The Letterkenski Science-Policy Dossier

Operating Register

Technical accuracy at constitutional weight, plain-language-with-precision diction, dry Manitoba-Canadian gallows-humor closing lines. Underlying premise: science is a product made by labor under management decisions made for documented reasons. Every institutional layer — NIH grant-making, FDA user-fee structures, RELX’s 38% margins, OpenAI’s nonprofit-to-PBC transition, DARPA’s commercialization conveyor — is a corporate-restructuring story dressed in lab coats.

Direct quotations under twenty words; longer source passages rendered as close paraphrase with attribution.

PART 1 — AUTHORITATIVE CITATION CORPUS

1.1 Carl Zimmer

  • She Has Her Mother’s Laugh: “The stubborn inequalities in the United States are not the result of some people living in a physical environment. Their environment is built by social forces, and those forces last for centuries because they are regenerated across the generations.”
  • On heritability: “Among children who grew up in affluent families, the heritability [of intelligence test scores] was about 60 percent. But twins from poorer families showed no greater correlation than other siblings. Their heritability was close to zero.” Use to dismantle “the genes show” rhetoric.
  • Watson-Crick-Franklin episode: prototype labor-extraction-in-science story — deploy whenever a contemporary model paper credits “the data” as if it generated itself.
  • A Planet of Viruses: “every liter of seawater contained up to one hundred billion viruses.” Useful for puncturing pandemic-exceptionalism rhetoric — the planet has always been viral; what changed in 2020 was the surveillance apparatus, not the biology.
  • Life’s Edge: treats the demarcation problem as historical-epistemological artifact — the move to use when LLM enthusiasts conflate “intelligence” with whatever benchmark they’ve just topped.

Operational rule: Zimmer is the workhorse citation when the science must be unimpeachable while you do the political-economy framing. He gives the molecular biology; you supply the question of who paid for the lab.

1.2 Ed Yong

  • I Contain Multitudes: “If there is a core [microbiome], it exists at the level of functions, not organisms.” Model for writing about any complex biological system: not “the gene for X,” not “the microbe for Y,” but the functional architecture and the labor that fills it.
  • “Much of modern medicine is built upon the foundations that antibiotics provide, and those foundations are now crumbling.” Pair with: antibiotic pipeline broken since the 1980s because antibiotics generate single-course revenue rather than chronic-disease revenue.
  • An Immense World: umwelt framing. Use whenever an AI proponent claims the model “perceives” or “reasons” — forces the question of whose umwelt the system operates in.
  • Yong’s pandemic structural moves: open with a researcher or clinician, embed the technical layer, then locate the failure in management decisions.

Operational rule: It is permissible to render judgment, provided the technical substrate is bulletproof.

1.3 Sabine Hossenfelder

  • Lost in Math: “We have not seen a major breakthrough in the foundations of physics for more than four decades. The belief in beauty has become so dogmatic that it now conflicts with scientific objectivity.”
  • Hossenfelder names “cognitive biases, collective action problems, and bad incentives (e.g., incentives to use the same approaches as everyone else, so that you can publish papers and get jobs).”

Operational rule: Citation when puncturing the prestige economy of “fundamental” science. Treat supersymmetry funding the way you treat WeWork’s valuation — a story about which incentive structures sustained the bubble.

1.4 Adam Becker

  • More Everything Forever: the TESCREAL bundle is “the ideology of technological salvation” that “channels attention into abstract future possibilities, thereby avoiding politically uncomfortable questions about present-day redistribution and power structures.”
  • Moore’s Law: “The fate of Moore’s Law is the fate of all exponential trends: they end, just as Moore himself said.”
  • Longtermist sleight: “For a strong longtermist, investing in a Silicon Valley AI company is a more worthwhile humanitarian endeavor than saving lives in the tropics.” Deploy when “AI safety” rhetoric tries to relocate moral attention away from algorithmic harm now.

Operational rule: Spine of bad-faith-technique section. Whenever writing about AI doom, plant a Becker citation in the second or third paragraph; it forecloses the argument that taking Yudkowsky-style claims seriously is sophisticated.

1.5 Steven Strogatz

  • Infinite Powers: “In mathematical modeling, as in all of science, we always have to make choices about what to stress and what to ignore. The art of abstraction lies in knowing what is essential and what is minutia, what is signal and what is noise, what is trend and what is wiggle.”

