Charles Murray and Steve Sailer — Liberal Social Science Claims, Admissions, and Evidence

Source and preservation record

This is the canonical dossier for the Murray–Sailer–Ash–Barry X exchange and the full source-audited analysis delivered in this chat on 20 October 2026. It was saved at Anthony’s explicit request: “Put in its entirety into a note and add any questions I ask incorporate it into note”.

The four posts were published on 16–18 October 2026. The analysis distinguishes the posts’ claims from analyst-selected studies, empirical results from normative judgments, and descriptive differences from causal explanations. Source details and locators are in the same-note native footnotes.

Preservation and continuing updates

Every paragraph, table row, original inline source link, and the original research receipt is retained below. Study headings and native footnote references were added for navigation and provenance. The full original chat answer is recoverable byte for byte after removing only those additions. The original chat-only receipt describes the state before vault installation; the current installation record is at the end of this note.

Anthony authorized incorporation of subsequent questions and researched answers from this chat into this same note. Record questions verbatim and date them, place the developed answer in the relevant section or a clearly marked addendum, update the linked contents, and retain a brief entry identifying the sections changed. Later explicit instructions can narrow or revoke this maintenance scope.

Book use status — 20 October 2026

Selected substantive passages from all eight supplied books now inform Substantive analysis of the eight supplied books. The addition reconstructs the authors’ arguments, checks consequential primary sources, and records corrections, current evidence, and unresolved questions. This was targeted reading and review, not complete reading or verification of every page and citation. The earlier answer in Whether the books have informed the analysis is retained as a historical status report.

Contents

• Source and preservation record
• Full analysis
• Overall assessment and the four-post exchange
• Admissions, hardship, and predictive validity
• Coleman and school resources
• Prison rehabilitation
• Job training
• Poverty and environmental interventions
• Race, IQ, and causal identification
• Sex differences in occupational interests
• Additional cases and the Head Start qualification
• The broad generalization and final assessment
• Original chat-only research receipt
• Follow-up questions and incorporated answers
• Further analysis that would change the assessment
• Three books to start with
• Optional books for specific questions
• A manageable reading order and comparison method
• Follow-up source and review limits
• Library check for the recommended books
• Books supplied and filed in the library
• Whether the books have informed the analysis
• Substantive analysis of the eight supplied books
• What was read and which editions were used
• Admissions — prediction, opportunity, and a denominator problem
• Genetics and intervention — the real Murray–Harden disagreement
• Racial means, moral equality, and sex differences in interests
• Welfare — incentives, poverty measures, and a revised marriage example
• Rehabilitation — revision, coercion, and the limits of treatment selection
• Additional cases — failed inputs, schoolwide bonuses, and vaccination uptake
• What this changes in the assessment
• Claim status and unresolved checks
• Related notes
• Sources and footnotes
• Vault integration record
• Correction and maintenance history
• Substantive book-analysis review and installation receipt

Full analysis

Overall assessment and the four-post exchange

My assessment is that Murray and Sailer point to a real history of disappointing social programs and overconfident explanations. Sailer’s claim that practically none of what he calls “liberal feel-good nostrums” survived empirical testing goes beyond the evidence supplied. Some examples are strong, some are mixed, and some bundle together questions that require separate answers.

The strongest version of their argument is this: good intentions do not establish that a policy works; persistent differences cannot simply be assumed away; and biological or individual differences deserve investigation alongside social conditions. That position is much better supported than Sailer’s sweeping generalization.

The earlier exchange helps explain how they got there:

• Redmond Barry¹ asked whether admissions should consider circumstances when comparing a wealthy student with a 3.9 GPA and a poor student with a 3.8.
• Hunter Ash² answered that the idea of hardship making a given credential more impressive could be tested by whether such students outperform predictions once in college. He asserted that they do not.
• Steve Sailer³ broadened this into a claim about decades of social science.
• Charles Murray⁴ supplied six research areas as examples.

