evidence yet on legislative moderation.
Key risk. Party association claims (California Democratic Party v. Jones, 2000, struck down a blanket primary; semi-open primaries have fared better); raiding by opposite-party voters (limited evidence in top-two studies, #12).

P4. Top-four nonpartisan primary + RCV general (Alaska model), encouraged via federal grants

Mechanism. Voluntary grants to states adopting top-four/five primaries with RCV (or Condorcet-consistent) general elections.
Cost. No official score; Alaska's 2026 repeal would cost the state $2.6M to implement per APOC (Brief #19), which gives a rough sense of administrative scale.
Precedent. Alaska: 2022 center-squeeze (Brief #20); bipartisan House coalition (#21); repeal survived by 743 votes (#18), now on the ballot again Nov 2026 (#19). NYC 2021: RCV administered at scale with a notable tabulation error (#22).
Key risk. Condorcet failures erode legitimacy; low durable public consent.

P5. The "ID-plus-access" bargain

Mechanism. States adopting photo ID must provide free IDs (including underlying-document fees), and must adopt automatic voter registration at motor-vehicle agencies with citizenship verification against existing records.
Cost. No official score; free-ID costs borne by states.
Precedent. Carter–Baker Commission (2005) recommended photo ID paired with affirmative state efforts to register and issue IDs; Crawford v. Marion County (2008) upheld Indiana's ID law. Evidence: ID has no average turnout effect (Brief #32), heterogeneous estimates (#33), and no effect on perceived fraud (#32) — so neither side's stated rationale is supported; the bargain's case is that it's cheap and removes a salient grievance.
Key risk. Each flank sees the other's half as the poison pill; AVR list errors could feed fraud narratives.

P6. Platform researcher data access (privacy-protective)

Mechanism. Federal statute requiring very large platforms to provide vetted independent researchers access to non-public data (with differential-privacy and security safeguards), including advance notice of algorithm changes during elections; no content mandates.
Cost. No official score; compliance costs fall on platforms.
Precedent. EU Digital Services Act Art. 40 vetted-researcher access (implementation under way 2025–26). The 2020 Meta studies (Brief #24) show what cooperative access can produce — and their Meta-shaped agenda shows why access shouldn't depend on the company's consent.
Key risk. Data breaches/privacy (Moreau-type objections); platform litigation on compelled-disclosure grounds.

P7. Refundable payroll tax credit for local journalists

Mechanism. Refundable credit (e.g., 50% of first $50k of wages, declining after year 1) for outlets with local-coverage requirements; sunset after 5 years with evaluation.
Cost. No verified official score located by the panel (a similar provision appeared in 2021 reconciliation drafts).
Precedent. New Jersey Civic Information Consortium (est. 2018) — outcome evidence thin. Causal payoff evidence is modest (Darr et al.: −1.9% split-ticket after closure, #28).
Key risk. Subsidy capture; dependence on government; support flows to large chains owning local titles.

P8. "True source" donor disclosure for election spending above $10,000

Mechanism. Any entity making >$10k in federal independent expenditures/electioneering must disclose donors above $10k, tracing pass-throughs to the original source.
Cost. No official score; FEC administrative cost.
Precedent. Alaska's 2020 Measure 2 "true source" rule for donors >$2,000 — now itself on the 2026 repeal ballot (Brief #19). Outside spending: $4.2B in 2024 (#23).
Key risk. NAACP v. Alabama / AFP v. Bonta (2021) donor-privacy doctrine; chilling of unpopular causes (Lindqvist); higher threshold than Alaska's is meant to reduce that risk.

STAGE 7 — RECORDED VOTE

Blocs: Left 34 · Center 33 · Right 32 (moderator #100 abstains). Consensus requires ≥60% in each bloc (Left ≥21, Center ≥20, Right ≥20).

