Appendix E — Ideas in Search of a Home

This appendix is a temporary holding place for ideas that belong in the book but do not yet have a settled location. Each idea should begin as its own second-level section. As the structure develops, ideas can be expanded, combined, or moved into the chapters and appendices where they belong.

For each idea, capture:

the claim or question in one or two sentences;
why it matters to the book;
evidence, examples, or cases worth pursuing; and
one or more possible destinations in the manuscript.

When an idea finds a permanent home, move the material rather than duplicating it here.

E.1 Science as a Principal–Agent Problem

Capital hires scientists because it needs knowledge that it cannot produce or independently verify. The scientist is therefore an agent whose expertise is valuable precisely because the principal—the firm, funder, investor, administrator, or state—cannot fully check the scientist’s work. This creates problems of asymmetric information, monitoring, goal divergence, and trust.

Why it matters: Scientific autonomy is not merely a professional privilege. It is partly a practical response to the inability of capital and management to specify, monitor, and evaluate frontier knowledge production. Attempts to eliminate that autonomy can destroy the capacity the principal intended to purchase.

Evidence or examples to pursue:

industrial research laboratories and their relationship to corporate management;
grantmaking, peer review, and delegated judgment;
managerial metrics used as substitutes for substantive evaluation;
research integrity, fraud, and the limits of internal monitoring;
AI laboratories, where unusually large capital investments meet highly specialized knowledge; and
cases in which principals successfully or unsuccessfully aligned scientific work with organizational goals.

Possible homes: Appendix A, Money; Chapter 4, Building Resilient Institutions; or Appendix C, AI and Capital Investment.

E.2 Fascist Tourists

Fascist tourists is a provisional term for outsiders, generalists, or lower-tier practitioners who enter a scientific domain and impose rigid, surface-level rules on more capable scientists. They police visible signs of correctness—approved procedures, formats, checklists, or methodological rituals—without being able to judge whether those rules matter to the actual scientific problem. The result is cargo-cult rigor: strictness about legible form without corresponding understanding of substance.

Why it matters: When institutions cannot directly evaluate scientific quality, procedural conformity can become a substitute for judgment. This may protect institutions from obvious failure, but it can also allow people with less substantive competence to constrain people with more, regardless of the importance or relevance of the rule being enforced.

Questions and examples to pursue:

Which scientific rules protect real epistemic values, and which persist mainly because they are easy to audit?
When do checklists, reporting standards, preregistration, or standardized methods improve work, and when do they become ritualized obstacles?
How do bureaucracies reward the enforcement of visible compliance over difficult substantive judgment?
Does this pattern become more common when managers cannot distinguish excellent science from merely legible science?
How should institutions prevent arbitrary procedural policing without treating scientific prestige as an exemption from legitimate scrutiny?

E.2.1 Working Note: Jason Arday, Academic Freedom, and Responsibility After a Death

On August 5, 2026, the University of Cambridge announced an investigation into new information about Professor Jason Arday’s qualifications and honorary appointments, alongside ongoing complaints of research misconduct. Arday resigned and was found dead on August 14 after weeks of intense coverage. Police said the death was not believed to be suspicious. The aftermath produced competing demands for scrutiny of Cambridge’s hiring and oversight, the allegations against Arday, and the scale and racial character of the media campaign directed at him. An Associated Press account reported that a communications scholar counted 289 press stories about Arday in 22 days and described calls for a public inquiry into the coverage (Associated Press).

The causal relationship between the scrutiny and Arday’s death has not been established. The working argument is instead about responsibility after a fatal outcome. In any other serious domain, a death following an organization’s activities would trigger examination of the entire chain of events even before anyone knew whether legal or causal responsibility existed. Academic and para-academic actors should not demand an exemption from that elementary duty to investigate consequences.

Academic freedom was supposed to mean: “We did excellent work and came to an unpopular conclusion.”