One qualification matters throughout: Coleman is the only specific study Murray names. The other studies below are evidence I selected to examine his examples, rather than a bibliography he supplied.

Admissions, hardship, and predictive validity

On admissions, Ash has evidence for a narrower claim. A 2025 study by John Friedman and colleagues examined admissions and academic records at participating Ivy Plus colleges. Disadvantaged students with the same SAT/ACT scores generally did not earn higher first-year grades than their more advantaged peers. This supports the tests’ usefulness as measures of current academic preparation and challenges the claim that their predictive meaning systematically understates disadvantaged students’ preparation. Friedman, Sacerdote, Staiger, and Tine, 2025⁵.

That finding does not directly resolve Barry’s comparison of two high-school GPAs. The study concerns standardized scores, selected students who enrolled at particular colleges, and first-year outcomes. Its authors acknowledge that comparisons of grades and tests vary across institutions and outcomes. It also cannot establish what those students would have achieved under equal childhood conditions.

Ash’s proposed test addresses one reason for considering hardship: a credential may underpredict future academic performance. Other reasons include recognizing achievement under obstacles or promoting opportunity and social mobility. Disadvantage can also continue after enrollment through financial pressure or family responsibilities.

A college could therefore accept the predictive finding and still decide that circumstances should count. The empirical result narrows one argument for contextual admissions; the value judgment about admissions remains open.

Coleman and school resources

The Coleman Report challenged a simple account of educational inequality. Its 1966 survey found that conventional facilities and curriculum differences explained relatively little variation in measured achievement, while background and student-body characteristics mattered substantially. Yet it also described relationships with teacher quality and greater sensitivity to school quality among minority pupils. Its own findings resist reducing it to a claim that schools do not matter. Original Coleman Report, especially its summary discussion⁶, official ERIC record⁷.

The central methodological issue is that an observational comparison of existing schools differs from estimating what happens when a policy changes school resources. Suppose struggling districts receive extra funding: they may still have low achievement afterward, even if the funding helped. Comparing funding and achievement without resolving that problem can miss a beneficial effect. This is an illustration of the statistical issue, rather than a claim about every district.

Later research uses changes induced by school-finance reforms to improve causal identification. Jackson, Johnson, and Persico’s 2016 study found benefits to educational attainment, adult wages, and adult poverty. Jackson and Mackevicius’s 2024 synthesis also found positive average effects of spending increases. These studies have assumptions and variation across settings; they establish neither a guarantee that every expenditure works nor general irrelevance of funding. Jackson, Johnson, and Persico’s final paper⁸, Jackson and Mackevicius, 2024⁹.

My assessment: Coleman is a serious challenge to the expectation that equalizing a few measured school inputs would readily equalize achievement. It does not establish a causal zero effect of school resources. Family background can strongly predict outcomes while particular school interventions still produce worthwhile improvements.

Prison rehabilitation

Prison rehabilitation supplies a genuine history of disappointment. Robert Martinson’s 1974 review drew on 231 evaluations of programs attempted between 1945 and 1967 and reached a broadly pessimistic conclusion, with important qualifications. It is a plausible example of the literature Murray has in mind, although he did not name it. Martinson, “What Works?—Questions and Answers About Prison Reform”¹⁰.

Later evidence changes the assessment. A 2018 meta-analysis by Bozick and colleagues found lower reoffending among prisoners receiving education. In its stronger-study pool, the estimated odds ratio was 0.72: 28% lower odds of recidivism. Odds differ from probabilities, so this should not be reported as a 28% reduction in a person’s risk. Bozick and colleagues, 2018¹¹.

This is encouraging evidence with limits. Much comes from studies that adjust for differences between participants rather than randomly assigning them; the randomized evidence base is small. Evidence of improved employment was less clear.

The defensible conclusion is that some programs reduce reoffending, with effects depending on the treatment and evidence quality. The history supports skepticism about promises, while a blanket impossibility claim goes too far. A reduction in reoffending can be valuable even when most participants remain at substantial risk.