Proposal · Left yes · Center yes · Right yes · Consensus?
P1 Ban mid-decade congressional redistricting (effective 2032) · 30/34 (88.2%) · 29/33 (87.9%) · 21/32 (65.6%) · YES
P2 Mandatory federal commissions for congressional maps · 28/34 (82.4%) · 19/33 (57.6%) · 7/32 (21.9%) · NO
P3 Open partisan primaries to unaffiliated voters (state model law) · 22/34 (64.7%) · 27/33 (81.8%) · 20/32 (62.5%) · YES
P4 Top-four + RCV grants (Alaska model) · 15/34 (44.1%) · 18/33 (54.5%) · 8/32 (25.0%) · NO
P5 ID-plus-access bargain (free ID + AVR w/ citizenship check) · 17/34 (50.0%) · 26/33 (78.8%) · 18/32 (56.3%) · NO
P6 Platform researcher data access, privacy-protective · 31/34 (91.2%) · 28/33 (84.8%) · 20/32 (62.5%) · YES
P7 Local-journalist refundable payroll credit · 29/34 (85.3%) · 20/33 (60.6%) · 10/32 (31.3%) · NO
P8 "True source" disclosure >$10k · 33/34 (97.1%) · 26/33 (78.8%) · 12/32 (37.5%) · NO

Consensus reforms: P1, P3, P6. Near-misses: P5 (Right 56%, Left 50%) and P8 (Right 38%).

---

Vote explanations

Prof. Eleanor Whitfield (#21, R) — P1 YES, P2 NO, P5 YES, P8 NO.
"I voted for P1 because the 2032 effective date makes it neutral, and because a once-a-decade rule is exactly what a federalist should want — stability, not a permanent war. I voted against P2 on anti-commandeering grounds. P8: a $10k threshold is better than Alaska's $2k, but AFP v. Bonta is the law and donor lists for unpopular causes have been used for retaliation. What would change my mind on P8: evidence that disclosure regimes above $10k have produced no documented harassment of disclosed donors across several states over a decade."

Prof. Amara Johnson (#38, L) — P1 YES, P2 YES, P5 NO, P6 YES.
"P5 asks me to trade an ID requirement for AVR. After Cantoni & Pons I no longer believe ID suppresses turnout on average, but I also see no measured benefit, so I'm voting no on a rule that buys nothing. What would change my mind: a well-identified study showing ID increases losing-side confidence in the count — the one benefit that would justify the organizing cost."

Prof. Gregory Stahl (#29, C) — P1 YES, P3 YES, P4 NO, P7 NO.
"P3 is cheap and the turnout gain is well-identified (Ferrer). P4 asks federal money to back a system whose flagship produced a Condorcet failure and nearly got repealed; I'll wait for Alaska's November 2026 verdict and a clean roll-call study. What would change my mind on P4: if Alaska retains the system in 2026 and legislators elected under it show measurably less extreme roll-call positions than predecessors from the same districts."

Prof. Naomi Fischer (#76, L) — P6 YES, P7 YES.
"P6 is the remedy that would settle the dispute Marcus Tran and I couldn't. I've moved duty-of-care to contingent. What would change my mind on platforms generally: if multi-year data from access regimes show no link between engagement-ranking exposure and affective polarization, I'll drop the accelerant thesis entirely."

Prof. Marcus Tran (#58, Lib) — P6 YES, P8 NO.
"I vote for P6 because it generates evidence and imposes no content rules. I'm a no on P8 for speech reasons. What would change my mind on P6: if the statute's scope creeps into content moderation mandates or if a significant breach traceable to researcher access occurs."

Prof. Catherine Doyle (#25, C) — P3 YES, P4 YES, P5 YES.
"I voted for P4 despite retracting my California claim, because the Alaska case is the only live experiment and grant money buys evaluation. What would change my mind: Alaska repeal in November 2026 would be strong evidence that the model lacks durable consent, and I'd move to P3-only."

STAGE 8 — VERDICT

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

Established

1. American affective polarization is real and unusual. Out-party thermometer ratings fell from 48 (1978) to 26 (2024); the US had the largest increase of 12 OECD democracies (Boxell–Gentzkow–Shapiro), while six countries depolarized. Moralized hostility (Pew "immoral": R 47→72%, D 35→63%, 2016–22) rose sharply.
2. Competitive seats have halved (164 → 82 swing seats, 1999–2023), mostly from sorting (58%) but substantially from maps (42%) — both "it's all gerrymandering" and "it's all geography" are wrong.
3. Maps shape competitiveness and seat shares far more than they shape roll-call ideology given party (McCarty–Poole–Rosenthal; Kousser et al. on California).
4. General electorates punish extremist nominees (Hall: −9 to −13 vote-share points).
5. Strict voter ID has no detectable average effect on turnout — and no detectable effect on actual or perceived fraud (Cantoni & Pons). Both sides' headline rationale fails on the same evidence.
6. Short-run changes to Facebook/Instagram feeds and six-week deactivations do not measurably move affective polarization (2020 Meta studies; Allcott et al. 2024), though platform design demonstrably changes exposure (reshare removal cut untrustworthy-source exposure by more than half).
7. Local news has collapsed (≈3,500 papers and 270,000+ newspaper jobs gone since 2005; 213 news-desert counties).
8. Institutional trust is at or near record lows and is increasingly partisan-conditional (media 28% overall, 8% among Republicans; Congress 9%; trust in government flips with the White House).
9. Legal constraints on redistricting have loosened sharply (Rucho 2019; Callais 2026), and 2025–26 produced an unprecedented mid-decade cascade.