Academic freedom protects inquiry and unpopular conclusions; it is not immunity from examining methods, proportionality, targeting, or consequences. Legitimate questions about credentials or plagiarism do not by themselves justify an unlimited public campaign. Conversely, the existence of an abusive campaign would not resolve the underlying academic questions. Both the allegations and the machinery of the takedown remain proper subjects of investigation.

The fascist tourist edge case is a person who claims the protections and authority of academic work while rejecting its responsibilities. The work can be shifted into newsletters, social media, podcasts, or fragmented side audiences; broken into individually deniable pieces; and aimed iteratively at members of a disfavored subgroup until an especially vulnerable target appears. This resembles hill climbing through targets rather than inquiry organized around a proportionate scientific question.

Questions to preserve from the thread:

When does scrutiny of an individual become racial opposition research rather than a proportionate investigation of conduct?
Who is responsible for assessing cumulative harm when a campaign is distributed across many formally independent participants?
Can actors invoke academic freedom for para-academic opposition research while disclaiming academic standards of care, relevance, and accountability?
What kind of inquiry should follow a death without presuming either causation or innocence on the part of the surrounding institutions and media ecosystem?
How should legitimate investigation continue without turning one allegedly undeserving Black academic at a time into a public proxy battle over race and institutional merit?

This case is one instance of the wider shift from gaming scientific institutions to gaming the legal, administrative, and media systems surrounding scientists. See Section E.6.

Possible homes: Chapter 4, Building Resilient Institutions; Chapter 5, Communicating Under Pressure; or Appendix B, People and Fields.

E.3 Rent Collection, Human Subjects, and the Last Fields to Automate

Fields that require people as research inputs may be both the most institutionally cumbersome and the last to automate. Human-subjects research depends on recruitment, consent, trust, access, intervention, and observation in the real world. These requirements create bottlenecks around which institutions and intermediaries can collect rents. They also leave a stubborn residue of human labor after AI has automated literature review, analysis, simulation, and writing.

Why it matters: The fields that are hardest to automate may not be the ones with the most sophisticated theories or instruments. They may simply be the ones whose essential inputs remain people. This could produce an inversion in which technically advanced fields automate quickly while noisy, expensive, and organizationally difficult human-facing fields retain more labor and institutional power.

Questions and examples to pursue:

Who controls access to human subjects—universities, hospitals, platforms, schools, governments, panel providers, or community organizations—and what rents can that control support?
Which costs reflect necessary protections for subjects, and which reflect institutional extraction, monopoly access, or administrative self-preservation?
Do clinical trials, field experiments, surveys, interviews, ethnography, education research, and public-health interventions share a common automation bottleneck?
Will AI automate analysis long before it can automate recruitment, consent, trust, treatment, and the measurement of real outcomes?
Can synthetic data or simulated subjects substitute for people, and where would that substitution become scientifically circular or politically dangerous?
Will the last fields to automate become relatively more expensive, lower status, more politically exposed, or more valuable?

Possible homes: Chapter 3, Science in the Age of AI; Appendix B, People and Fields; Appendix C, AI and Capital Investment; or Appendix A, Money.

E.4 Shallow Proxies for Quality

When scientific quality is difficult or expensive to judge directly, institutions substitute shallow proxies: credentials, institutional affiliation, journal prestige, citation counts, grant totals, statistical thresholds, methodological badges, polished presentation, or conformity to an expected format. These signals can contain information, but they are easier to observe—and often easier to game—than the underlying quality they are meant to represent.

Why it matters: Science must allocate attention, money, jobs, and authority under severe uncertainty. Proxies make those decisions administratively possible, but once a proxy becomes a target it can displace the quality it was intended to measure. AI may accelerate this problem by making polish, volume, citation-like language, and procedural conformity extremely cheap to produce.

Questions and examples to pursue:

Which proxies are used for discovery, hiring, promotion, funding, publication, and public credibility?
When does a useful signal become a target, a ritual, or a barrier to unfamiliar work?
Who benefits from a proxy, and who bears the cost of false positives and false negatives?
Which dimensions of scientific quality remain illegible to administrators, funders, reviewers, and the public?
Can multiple imperfect signals discipline one another, or do they merely create a thicker performance of quality?
How does AI change the relative cost of producing the appearance of quality versus quality itself?