Job training

Job training offers especially clear examples on both sides. The national randomized Job Corps evaluation followed applicants for roughly 20 years. It did not demonstrate sustained employment or earnings gains for the sample overall, although older entrants had some favorable outcomes. This is substantial evidence against assuming that intensive training automatically produces lasting economic advancement. Credentials, program completion, and short-term gains are insufficient substitutes for that endpoint. Department of Labor’s 20-year Job Corps evaluation¹².

Year Up supplies a strong counterexample. It combines training, employer internships, and support for selected young adult applicants. Its randomized evaluation found an earnings increase of $1,895 per quarter in quarters 23–24 after assignment, roughly six years later. That was approximately 27.5% above the control group’s mean; the report followed outcomes through seven years. Year Up long-term evaluation, particularly Exhibit 2-1¹³.

Year Up does not show that a generic training class would have the same effect. Eligibility, employer connections, implementation, and the intervention bundle matter. But its result makes universal pessimism about training untenable.

The useful question is which model works for which participants, for how long, and at what cost. “Job training works” and “job training fails” are both too coarse to guide a decision.

Poverty and environmental interventions

Poverty requires distinguishing several objectives. Helping a family consume more today, raising an adult’s own future earnings, improving a child’s adult prospects, and reducing the national poverty rate are separate achievements. A program may accomplish one without accomplishing all four.

The Moving to Opportunity experiment illustrates why timing matters. Families were randomly offered vouchers enabling moves out of high-poverty neighborhoods. Children younger than 13 at assignment had better adult earnings and college-attendance outcomes. The results were much less encouraging for adolescents, and adult employment and earnings findings were disappointing. Chetty, Hendren, and Katz’s published study¹⁴.

This is evidence that changing childhood environments can change economic outcomes. It does not reveal the complete cause of poverty or isolate every feature of the neighborhood intervention. It also shows why a failed adult intervention cannot establish that childhood conditions were irrelevant.

Another example is childhood access to food stamps. Research using the program’s historical county rollout found improved adult metabolic health and greater economic self-sufficiency among women. This used a policy rollout rather than random assignment, so its causal interpretation depends on the design’s assumptions. It nevertheless provides evidence of benefits beyond immediate consumption. Hoynes, Schanzenbach, and Almond, published in 2016¹⁵.

My assessment: Murray’s reference to “causes of poverty” is too unspecified to receive a single verdict. Evidence that individual characteristics predict poverty, evidence that some programs disappoint, and evidence that childhood conditions have causal effects can all coexist. A serious explanation must determine how those factors interact.

Race, IQ, and causal identification

On race and IQ, several distinct propositions must remain separate:

1. Cognitive tests can measure meaningful differences and predict important outcomes.
2. U.S. samples have documented average Black–White test-score differences.
3. Genetic differences contribute substantially to variation among individuals in studied populations.
4. Genetic differences explain some particular share of an average gap between racial groups.

The first three do not automatically establish the fourth. Nisbett and colleagues review the historical U.S. group-difference evidence; Deary, Cox, and Hill describe the substantial measurement and genetic evidence concerning individual differences. Nisbett and colleagues, 2012¹⁶, Deary, Cox, and Hill, 2022¹⁷.

Heritability concerns variation in a particular population and environment. It does not tell us what percentage of a difference between two populations is genetic. Schraiber and Edge’s 2024 mathematical analysis explains why information about variation within groups cannot, by itself, separate between-group differences into genetic and environmental components. Schraiber and Edge, 2024¹⁸.

As a hypothetical illustration, imagine two gardens planted with the same mixture of seeds but receiving different nutrients. Genetic variation could explain much of the height variation within each garden while the average difference between gardens comes from the soil. This illustrates the inference problem; it does not establish the cause of a human racial score gap.