Operational rule: Tonal model for plain-language-with-technical-precision. When explaining Bayesian methods, regression toward the mean, or association vs. causation, admit the abstraction explicitly and name what is being ignored.

1.6 Naomi Oreskes

  • Merchants of Doubt: “the same individuals surface repeatedly — some of the same figures who have claimed that the science of global warming is ‘not settled’ denied the truth of studies linking smoking to lung cancer, coal smoke to acid rain, and CFCs to the ozone hole.”
  • Strategic objective: opponents of regulation “keep the controversy alive” by spreading doubt long after consensus has been reached. Internal Brown & Williamson memo: “Doubt is our product.”
  • On consensus: “A conclusion becomes established not when a clever person proposes it, or even a group of people begin to discuss it, but when the jury of peers — the community of researchers — reviews the evidence and concludes that it is sufficient to accept the claim.”
  • Science on a Mission: military funding produces military-shaped science — proposition that travels directly to DARPA-to-commercial AI extraction.

Operational rule: Keystone citation for bad-faith-techniques section.

1.7 Paul Offit

  • On aluminum adjuvants: “Various preparations of aluminum salts have been used in vaccines since the late 1930s. So, the safety of aluminum in vaccines has been assessed for more than seventy years.”
  • On rhetorical drift: “anti-vaccine activists today define safe as free from side effects such as autism, learning disabilities, attention deficit disorder, multiple sclerosis, diabetes, strokes, heart attacks, and blood clots — conditions that aren’t caused by vaccines.”
  • Empirical magnitude: “the risk of contracting measles in five- to nine-year-olds whose parents had chosen not to vaccinate them was one hundred and seventy times greater than for vaccinated children” (Salmon et al., Johns Hopkins, 1999).
  • Vaccinated (2007): Hilleman biography, the definitive accessible account of how vaccines are actually made — mid-century industrial bioscience as labor under management.

Operational rule: Substantive immunology citation. Offit is co-inventor of RotaTeq — disclose this in any extended use, then make the meta-point: the conflict-of-interest accusation, applied to a researcher who developed an actual vaccine that prevents rotavirus deaths, illustrates how conflict-of-interest rhetoric gets weaponized to flatten distinctions between paid PhRMA spokespersons and people who invented working products.

1.8 Eula Biss

  • On Immunity: “we cannot immunize our children, or ourselves, against the world.” Vaccination as inherently social act — individual bodies participating in a public-good infrastructure.
  • Documents how middle-class American mothers in the late 2000s and early 2010s organized vaccine refusal through networks mirroring the structure of consumer organic-food activism.

Operational rule: Pair with Offit. Offit for the immunology, Biss for the cultural register. Together they let you write about the childhood vaccine schedule without either flattening it into “trust the experts” or dignifying VAERS-mining as legitimate methodology.

1.9 Nancy Tomes

  • The Gospel of Germs: late-nineteenth-century bacteriology moved from the laboratory into household-hygiene practice — a media-and-marketing history of a scientific concept’s diffusion. The apparatus was built between 1880 and 1955 by specific people for specific reasons, and we can read the records.

1.10 Eric Topol

  • Deep Medicine: “The greatest opportunity offered by AI is not reducing errors or workloads, or even curing cancer: it is the opportunity to restore the precious and time-honored connection and trust — the human touch — between patients and doctors.”
  • Candor: “The field is long on AI promise but very short on real-world, clinical proof of effectiveness.”

Operational rule with caveat: Cite Topol for what AI demonstrably does in medical imaging and pattern recognition; deploy political-economy critique against Topol’s policy implications. His optimism systematically underweights “technochauvinism” — the assumption that economic-institutional arrangements will let AI’s time savings flow to patient care rather than administrative cost extraction.

1.11 Kate Crawford

  • Atlas of AI: “AI is made from vast amounts of natural resources, fuel, and human labor.”
  • “AI seeks to systematize the unsystematizable, formalize the social, and convert an infinitely complex and changing universe into a Linnaean order of machine-readable tables… this is transmuting difference into computable sameness.”
  • “Over and over, we see the ideology of Cartesian dualism in AI: the fantasy that AI systems are disembodied brains that absorb and produce knowledge independently from their creators, infrastructures, and the world at large.”
  • Six-chapter structural template: Earth, Labor, Data, Classification, Affect, State — AI as logistical-extractive system rather than cognitive technology.