Possible homes: Chapter 4, Building Resilient Institutions; Appendix A, Money; Appendix B, People and Fields; or Chapter 3, Science in the Age of AI.

E.5 Neither Side Is Your Friend: Cheerleaders and Crazies

Scientists should not mistake hostility from one political coalition for epistemic reliability in the other. The most visible opponents of science may traffic in conspiracy, denial, or open institutional attack. But many self-described supporters treat science as a partisan symbol rather than a method. Democratic support for science can coexist with superstition, weak scientific literacy, motivated reasoning, and shallow deference to whatever carries the approved expert label.

Cheerleaders and crazies is provisional shorthand for two different hazards. Crazies attack scientific institutions or replace evidence with conspiratorial explanation. Cheerleaders praise science in the abstract, overstate convenient findings, weaponize the phrase “science says,” and expect scientists to validate coalition goals. Neither posture guarantees support for scientific autonomy, uncertainty, correction, or inconvenient results.

Why it matters: Scientists can become politically dependent on supporters who value the authority of science more than its disciplines. Cheerleading may feel safer than hostility, but it can reward overclaiming, suppress internal criticism, and convert scientific credibility into a partisan asset. When findings change or cease to serve the coalition, symbolic support may prove brittle.

Questions and examples to pursue:

How much expressed trust in science reflects scientific understanding, generalized institutional trust, partisan identity, or opposition to another party?
Which superstitious, conspiratorial, or unsupported beliefs coexist with nominally pro-science political identities?
When do supporters tolerate uncertainty, failed replication, dissent, and findings that cut against their policy preferences?
How do slogans such as “believe science” obscure the conditional, adversarial, and self-correcting character of scientific practice?
What forms of coalition-building protect science without making it intellectually subordinate to a coalition?
Where are the asymmetries between political camps, and where would a claim of symmetry become lazy or empirically false?

Possible homes: Chapter 2, Science in a Populist Age; Chapter 5, Communicating Under Pressure; Chapter 6, Building Coalitions for Science; or Appendix D, Voter Beliefs and Attitudes towards Science.

E.6 The Gamification of Science

Science is governed through rules, incentives, rankings, thresholds, and procedures. Those devices make collective knowledge production possible, but they also create games. Participants can optimize what the institution scores while evading the substantive purpose of the score. The result is a widening gap between doing excellent work and becoming skilled at the systems that certify, publicize, fund, or attack it.

Games played inside science:

p-hacking, specification searching, and selective reporting;
hypothesizing after results are known while presenting the result as confirmatory;
venue shopping until a favorable editor or reviewer accepts the work;
salami slicing and optimizing publication counts rather than cumulative understanding;
obfuscating a weak step or shifting a contestable assumption into a more technical, distributed, or opaque part of the analysis;
choosing benchmarks, outcomes, time windows, or comparison sets that flatter the preferred result; and
manufacturing the shallow quality signals described in Section E.4.

These strategies exploit the difference between the epistemic objective and the institution’s observable scoreboard. They can produce work that is formally successful and substantively weak without requiring anyone to fabricate data or state an easily falsified lie.

Games played against scientists:

The fascist tourist variation in Section E.2 treats the accountability environment itself as a competitive game. The objective is not necessarily to refute an opponent’s work. It is to discover a sequence of procedural, reputational, financial, or legal moves that makes the opponent costly to employ, publish, fund, defend, or associate with.

Possible moves include complaint shopping across institutions, strategic misconduct allegations, expansive records requests, threats of litigation, demands directed at employers or funders, platform reports, donor pressure, jurisdiction shopping, and the serial repackaging of an allegation for new audiences. Each mechanism may have a legitimate purpose. The game consists of combining them asymmetrically: the attacker selects cheap moves while the target and the target’s institution must pay the cost of answering every move.