Environmental changes can also shift population test scores. Bratsberg and Rogeberg’s Norwegian sibling study found that both rising scores and their later reversal could be recovered within families, supporting an environmental explanation of those cohort changes. That finding does not identify the cause of a U.S. racial gap. Bratsberg and Rogeberg, 2018¹⁹.

My assessment: dismissing intelligence measurement or genetic influences on individual variation is poorly supported. Using those established findings to announce a settled genetic explanation of racial mean differences is also unwarranted. Such a causal explanation requires additional evidence capable of distinguishing the alternatives.

Murray’s tweet itself names the topic without stating that the gaps are genetic. I would not attribute a stronger assertion to that wording.

Sex differences in occupational interests

Sex differences in occupational interests are one of Murray’s stronger examples, especially against an exclusively social explanation. Su, Rounds, and Armstrong’s 2009 synthesis included more than 500,000 respondents and found a large average difference along a things–people interest dimension: men tended toward things-oriented activities and women toward people-oriented activities. The standardized mean difference was about 0.93, nearly one standard deviation. These are average differences with overlapping individual distributions. Su, Rounds, and Armstrong, 2009²⁰.

There is also evidence for a biological contribution. Beltz, Swanson, and Berenbaum studied 125 participants with congenital adrenal hyperplasia or their unaffected siblings. Females with the condition, involving atypically high prenatal androgen exposure, showed more things-oriented interests than unaffected females, with scores between the female and male comparison-group averages. The association with indicators of androgen exposure strengthens the hormonal interpretation. Beltz, Swanson, and Berenbaum, 2011²¹.

That is more informative about a possible mechanism than simply observing different occupations. It remains an imperfect natural experiment: the clinical groups can differ in other ways, the sample is small, and the study cannot estimate what fraction of the general population interest difference is hormonal. The authors also emphasize socialization, overlapping interests, and the multiple determinants of career choice.

A 2025 study of 109,460 online respondents across 57 countries adds nuance to the gender-equality argument. Some interest differences were larger in more gender-equal countries; others were smaller. This challenges the expectation that greater equality uniformly erases interest differences. But English-language online volunteers are not national probability samples, and correlations between countries do not identify biological causation. Du, Wee, and Rounds, 2025²².

My assessment: meaningful average interest differences and a partial biological contribution deserve substantive recognition. Explaining a particular occupational disparity still requires evidence about interests, abilities, incentives, caregiving, access, and discrimination. Neither an observed disparity nor a biological contribution settles that whole question.

Additional cases and the Head Start qualification

Several additional cases help test the argument more broadly. These are illustrative cases selected for informative evidence, rather than a representative sample from which to calculate an ideological success rate.

Head Start deserves particular caution. Controls could access other preschool services, and children in the younger comparison cohort later gained access to Head Start. Follow-up attrition also matters. The experiment estimates the added benefit of the studied offer relative to those alternatives, while the historical adult-outcome studies concern different cohorts and comparisons. Treating either side as a complete verdict on all preschool programs would overstate it. Head Start follow-up’s methods and findings²³.

The broad generalization and final assessment

The broad problem with Sailer’s claim is its undefined universe. Which propositions qualify as “liberal feel-good nostrums”? What counts as a test? Does failure mean no benefit, insufficient benefit for the cost, or failure to eliminate a disparity? Those standards yield different judgments. Selecting failed examples after observing their results cannot establish how common failure is across an unspecified class of claims.

His 1972 remark likewise supplies a personal historical impression rather than a demonstrated date when the whole issue was settled. Coleman predates it; much of the decisive evidence reviewed here comes later. Martinson’s 1974 publication drew on earlier research, so publication dates alone do not tell us everything about what was known in 1972. Establishing Sailer’s broader historical assertion would require a defined proposition and contemporary evidence.

There is also a distinction between a favored intervention failing and the scientists evaluating it failing. Researchers who rigorously document disappointing outcomes are performing an essential scientific function. Sailer’s claim about the investigators’ politics would itself need evidence; their field or research topic cannot establish it.