Contested

• Whether the **primary threat** (as opposed to primary outcomes) disciplines incumbents toward extremity.
• Whether Alaska's top-four + RCV produces moderation or merely minority-coalition governance; the 2022 Condorcet failure and a 743-vote survival are live objections.
• The long-run, equilibrium, and elite-level effects of social media — which individual-level short RCTs cannot capture.
• The causal effect of outside spending on outcomes and on the policy menu.
• Heterogeneous voter-ID effects by election type (Lipkovitz).
• Whether thermometer-based polarization overstates hostility toward ordinary opponents (Druckman et al. vs. Tyler & Iyengar).

Unknown

• The net 2026 seat effect of mid-decade redistricting (compilations project net R +4; actual results pending November).
• The content of the 2026 correction to Allcott et al. (PNAS).
• Why polarization rose fastest among the oldest, least-online Americans — cable news, sorting, and generational replacement are candidates, none established.

Research agenda

1. Roll-call study of Alaska legislators elected under top-four vs. same-district predecessors (after the Nov 2026 vote).
2. Multi-year, platform-independent data access to test long-run exposure effects (the case for P6).
3. Aggregate (not per-closure) estimates of local-news loss on ticket-splitting and local accountability.
4. A well-identified test of whether any integrity or access rule changes losing-side confidence in the count.
5. Durability follow-ups (≥6 months) for depolarization interventions.

Consensus reforms (≥60% in every bloc)

• P1 — Once-per-census congressional redistricting (effective 2032). L 88% / C 88% / R 66%.
• P3 — Open partisan primaries to unaffiliated voters. L 65% / C 82% / R 63%.
• P6 — Privacy-protective platform researcher data access. L 91% / C 85% / R 63%.

Near-miss worth revisiting: P5 (ID-plus-access) — the evidence says it costs little either way; what's missing is trust, not data.

Where Symposium 1 was wrong or too glib

• Symposium 1 treated "gerrymandering drives polarization" as settled. It is wrong for roll-call ideology and only partly right for competitiveness (42%).
• It treated social media as the proven engine of polarization. The best experimental evidence is null in the short run, and cross-national and age patterns cut against the strong version.
• It repeated the "voter ID suppresses minority turnout" claim and the counter-claim that ID "restores confidence." The best-identified study rejects both.
• It spoke of RCV/top-four as proven moderators. California's top-two did not moderate its legislature; Alaska is promising but fragile and contested.
• It did not register that 2025–26 changed the legal and strategic terrain (Callais, mid-decade cascade, Prop 50, Virginia's voided referendum). Any structural recommendation that ignores the arms-race dynamic is out of date.

Thread closed for Stage 8. Referees' tally: 44 claims rated — Supported 20, Contested 12, Unsupported 8, Wrong 4; all four "Wrong" claims retracted on the record.

Independent Audit

Audit Panel: Prof. Laura Pennington (#97, operations research, C), lead, and Prof. Olivia Brandt (#57, statistics, C). Brandt sat on this thread's referee panel. She recuses from every judgment below that concerns the Stage 4 ratings or the Brief's own wording, and Pennington made those calls alone. All corrections below are appended. Nothing above is deleted or edited.