Operational rule: Keystone citation for any AI piece. Both registers refuse the marketing voice; both treat infrastructure as the actual subject.

1.12 Public-Health-Policy Citation Pack

  • Daniel Carpenter, Reputation and Power (2010). FDA’s regulatory power derives from accumulated reputation for competence; reputation itself is a political asset. Three powers: directive, gatekeeping, conceptual. Cite when an op-ed treats FDA approval as a binary technical fact rather than a reputation-managed political decision.
  • Marcia Angell, The Truth About the Drug Companies (2004). “The few drugs that are truly innovative have usually been based on taxpayer-supported research done in nonprofit academic medical centers or at the National Institutes of Health.” Actual cost of bringing a new drug to market closer to $100 million than the $800 million industry figure.
  • Sheldon Krimsky: “The prophylactic measures that are taken to prevent conflict of interest in public affairs are considered irrelevant in science precisely because scientists view themselves as participating in a higher calling than that of public officials.”
  • Additional: David Healy, Pharmageddon; Ben Goldacre, Bad Pharma; Vinay Prasad, Ending Medical Reversal; Jonathan Cohn, Sick.

PART 2 — INSTITUTIONAL DOCUMENTARY SUBSTRATE

2.1 NIH Grant-Making

Core mechanisms. R01 — standard investigator-initiated, four to five years, FY2021 total costs averaging ~$570,000/year. R21 — two-year, $275,000 direct-cost exploratory. K-award series supports postdoc-to-PI transition. OER administers across NIH’s 27 institutes; CSR provides peer review.

The doubling, FY1998–FY2003. Senate voted 98–0 in 1997. Appropriations rose from $13.6 billion (FY1998) to $27.1 billion (FY2003), 14–16% annual increases. RPG funding rose from $7.7B to $13.8B. Mean real total costs per RPG rose from ~$530,000 to ~$610,000.

Post-doubling collapse. Real NIH funding peaked in FY2003 and did not match that peak again until ~FY2024. Sequestration March 2013 cut NIH 5%. Real per-grant costs hit nadir of ~$520,000 in FY2013.

FY2025–FY2026 (Bhattacharya/Trump period). Final FY2025 NIH+ARPA-H: $48.495B (0.7% below FY2024). FY2026 request proposes $27.915B — a $19.1B (–40.6%) reduction — eliminating four institutes and consolidating remaining 19 into eight. NIEHS and ARPA-H moved out of NIH. Inflation-adjusted, FY2026 request would put NIH 47.3% below FY2003 peak. Workforce dropped from ~21,000 (early 2024) to 17,300 (late 2025); 1,100 PhDs gone.

Grant terminations. From March 2025, HHS ordered NIH to cancel hundreds of grants flagged for DEI content. By November 2025, terminations exceeded 1,000 grants on Alzheimer’s, HIV, vaccines. November 2025 JAMA Internal Medicine analysis: cuts and freezes halted ~380 clinical trials enrolling 74,000 patients.

Indirect cost rates. Universities negotiate F&A rates of 50–70%. Trump February 2025 attempted 15% cap; litigated and partially blocked. Indirect costs cover building maintenance, IRB administration, animal care, IT, grant management — the lab as physical-and-administrative apparatus.

2.2 NSF

  • Vannevar Bush, Science: The Endless Frontier (July 1945). Postwar template: federal funding of basic research, channeled primarily through universities, peer review, national-security/public-health/economic-prosperity rationales. NSF created 1950.
  • CISE Directorate grew faster than any other directorate over the past two decades. National AI Research Institutes program: 25+ institutes at $20M each over five years. CISE annual budget exceeded $1B by FY2023.
  • Dear Colleague Letters are documents you can read directly to track priority drift.