I would retain the narrower lesson: moral conviction, spending, and attractive mechanisms need to face evidence about actual outcomes. I would reject the inference that this selected record establishes near-universal failure across Sailer’s undefined category of social-scientific claims. The practical standard is to specify the claim, intervention, population, comparison, outcome, follow-up, and cost—and apply that same standard to successes and failures.

Original chat-only research receipt

Research receipt: Chat only; no vault files or graph connections changed. Five specialist roles completed: intake auditor, claim mapper, primary-source researcher, adversarial reviewer, and final auditor. Intake passed; final audit passed with disclosed limits. Attribution was checked against my direct browser capture, without a second independent X capture. Coleman’s full report was only partially inspected; some supporting sources were available through abstracts or indexed text. Arithmetic and counterargument checks passed. Specialists followed read-only instructions under your authorized permission exception; the parent session was not technically restricted to read-only. No blind replication was performed. Protocol 1.2; reviewed October 20, 2026; research and correction record retained in this chat.

Follow-up questions and incorporated answers

Further analysis that would change the assessment

Question from Anthony, 20 October 2026: “any further analysis need to be done? books that might give more information?”

Yes. Some targeted follow-up would make the assessment more complete, especially on policy effectiveness and admissions. The evidence already examined is sufficient to reject treating the selected cases as proof of near-universal failure across Sailer’s undefined category. Further investigation could change judgments about particular programs, mechanisms, or historical interpretations. A broader verdict needs a defined set of claims and a defensible way of selecting evidence.

My priorities would be these five questions. The first two offer the most immediate practical value; the others deepen particular parts of Murray’s argument.

1. How often do clearly defined interventions succeed, for whom, and at what cost? This is the largest gap in Sailer’s generalization. A collection of failures cannot establish an overall failure rate, and a collection of successes cannot establish an overall success rate. Choose one manageable area, such as U.S. employment training; define the eligible programs, study quality, outcomes, follow-up periods, and success criteria before classifying the results. Include favorable, null, and harmful findings, effect sizes and uncertainty, costs, and evidence about implementation at scale. Consider whether participants’ gains reflect gains in total employment or displacement of other workers.

The existing Job Corps and Year Up findings make this a useful starting point: different interventions can produce different long-term results. They were evaluated in different populations and settings, however, so their contrast is not a randomized head-to-head comparison. A good audit would explain which features might account for the difference and whether the evidence actually identifies those features. Even a representative employment-training audit would support conclusions about that defined domain, not a success percentage for all “liberal” ideas.¹²¹³

2. What does considering hardship in admissions actually change? Separate three questions: how well grades and tests predict college performance; how much admission to a particular institution causally benefits a student; and which goals should govern selection when places are limited. Ash’s proposed test primarily addresses the first question. A disadvantaged student could earn the same grades as an equally scoring affluent student and still benefit substantially from the opportunity. Conversely, an individual benefit does not by itself establish the fairest or most productive allocation of a fixed number of places.

The next analysis should therefore examine contextual-admissions policies, enrollment, graduation, later outcomes, institutional support, and effects on other applicants. It should distinguish comparisons among enrolled students from credible evidence about what happens because a student is admitted. The Ivy Plus score-calibration study is relevant evidence about preparation, but it does not settle all of these policy questions.⁵

3. Which evidence can identify the causes of racial mean differences in cognitive scores? Individual differences, the heritability of variation within a population, and the causes of a difference between populations are separate quantities. Further work should assess what within-family, ancestry, admixture, and other genetic designs can establish while addressing environmental confounding, population structure, and whether genetic prediction transfers across populations. Educational attainment and cognitive-test performance also need to remain distinct outcomes.