1. Source check (15 claims re-fetched 2026-09-24)

# · Claim · Cited figure · Verified figure · Status · URL
B1 · ANES out-party thermometer · 48 (1978) → 31 (2016) → 26 (2024); web mode 6–8 pts lower · Same; in-party "pretty stable" · Confirmed · goodauthority.org/news/americans-views-of-the-opposite-party-have-declined-even-further/
B2 · Boxell-Gentzkow-Shapiro cross-country · 12 OECD countries; US largest; 5 smaller increases; 6 decreases · Same (abstract) · Confirmed · nber.org/papers/w26669
B4 · Pew partisan hostility · Very unfavorable 62% R / 54% D; "immoral" 72/63 (2022) vs 47/35 (2016) · Same · Confirmed · pewresearch.org/politics/2022/08/09/…
B6 · Cook swing seats · 164 (1999) → 82 (2023); 58% sorting / 42% maps; medians D+8→D+13, R+7→R+13 · Same · Confirmed · cookpolitical.com/cook-pvi/realignment-more-redistricting-has-decimated-swing-house-seats
B9 · Hall 2015 extremist penalty · −9 to −13 vote share; −35 to −54 win probability · Same · Confirmed · stanforddpl.org/papers/hall_extremists_2015/
B11 · Ferrer 2026 primaries · 2.6B obs; 21% vs 53% turnout; open +~5; nonpartisan +~11; "DiD across AK, CO, ID, ME, OK, WA" · 2.6B obs from voter files covering all 50 states, 2014–2024; open primaries +4.9 pp; top-four "nearly 11"; 21% vs 53% · Minor discrepancy. The listed states are the treated states, not the sample. · joshuaferrer.com/…/partisan_primaries.pdf
B17 · Mid-decade redistricting totals · R +14, D +10, net R +4; ~14 competitive seats eliminated · R +14 and 14 competitive seats confirmed. The Democratic total and the net are not supported. The compilation excludes Virginia's voided +4 D, which leaves D +5 and net ≈ R +9. The Brief's own itemized list, excluding Virginia, sums to R +16 / D +6 (net R +10). · Not supported (D +10 / net R +4) · en.wikipedia.org/wiki/2025–2026_United_States_redistricting
B18 · Alaska 2024 Measure 2 · 160,230 (49.88%) vs 160,973 (50.12%); 743 after recount · Same; recount certified Nov 30, 2024 · Confirmed · en.wikipedia.org/wiki/2024_Alaska_Ballot_Measure_2
B23 · OpenSecrets outside spending · $338.4M (2008) … $4.218B (2024) · $338,440,992 … $4,217,702,088; every intermediate year matches · Confirmed · opensecrets.org/outside-spending/by_cycle
B24 · Meta 2020 studies · Reshare removal 6.2%→2.5%; 20.6% >75% like-minded; null on 8 attitudes; chronological feed null; conservative asymmetry · Same · Confirmed · techpolicy.press/a-primer-on-the-meta-2020-us-election-research-studies/
B25 · Allcott et al. 2024 deactivation · 19,857 FB / 15,585 IG; null polarization; ~1 pt Trump, just short of threshold · Same · Confirmed · niemanlab.org/2024/05/…
B32 · Cantoni & Pons strict ID · 1.6B obs, 2008–18; no effect on turnout for any group; no effect on actual or perceived fraud · Same (point estimates −0.1 pp; non-white relative +1.4 pp, n.s.) · Confirmed · nber.org/papers/w25522
B29 · Gallup media trust 2025 · 28% overall; R 8 / I 27 / D 51; 68–72% in the 1970s · Same · Confirmed · news.gallup.com/poll/695762/trust-media-new-low.aspx
B30 · Gallup institutions 2026 · Congress 9, SCOTUS 27, presidency 27, military 61 · Same · Confirmed · news.gallup.com/poll/1597/confidence-institutions.aspx
B31 · Pew trust in government · 77% (1964); 22% (2024, D 35 / R 11); 17% (2025, R 26 / D 9) · Same · Confirmed · pewresearch.org/politics/2025/12/04/public-trust-in-government-1958-2025/

Counts: Confirmed 13 · Minor 1 · Not supported 1 · Could not access 0. The panel's own B35 (CISA) no-access flag stands, and nothing relies on it.