Side audiences reduce the evidentiary burden further. A campaign need not persuade a common public if it can find a smaller audience already disposed to accept the story and capable of imposing costs. See Section E.7.

Lawfare as a scientific meta-game:

Lawfare moves the contest from the truth of a claim to the target’s ability to survive process. A complaint need not prevail to consume time, money, attention, insurance, legal counsel, institutional goodwill, and emotional capacity. Fragmented actors can coordinate effects without accepting collective responsibility, while every response by the target creates new statements and discrepancies that can become material for the next round.

The central problem is not that lawsuits, complaints, investigations, or records requests are inherently illegitimate. Scientific institutions need all of them in some circumstances. The problem is distinguishing accountability from adversarial process abuse when both use the same formal tools.

Questions and examples to pursue:

What are the scientific equivalents of a game’s score, exploit, dominant strategy, and unwritten rule?
When does optimization reveal a badly designed institution rather than a uniquely bad actor?
What features distinguish a good-faith complaint from a campaign whose real product is cumulative process cost?
How should institutions evaluate proportionality, consistency of standards, target selection, escalation, and stopping rules?
Can due process protect targets without shielding fraud, misconduct, or genuine abuse?
Does AI make these games cheaper by automating anomaly searches, dossiers, complaints, legal drafts, audience testing, and distributed pressure?
What institutional designs align winning the game more closely with producing reliable knowledge and correcting real wrongdoing?

Possible homes: Chapter 3, Science in the Age of AI; Chapter 4, Building Resilient Institutions; Chapter 5, Communicating Under Pressure; or Chapter 6, Building Coalitions for Science.

E.7 Side Audiences, Self-Selection, and Prior-Compatible Stories

Fragmented media make it possible to find an audience rather than persuade one. A story does not need enough evidence to convince readers with diverse priors. It needs to reach people whose existing beliefs make the story plausible, emotionally useful, or politically actionable. Those readers then self-select into becoming its audience.

In Bayesian terms, a weak signal can produce confident belief when it reaches people with sufficiently strong priors. The storyteller can therefore optimize audience selection instead of improving the likelihood value of the evidence. Apparent persuasiveness becomes endogenous: the people who encounter, share, subscribe to, or act on the story are disproportionately those for whom it already fits.

Side messaging:

The same campaign can maintain different messages for different audiences. A public-facing version may be cautious and procedural; a newsletter version accusatory; a donor version urgent; a legal version narrowly actionable; and a social-media version memetic or cruel. No single audience sees the whole campaign, and each message can be defended as appropriate to its local context.

Side messaging provides:

audience-specific evidentiary standards and emotional cues;
plausible deniability about the campaign’s strongest implications;
separation between respectable messengers and inflammatory ones;
repeated opportunities to test which framing recruits attention;
insulation from correction issued in another venue; and
the appearance of independent convergence when related claims return from different channels.

Why small audiences can be enough:

The relevant audience may not be the public or even the scientific field. It may be one editor, employer, funder, donor, lawyer, regulator, administrator, or platform moderator. A story can fail as mass persuasion and succeed as targeted activation if a small audience controls a consequential veto point. This changes the objective from establishing consensus to locating someone with both compatible priors and institutional leverage.

Questions and examples to pursue:

How do newsletters, podcasts, group chats, niche publications, and social platforms help stories locate prior-compatible audiences?
When does audience self-selection masquerade as broad evidentiary agreement?
How are different versions of a claim coordinated while remaining formally separate?
Which side audiences possess veto power over employment, funding, publication, reputation, or legal exposure?
Why do corrections fail when they do not travel through the same trusted side channels as the original claim?
How do recommendation systems and AI reduce the cost of discovering receptive audiences and tailoring messages to them?
What would a scientific institution need to monitor in order to see the distributed campaign rather than only the individual message addressed to it?

Possible homes: Chapter 2, Science in a Populist Age; Chapter 3, Science in the Age of AI; Chapter 5, Communicating Under Pressure; Chapter 6, Building Coalitions for Science; or Appendix D, Voter Beliefs and Attitudes towards Science.