The central question is what each design identifies, including its assumptions. More heritability estimates or ancestry associations alone do not settle the causal share of a particular racial score gap. An unresolved quantity may remain unresolved after extensive investigation. That uncertainty does not establish a zero genetic contribution either.¹⁷¹⁸

4. How much do measured interests explain actual occupational differences? The existing evidence supports average differences in interests and some biological contribution. The next step is to follow the pathway from prenatal influences and later social experience to interests, education, hiring, working conditions, caregiving, retention, and career choice. Measure how much each mechanism explains a specified outcome in a specified population.

This would strengthen the analysis because the existence of an average interest difference does not establish how much of a particular occupational disparity it explains. Cross-country correlations also do not, by themselves, isolate innate causes. A useful study would allow for biological and social mechanisms to interact rather than forcing the entire disparity into one explanation.²⁰²¹²²

5. What did Coleman and the pre-1972 literature actually establish at the time? A fuller reading of the original Coleman report, its qualifications, and contemporary responses would improve the historical audit. Ask what was known then, which claims the report actually challenged, and how later retellings broadened its conclusions. Martinson’s 1974 publication drew on earlier studies; its publication date alone cannot establish either that its findings were unavailable in 1972 or that the whole subject had been settled by then.

This is primarily a historical precision task. The present analysis already distinguishes an observational finding about measured school resources from evidence that changing particular resources has no causal effect. Completing the original-document audit could refine that account without making the old report a substitute for later causal research.⁶⁷¹⁰

Three books to start with

My recommended core route is Andrew Leigh’s Randomistas, followed by Charles Murray’s Human Diversity alongside Kathryn Paige Harden’s The Genetic Lottery. These serve different purposes: understanding policy evaluation, reconstructing Murray’s fuller argument, and examining how genetic evidence relates to claims about equality and policy. This is a recommendation about reading value, not certification that every claim in any of the books is correct.

Human Diversity is already marked Planned, not completed, in the live and is unchecked on the . This recommendation fits that existing plan; it does not create another reading commitment.

The reason to pair Murray and Harden is the comparison of empirical claims, causal inference, and political reasoning. Acknowledging genetic influences does not uniquely determine a welfare policy or theory of justice. Likewise, preferring an egalitarian outcome does not determine whether an empirical claim is true. Where the books address different populations, traits, or questions, do not force them into a direct disagreement.

For a selective first pass through Harden, the verified contents identify chapters 4–7 as the sequence covering ancestry and race, life chances, causal reasoning, and mechanisms. Chapters 8–9 then connect possible alternative environments and the study of nurture. These are navigation suggestions based on the publisher-hosted contents and available opening excerpts, not a claim that those chapters have been read in full.³¹

Optional short foundation: If intelligence testing and heritability are unfamiliar, put Ian J. Deary, Intelligence: A Very Short Introduction, second edition (2020) before the Murray–Harden pair. Oxford’s contents include genetic and environmental contributions, school and work, changes across generations, and psychologists’ areas of agreement. It is a compact orientation to intelligence research; particular disputes still require the underlying studies.³²

Optional books for specific questions

These are alternatives for a narrower interest, rather than four additional required books.

• U.S. welfare and incentives — Charles Murray, Losing Ground: American Social Policy, 1950–1980 (first published 1984; later reissues). Read this to understand Murray’s historical welfare argument more directly. It is a sustained case that policies can change incentives and produce unintended consequences. Treat its historical scope and causal claims as matters to examine against the evidence, and check later research before applying the argument to current programs.³³
• Admissions and the purpose of higher education — Nicholas Lemann, Higher Admissions: The Rise, Decline, and Return of Standardized Testing (2024). Useful for understanding testing, meritocracy, and the institutional choices behind the original exchange. It broadens the question from selecting a small number of elite students to the opportunities the higher-education system provides. Its historical and normative argument does not replace current evidence on test validity or the causal benefits of particular admissions policies.³⁴
• Prison rehabilitation — Francis T. Cullen and Karen Gilbert, Reaffirming Rehabilitation, second edition (2012). Useful for the history of the rehabilitative ideal, the “nothing works” debate, and the affirmative case for rehabilitation. It is an argued position with an older evidence cutoff; pair it with later systematic reviews and program-specific evaluations.³⁵
• Poverty and practical interventions — Abhijit V. Banerjee and Esther Duflo, Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty, revised edition (2025). A strong choice if poverty is the main interest: it examines how the lives and choices of poor people interact with particular interventions. The verified 2025 revision adds material beyond the 2011 original. Keep the developing-country settings distinct from claims about U.S. welfare, Job Corps, or genetic causes of differences.³⁶