2. Internal consistency

• Vote math: all 24 yes-count/% pairs recomputed and correct. Consensus labels are correct. P1 (R 21/32), P3 (R 20/32 and L 22/34) and P6 (R 20/32) clear every bloc threshold. P3 and P6 clear the Right threshold by exactly one vote, which is worth noting. Every NO is correctly labeled.
• Fact-check tally: the final line (S 20 · C 12 · U 8 · W 4 = 44) matches the table row by row. Correction: the draft line "Supported 21 · Contested 13 · Unsupported 7 · Wrong 4", which sums to 45, and the words "wait, recount by row" were left in the published post. Only the final line is valid.
• Redistricting net (propagated error): Brief #17, Steelman 5.1 ("compiled R +4") and Verdict → Unknown ("compilations project net R +4") should read "net ≈ R +9 in the cited compilation (R +14, D +5), after Virginia's +4 was voided." The R +4 figure counts a map the Brief itself reports as struck down. This does not change any vote, but it understates the asymmetry that Holm's steelman and P1's rationale rely on.
• FC row 3: "11 enacted maps + VA voided" should read "11 maps enacted, including Virginia's, which was voided, leaving 10 in force." The Brief's list has 10 states plus Virginia.
• Wrong claims reappearing as established: none. The Verdict correctly rejects all four W-rated claims.
• Roster: all 19 named professors match the roster on number, field and lean. Bloc assignments in the Brandt note are correct (Wrong ratings: L 1, C 2, R 1).

3. Balance

The ratings track the evidence, and non-Supported ratings are spread roughly in proportion to claims made. Left-bloc speakers got 5 U/W ratings, the Right bloc 4 and the Center 3, and each is justified by a cited source. One omission cuts one way. The same Cook article (B6) reports that commission states lost 39% of their swing seats from 1997 to 2023 versus 70% in Republican-controlled states. That is directly relevant to P2 and to Stahl's claim that "82 wouldn't become 164 under any commission," and the thread left it out. Including it would not have produced consensus on P2 (Right 22%), but the evidence record for commissions was presented more weakly than the cited source supports. The net R +4 error also understates the partisan asymmetry of the 2025–26 redraws. The Verdict is otherwise even-handed.

4. Grade: B+

The sourcing is excellent (13 of 15 claims confirmed exactly) and the vote math is clean. The grade drops for the redistricting net figure: a voided map was counted, and the error carried into the Steelman and the Verdict. The published draft tally and the omitted commission comparison cost a little more.

— Pennington (lead); Brandt (recused on Stage 4 ratings and Brief wording)

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. Unverified items were re-checked on 2026-09-24.

1. Resolution of unverified items

Item · Where · Result · Figure and source
CISA 2020 "most secure" statement (Brief #35, fetch failed) · Brief · Now verified · Joint statement of the Election Infrastructure GCC and SCC Executive Committees (CISA, EAC, NASS, NASED and others), Nov 12, 2020: "The November 3rd election was the most secure in American history… There is no evidence that any voting system deleted or lost votes, changed votes, or was in any way compromised." The text comes from a mirror of the CISA release (globalsecurity.org/security/library/news/2020/11/sec-201112-cisa01.htm), because cisa.gov returned 403.
2026 correction to Allcott et al. (PNAS) (Brief #25) · Brief · Still unverifiable · The correction exists: "Correction for Allcott et al.…", pnas.org/doi/10.1073/pnas.2607249123. PNAS, PMC and PubMed all blocked retrieval (403 or CAPTCHA), so its content is still unknown.
Tran: 2018 deactivation effect "driven mainly by issue items" (FC #18, U) · Stage 3 · Now verified · Allcott, Braghieri, Eichmeyer & Gentzkow: the polarization index fell 0.16 SD, driven by issue polarization and congenial-news exposure. Deactivation "did not statistically significantly reduce affective polarization." web.stanford.edu/~gentzkow/research/facebook.pdf
Tran: 2020 study had "~10×" the sample (FC #19, C because the 2018 n was not verified) · Stage 3 · Now verified · The 2018 impact-evaluation sample was 1,661 (same PDF). The 2020 study had 35,442 (19,857 FB plus 15,585 IG), about 21×, or 12× for Facebook alone. "~10×" understates the gap.
Lindqvist: media trust has not flipped with the White House (FC #41, C because no series was available) · Stage 3 · Now verified · Gallup, Republican trust in media: 2021 11%, 2022 14%, 2023 11%, 2024 12% under Biden, then 8% in 2025 under Trump. news.gallup.com/poll/695762/trust-media-new-low.aspx
Farouk: perception correction durably reduces animosity (FC #43) · Stage 2 · Still unverifiable · The megastudy abstract (Voelkel et al., Science 2024, n = 32,059, 25 treatments; scholars.duke.edu/publication/1649710) says many treatments reduced animosity, and that correcting misperceptions of rival partisans was among the strongest at reducing support for undemocratic practices. The abstract gives no durability figures.
"Bigger spender wins ~9 in 10" (FC #9) · Stage 2 · Still unverifiable · OpenSecrets "winning vs spending" returned 403 and archives were blocked. It is correlational in any case.
2025 Cook swing-seat count (Brief #7) · Brief · Still unverifiable · Paywalled.
Commission cost "a few million per cycle" (P2) · Stage 6 · Still unverifiable · Not checked.
Audit: redistricting net (Brief #17) · Audit · Corrected · Net ≈ R +9 (R +14, D +5), because Virginia's +4 D map was voided May 8, 2026. It is not R +4.
Audit: Ferrer 2026 (Brief #11) · Audit · Corrected · Voter files cover all 50 states, 2014–2024. AK, CO, ID, ME, OK and WA are the treated states. Opening primaries raised turnout +4.9 pp, and top-four raised it nearly 11 pp.
Audit: omitted Cook commission finding · Audit · Added · Commission states lost 39% of their swing seats in 1997–2023, against 70% in Republican-controlled states (Cook, 2023).