2.3 DARPA

  • Created 1958 as ARPA in response to Sputnik. Program managers on three- to five-year tours rather than career bureaucrats — structure designed to be insulated from civil-service inertia.
  • ARPANET/Internet. Licklider’s 1963 “Intergalactic Computer Network” memo; BBN built IMPs in 1968; first ARPANET node at UCLA October 1969; TCP/IP deployed January 1, 1983; ARPANET decommissioned 1990; NSF shut down its backbone 1995.
  • GPS. TRANSIT/NAVSAT from Johns Hopkins APL after 1957. GPS launched 1973 as military system; civilian access authorized 1983 after KAL 007.
  • Siri. SRI’s CALO project, 2003–2008, $200M DARPA contract under PAL. SRI spun out Siri Inc.; Apple acquired April 2010; integrated into iOS October 2011. The canonical commercial-extraction pattern: DARPA grant → SRI nonprofit → spinout → tech-giant acquisition.
  • DARPA Grand Challenges (2004, 2005, 2007). Direct precursors to Waymo, Cruise, Argo, Aurora. DARPA paid for basic research and supplied prestige; commercial firms extracted value.
  • Project Maven. Established April 2017. Google joined 2017; March 2018 NYT published internal letter signed by 3,100 Google employees; June 2018 Google announced non-renewal; subsequently published AI Principles. Pentagon distributed contract across Palantir, Microsoft, Anduril, Anthropic, and (2025) OpenAI. 2026 Fortune reporting: employee leverage that worked in 2018 has been systematically eroded — Google decommissioned internal mailing lists and social platforms after Maven specifically to prevent recurrence.

2.4 CDC/FDA/NIH Structural Patterns

  • FDA user-fee structure (PDUFA). PDUFA I (1992). By FY2025, user fees provided 77% of PDUFA program total costs, versus 7% in FY1993. PDUFA reauthorized seven times. Critique: FDA’s budgetary dependence on industry fees “may advantage industry in negotiating favorable policy changes through PDUFA,” including evidentiary-standard adjustments, accelerated-approval pathways, and industry involvement in agency decision-making.
  • CDC reorganization (2022). Walensky’s August 17, 2022 announcement: “For 75 years, CDC and public health have been preparing for COVID-19, and in our big moment, our performance did not reliably meet expectations.” Eliminated 20 offices/divisions/centers; created 16 new entities; expanded Office of the Director’s authority.
  • NIH director succession. Collins (2009–2021) → Tabak (acting) → Bertagnolli (Nov 2023–Jan 2025) → Memoli (acting) → Bhattacharya (April 1, 2025–). On Bhattacharya’s first day, directors of NIAID, NICHD, NIMHD, NINR were placed on administrative leave (offered reassignment to Indian Health Service in Alaska, Montana, Oklahoma); these institutes administered $9 billion.
  • HHS political pressure (2024–2026). RFK Jr. (confirmed Feb 2025) ordered NIH to launch autism study framed around environmental and vaccine factors; explore “regret” among transgender people who underwent hormone treatment; cut recommended childhood vaccine schedule “to match those in Denmark” (a country with 1/57th the U.S. population and universal health care). HHS in early May 2025 cut $2.6B in NIH contracts.

2.5 Scientific Publishing — Standalone Pillar

2.5.1 Elsevier and RELX

1993 merger of Reed International and Elsevier; renamed RELX 2015. Elsevier acquired Pergamon Press (Maxwell’s portfolio) 1991.

The financials. RELX’s STM segment FY2024 revenue £3.05B; adjusted operating profit £1.17B — a 38% margin. By year: FY2019 38%, FY2022 37.8% (€1.2B on €3.26B), FY2023 ~38%, FY2024 38%. For comparison: Apple’s net margin 2024 ~24%, Google’s ~21%. Elsevier publishes ~600,000 articles annually across 2,800+ journals — about 16–17% of global academic-publishing market. Journals division operates at margins exceeding 40%.

The structural critique: “Every time we pay a $3000 article processing charge, only $1800 supports the publishing process, while the remaining $1200 goes directly to Elsevier shareholders.” Researchers write, peer-review, and edit for free; Elsevier provides server infrastructure and brand prestige; public funds underlying research. Annual profit Elsevier extracts exceeds 90% of HHMI’s annual budget.

The “big five” — Elsevier, Springer Nature, Wiley, SAGE, Taylor & Francis — together control approximately half of all scholarly publishing.

2.5.2 Aaron Swartz and JSTOR

September 2010–January 2011: Swartz used a script (keepgrabbing.py) on the MIT network to download ~4.8 million JSTOR articles, ~80% of database. Arrested January 6, 2011. By September 2012 superseding indictment, total 13 felony counts under CFAA and wire-fraud, cumulative maximum $1M fines and 35 years. JSTOR settled civilly and stated no interest in prosecution. MIT did not request prosecution but did not oppose. Ortiz and Heymann pressed forward; offered plea required six-month sentence and guilty pleas on all 13 felonies. Swartz died by suicide January 11, 2013.

Swartz’s 2008 “Guerilla Open Access Manifesto”: “We need to take information, wherever it is stored, make our copies and share them with the world… We need to download scientific journals and upload them to file sharing networks.”

2.5.3 SciHub

Alexandra Elbakyan founded SciHub 2011. Elsevier won $15M judgment in 2017 plus separate $4.8M to ACS. SciHub remains operational. Used routinely by working scientists at well-funded institutions, not only researchers in low-income settings — a fact that punctures the official position that paywalls only burden the global poor.

2.5.4 Open Access

  • BOAI (February 2002). Two strategies: self-archiving (Green OA) and OA journals (Gold OA).
  • PLOS. Founded 2000 by Varmus, Brown, Eisen. PLOS ONE (2006) became largest scientific journal by volume. APCs currently $1,805–$5,300.
  • Plan S (September 4, 2018; effective January 1, 2021). “S” originally for “shock.” Covers ~6% of worldwide articles, including roughly one-third of Nature and Science. From January 2025 cOAlition S no longer financially supports transformative agreements or transformative journals.
  • APC critique. Author-pays Gold OA reproduces rent-extraction structure under a different label. Hybrid journals constitute “double-dipping”: library subscriptions and author APCs both pay for the same content.

2.5.5 Preprint Servers

  • arXiv founded 1991 by Ginsparg. 2.5M+ papers by 2024.
  • bioRxiv launched November 2013 by Inglis and Sever at Cold Spring Harbor. ~Two-thirds of bioRxiv preprints subsequently published in peer-reviewed journals. ~180,000 preprints by late 2022.
  • medRxiv launched June 2019.
  • 2020 inflection. Preprint servers became central to actual practice of science. By January 2022, medRxiv had 50% more papers than January 2020. Also exposed preprint failures: hydroxychloroquine, ivermectin circulated widely.
  • openRxiv. March 11, 2025: ownership of bioRxiv and medRxiv transferred from CSHL to newly formed nonprofit.

2.6 Vaccine Documentary Record

2.6.1 Childhood Schedule

Set by ACIP (CDC): 15 voting members appointed by HHS Secretary; recommendations forwarded to CDC Director.

2.6.2 1986 NCVIA and VICP

Signed Reagan November 14, 1986. Background: 1970s-80s DPT controversy. Substantial jury awards led most vaccine manufacturers to exit market by 1985; only one company still made pertussis vaccine in U.S. by end-1985. VICP: no-fault federal compensation funded by excise tax (originally $0.75) on every dose. U.S. Court of Federal Claims adjudicates; Special Masters as fact-finders. By January 2024: 26,961 petitions (1,413 death, 25,548 injury); 10,443 (39%) compensated; total compensation paid exceeded $4.6B.

2.6.3 VAERS

Established 1990 by FDA and CDC. Passive reporting; reports unverified. CDC explicitly states reports cannot be used to determine causation. Cormack & Grossman (Waterloo, 2022): “VAERS reporting rate cannot be used to infer incidence of death or any other adverse event following vaccination.”

Established 1990. Active surveillance counterpart to VAERS. CDC partners with major integrated health systems covering 12M+ people; data allow rate comparisons (vaccinated vs. unvaccinated cohorts).

2.6.5 Adjuvants

Aluminum salts used as adjuvants since 1926 (Glenny). Mechanism: antigen-aluminum complex enhances phagocytosis by antigen-presenting cells, slows antigen diffusion, activates NLRP3 inflammasome, triggers Th2-skewed adaptive response. Newer licensed adjuvants: MF59 (Fluad); AS01 (Shingrix); AS04 (Cervarix); CpG 1018 (Heplisav-B).

2.6.6 Specific Vaccine Histories

  • Smallpox. Jenner 1796; WHO certified eradication 1980. Variola virus exists today only at CDC and Russian VECTOR Institute.
  • Polio. Salk IPV licensed 1955; Sabin OPV 1961. 1955 Cutter incident — improperly inactivated vaccine causing 40,000 cases of polio — established the modern vaccine-safety regulatory regime.
  • MMR. Measles 1963; mumps 1967 (Hilleman); rubella 1969. Combined 1971.
  • HPV. Gardasil 2006.
  • Varicella. Oka strain, Hilleman/Merck, U.S. licensure 1995.