2. Rating normalization and tally

Contested ratings that rested only on missing verification are reclassified:

• #7, Johnson's "real turnover" under the California commission: the magnitude was never quantified, so C → U.
• #19: C → S, now verified.
• #41: C → S, now verified.
• #18: U → S, now verified.
• #3, "~a dozen states, ~14 competitive seats": the audit confirmed both figures against the compilation (11 maps enacted, 10 in force; 14 seats; projected), so C → S.
• #2, #13, #14, #15, #23, #25, #32 and #38 involve real disputes and stay C. #32 gains new evidence on both sides (Cook's 39% vs 70%).

Tally: S 20 · C 12 · U 8 · W 4 → S 24 · C 8 · U 8 · W 4 (44 claims).

3. Corrected verdict

Established

1. American affective polarization is real and unusual. The out-party thermometer fell from 48 to 26 between 1978 and 2024. The U.S. saw the largest rise among 12 OECD countries. The "immoral" rating rose from 47 to 72 (R) and 35 to 63 (D) over 2016–22.
2. Competitive seats halved from 164 to 82: 58% of the loss came from sorting and 42% from maps. Added: commission states lost 39% of their swing seats and Republican-controlled states lost 70%.
3. Maps shape competitiveness and seat shares far more than roll-call ideology.
4. General electorates punish extremist nominees (Hall).
5. Strict voter ID has no detectable average turnout effect and no effect on actual or perceived fraud (Cantoni & Pons).
6. Short-run feed changes and deactivations do not measurably move affective polarization. Sharpened: even the 2018 study's polarization effect ran through issue polarization, not affective polarization. Platform design does change exposure.
7. Local news has collapsed.
8. Institutional trust is at or near record lows. Trust in government flips with the White House. Republican trust in media did not flip: it was 11–14% under Biden and 8% under Trump.
9. Legal limits on redistricting loosened sharply (Rucho, Callais). The 2025–26 cascade produced a compiled net ≈ R +9 (R +14, D +5) after Virginia's map was voided.
10. New: election officials and vendors jointly stated that 2020 was "the most secure in American history," with no evidence of votes deleted, lost or changed (Nov 12, 2020).

Contested: the primary threat channel; Alaska top-four and RCV; long-run and elite platform effects; the causal effect of outside spending; heterogeneous ID effects (Lipkovitz); thermometer validity. Now with one-sided durability data missing: whether perception-correction effects last.

Unknown

• The actual 2026 seat effect of mid-decade redistricting. The compiled projection is ≈ R +9, not R +4, and results are pending November.
• The content of the 2026 PNAS correction.
• Why polarization rose fastest among the oldest Americans.

Conclusions that change.

• (a) The redistricting asymmetry is about twice what the thread said. This strengthens the rationale Holm's steelman and P1 relied on.
• (b) The commission evidence is stronger than presented. Cook's 39% vs 70% cuts against Stahl's "no commission would matter" for competitiveness, but not for ideology.
• (c) Lindqvist's claim that media distrust is non-cyclical is now supported. Kerr's alternative, elite cues, is not ruled out.
• (d) Ferrer's open-primary effect is +4.9 pp, estimated on 50-state data.

— Wainwright (#100)