2.6.7 1976 Swine Flu Episode

Virus resembling 1918 strain isolated at Fort Dix February 1976. Ford administration committed to mass immunization. Campaign began October 1, 1976; suspended December 16, 1976 after detection of Guillain-Barré among recipients. ~40 million vaccinated. Attributable risk: ~1 additional GBS case per 100,000 vaccinees. Federal government paid millions in damages. The canonical example of post-licensure surveillance working: signal detected, program halted, subsequent programs redesigned.

2.6.8 mRNA Platform

Karikó (Penn from 1989) and Weissman (Penn from 1997) collaborated from late 1990s on mRNA-immunogenicity problem. 2005 Immunity paper: substituting pseudouridine for uridine reduced inflammatory signaling. Karikó struggled to obtain NIH funding; position at Penn was downgraded; joined BioNTech as SVP 2013. Moderna incorporated 2010. Pfizer-BioNTech BNT162b2 and Moderna mRNA-1273 reached EUA December 2020. Karikó and Weissman won 2023 Nobel.

2.6.9 Operation Warp Speed

Announced May 15, 2020; ~$18B in pre-purchase commitments and direct development funding. EUAs December 2020. DPA invoked. Distinguish OWS (procurement and logistics) from the underlying scientific work (decades of basic research at NIH, Penn, BioNTech, Moderna).

2.7 University Research-Funding Pipeline

  • Bayh-Dole Act of 1980. Universities, small businesses, nonprofits retain title to inventions made under federally funded research. Pre-1980, only ~25–30 universities had active tech-transfer offices; today essentially every R1 does. Critique (Mowery et al., Ivory Tower and Industrial Innovation): productivity story overstated relative to pre-existing university-industry ties; patent count stagnated since late 1990s; downstream effects on research norms (delayed publication, restricted material transfer, conflict-of-interest proliferation).
  • Foundation patterns. Gates Foundation by 2020 disbursing ~$5B annually — comparable to entire developed-country foreign-aid budgets. “Bill Chill” describes reluctance of grantees to criticize. Structurally accountable only to itself. Preference for vaccine-and-pharmaceutical interventions over broader social determinants reflects managerial-engineering theory of public health. HHMI funds investigators rather than projects. CZI (LLC, not foundation, allowing political activity) launched 2015 with Biohub network.

PART 3 — BAD-FAITH TECHNIQUES

3.1 The AI-Existential-Risk Pivot

Mechanism. Reframe every concrete present harm of AI deployment — algorithmic discrimination in lending and policing, copyright extraction without compensation, displacement of working artists and writers, water and electricity demand of data centers, labor abuses of data-labeling subcontractors in Kenya and the Philippines — as secondary to a hypothetical future “alignment” problem.

Documented examples. 2023 Center for AI Safety statement signed by Altman, Hassabis; 2023 Yudkowsky Time op-ed calling for international airstrikes on rogue data centers; EA channeling of charitable funding toward AI-existential-risk research. Becker: this rhetoric serves “as a kind of moral cover story that allows billionaires to justify vast concentrations of wealth and influence under the guise of working for humanity’s far future good.”

Dismantling. The technique works by abstraction. Counter-move is concretization: name the labor displaced, the watt-hours consumed, the copyrighted training data, the lawsuit list. Closing-line register: the future is always hypothetical; the data center next door is already drinking the aquifer.

3.2 Manufactured Vaccine-Safety Controversy

Mechanism. Recruit a discredited or fringe finding, present it as suppressed truth, use the cherry-picked particular to destabilize the general evidence base.

Wakefield. 1998 Lancet paper, 12 children, Royal Free Hospital. Brian Deer documented: (a) Wakefield was paid more than £400,000 by a UK solicitor preparing litigation against vaccine manufacturers; (b) Wakefield held a patent on a single-measles vaccine alternative; (c) case histories were falsified. Ten of 12 co-authors retracted interpretation 2004. Lancet fully retracted February 2, 2010. GMC struck Wakefield from medical register May 24, 2010.

VAERS misuse. Anti-vaccine activists cite VAERS death-report counts as if causation-confirmed.

Dismantling. Use Offit on aluminum adjuvants and Biss on social construction of vaccine fear. Refuse to fight on the cherry-picked terrain; reframe to the systemic record. Closing-line register: Wakefield kept his patent. The journal kept its publishing fees. The children kept the measles.

3.3 Open-Source / Open-Science Washing

Mechanism. Adopt the language and aesthetic of openness while preserving the closed apparatus.

OpenAI. Founded 2015 as nonprofit “to ensure that artificial general intelligence benefits all of humanity.” 2019: capped-profit subsidiary OpenAI LP. October 2025: restructuring into OpenAI Foundation (nonprofit, retaining 26% equity in ~$130B-valued for-profit) and OpenAI Group PBC (Microsoft 27%, employees and investors 47%). Concurrent: OpenAI deleted “safely” from its mission statement (November 2025 IRS Form 990). The company that called itself “Open” closed model weights in 2023, restructured profit caps in 2023 and again in 2025, and now faces multiple wrongful-death and copyright lawsuits.

Meta. Releases LLaMA weights under licenses with commercial-use carve-outs (over 700M monthly active users). The “open” framing harvests goodwill from the open-source community while preserving Meta’s monopoly on training data, compute, and iteration capacity.

Dismantling. Treat the corporate-form transition the way you’d treat a SPAC merger — read the 8-K, identify who ended up holding equity. Closing register: the only thing OpenAI will reliably open is a new round of fundraising.

3.4 What-AI-Can-Do Begging-the-Question

Mechanism. Cite a narrow capability gain (often genuine) as evidence for an unbounded capability claim. The “AI scientist” press release inflates AlphaFold-style structural prediction into general-purpose autonomous research.

Distinguish carefully. AlphaFold2 (2020) and AlphaFold3 (2024) genuinely solved protein-structure-prediction at the level of single-domain accuracy. Hassabis and Jumper won 2024 Nobel in Chemistry. This is real work. It is also narrowly bounded: AlphaFold predicts static structures from sequences, not protein dynamics, not protein-protein interaction at full complexity, not function.

Dismantling. Get the technical layer right. Cite the Nature paper. Then ask the political-economy question: who owns the model, who paid for the training data, who controls access to the API, what happens when DeepMind decides AlphaFold is no longer free.

3.5 “The Science Is Unsettled”

Mechanism. Invoke residual uncertainty to delay action. “Doubt is our product.”

Dismantling. Distinguish epistemological humility (always appropriate) from manufactured doubt (a specific rhetorical operation funded by specific actors). Cite the Brown & Williamson memo, the API memos, the ExxonMobil internal documents.

3.6 Cherry-Picking Preliminary Studies

Mechanism. Elevate a single preliminary finding while ignoring the systematic-review/meta-analytic record.

Dismantling. Cite Ioannidis’s “Why Most Published Research Findings Are False” (PLOS Medicine 2005) — with corollary: this is not a license for nihilism — it is a structural argument for systematic review and replication infrastructure. The single study is a candidate; the meta-analysis is the verdict.

3.7 In-Vitro / In-Vivo / Clinical-Trial Bait-and-Switch

Mechanism. Finding from cell culture (in vitro) or animal models (in vivo) presented as if it were clinical-trial result.

Examples. Hydroxychloroquine 2020 (in-vitro inhibition of SARS-CoV-2 → clinical-trial failure in RECOVERY and SOLIDARITY). Most “AI-discovered drugs” press releases (computational hits → preclinical → no IND).

Dismantling. Insist on the phase. Deploy a small recurring inset: “Stage of evidence: in vitro / in vivo / Phase 1 / Phase 2 / Phase 3 / regulatory approval.” Forces the rhetorical operation to surface.

Closing Operational Frame

Science is institutionally produced under specific funding and labor conditions, by specific people with specific incentives, and those conditions are documented in primary sources anybody patient enough can read. Bush’s Endless Frontier is online. NIH appropriations are tabulated by CRS. RELX’s annual reports are filed with three stock exchanges. PDUFA reauthorizations are public legislation. The MIT Swartz report is online. Wakefield’s GMC tribunal record is public. OpenAI’s IRS Form 990 disclosures are public. DARPA’s program descriptions are on darpa.mil. AlphaFold’s Nature paper is open access. The 1976 swine flu surveillance data is in Annals of Epidemiology.

The patient-dossier register works because the documents exist. Put them in front of a reader who has been told repeatedly that science is too complicated to interrogate, and demonstrate that the institutional decisions around science are no harder to read than any other 10-K, and considerably more consequential than most.

The records are open. The labor is documented. The margins are 38%. Manitoba is cold, but at least the snow is honest about what it is.