Your existing A Conflict of Visions and Social Justice Fallacies notes can provide conceptual companions on constraints, incentives, and equality arguments. Those notes are not substitutes for checking the empirical studies relevant to a particular claim. Their existence does not establish that either book has been finished.

A manageable reading order and comparison method

Start with Randomistas. Then use the already planned Human Diversity and read The Genetic Lottery alongside it. Add Deary first if a short psychometric foundation would help. Choose one topical book afterward only if its subject becomes the focus. If poverty is the main interest, the revised Poor Economics can take priority over the genetics pair.

For Murray and Harden, keep a single comparison page in Goodnotes or Obsidian and ask the same four questions of each substantial argument:

1. What is the empirical claim? Specify the trait, population, outcome, and comparison.
2. What does the evidence identify? Distinguish a measured difference, a prediction, and a causal explanation.
3. Which additional value judgment supports the policy conclusion? A factual result often leaves several defensible choices open.
4. What evidence would change the conclusion? Record a plausible contrary finding rather than only supporting quotations.

The highest-value further research would be a systematic audit of one policy domain and a closer analysis of the admissions questions. The reading route builds the tools to assess those questions and the biological claims without requiring all eight books at once. The suggested investigations remain proposed next steps; this follow-up does not claim they have already been completed.

Follow-up source and review limits

Book identities, editions, and subject coverage were checked against publisher pages, publisher-hosted book records, and an author university page. Where available, contents and opening excerpts were inspected. The books were not all read in full, and their bibliographies were not independently audited for this recommendation. Publisher descriptions establish scope and metadata, not the truth of the books’ arguments. The priority order is the coordinator’s judgment, informed by an independent adversarial review.

The standing source-access and attribution limits of the original analysis remain in its receipt and footnotes. This addendum proposes research priorities; it does not silently revise the original findings or report new causal estimates. Reading-list records, book statuses, and existing book notes were inspected but not changed.

Library check for the recommended books

Question from Anthony, 20 October 2026: “Heres my library, if the books you’re looking for might already be in my library, you can give me a list of the ones I don’t have”

Library supplied: `

Earlier search result, before the eight PDFs were supplied: I could not find any of the eight specifically recommended titles in the supplied library. PDFs of both Sowell companion books discussed above are present. This is a list of titles not found in this library check, rather than proof that Anthony does not own a copy elsewhere or inside an uninspected archive.³⁷

Updated availability after the earlier search

Anthony subsequently supplied all eight titles. They have been filed and verified; see Books supplied and filed in the library for the current locations and edition details. The supplied Poor Economics is the 2011 original; access to the 2025 revision is not established by that file.

Core recommendations not found in that earlier search:

1. Andrew Leigh — Randomistas: How Radical Researchers Are Changing Our World (2018). First priority for understanding policy evaluation.²⁹
2. Charles Murray — Human Diversity: The Biology of Gender, Race, and Class (2020). Already planned in Obsidian; no matching library file was found. A planned reading record does not establish possession of the book.³⁰
3. Kathryn Paige Harden — The Genetic Lottery: Why DNA Matters for Social Equality (2021). Companion to Murray for comparing scientific claims and political reasoning.³¹

Optional foundation not found:

• Ian J. Deary — Intelligence: A Very Short Introduction, second edition (2020). No matching title or author file was found, including an older-edition match.³²

Topic-specific alternatives not found:

• Charles Murray — Losing Ground: American Social Policy, 1950–1980 (first published 1984).³³
• Nicholas Lemann — Higher Admissions: The Rise, Decline, and Return of Standardized Testing (2024).³⁴
• Francis T. Cullen and Karen Gilbert — Reaffirming Rehabilitation, second edition (2012).³⁵
• Abhijit V. Banerjee and Esther Duflo — Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty, revised edition (2025). No matching copy of the original edition was found either.³⁶

Companion books already present as PDFs:

• Thomas Sowell — A Conflict of Visions. Files found at `
• Thomas Sowell — Social Justice Fallacies. File found at `

The three Sowell files exist, have nonzero sizes, and begin with the PDF signature. Their presence does not establish that either book has been read, and this check did not audit every page or determine the editions of the copies.

Scope and limits: The completed recursive filename inventory contained 107,345 file paths, not 107,345 books. Title and author searches included punctuation and capitalization variants. The library’s Zotero database and the archived Calibre catalog were also searched for the target titles; neither yielded a match. The archive catalog was used as a secondary cross-check rather than evidence that archived records describe active copies. Filesystem traversal completed without reported errors. Archive contents and unrecognizably named files were not exhaustively inspected. The library and reading records were not modified.

The reading priorities above remain unchanged. If obtaining only one new book initially, Randomistas remains the first recommendation; the two Sowell books are available as existing conceptual companions.

Books supplied and filed in the library

Request from Anthony, 20 October 2026: “Put all the books in the main folder of Library for you to use, and put them in the correct folders in my library”

Current availability: All eight recommended titles were subsequently supplied as PDFs in `

The filenames now identify the title, author, and edition or year where verified. Filing followed the library’s existing categories, with no new folder hierarchy. The title and copyright pages were inspected for identity and edition details; this does not claim that all eight books have been read or their scientific arguments audited.

Randomistas — Andrew Leigh

• Current file: `
• Supplied edition: 2018.

Human Diversity — Charles Murray

• Current file: `
• Supplied edition: 2020.

The Genetic Lottery — Kathryn Paige Harden

• Current file: `
• Supplied edition: 2021.

Intelligence: A Very Short Introduction — Ian J. Deary

• Current file: `
• Supplied edition: Second edition, 2020.

Losing Ground — Charles Murray

• Current file: `
• Supplied edition: Original 1984; supplied edition includes 2015 introduction.

Higher Admissions — Nicholas Lemann

• Current file: `
• Supplied edition: 2024.

Reaffirming Rehabilitation — Francis T. Cullen and Karen E. Gilbert

• Current file: `
• Supplied edition: 30th anniversary edition; copyright 2013, Routledge reissue 2015.

Poor Economics — Abhijit V. Banerjee and Esther Duflo

• Current file: `
• Supplied edition: 2011 original; not the 2025 revision.

Edition and attribution details:

• Poor Economics is the 2011 original, confirmed by the copyright page and eISBN 9781610390408. It is available for use, but the supplied file is not the revised 2025 edition discussed in the recommendations.
• Deary’s Intelligence is the second edition, 2020, confirmed on its title and copyright pages.
• Reaffirming Rehabilitation is the 30th anniversary edition by Francis T. Cullen and Karen E. Gilbert. Its copyright page identifies Anderson publication/copyright in 2013 and Routledge publication in 2015; its ISBN 9781455731305 and DOI 10.4324/9781315721361 match the recommended work’s identifiers. These actual-copy dates are retained separately from the publisher’s earlier release listing.
• Losing Ground contains an introduction explicitly dated 2015, in addition to the original 1984 text.
• The Higher Admissions title page credits Nicholas Lemann. Its original filename and embedded author metadata named Marvin Krislov, who contributes a commentary. The filename now credits Lemann. The embedded PDF metadata was left intact because the filing operation preserved each PDF’s contents exactly.

Verification: All eight destination files exist and match their original byte sizes and SHA-256 hashes. None of the eight source files remains in the main Library folder. No destination was overwritten. The existing reading plans and reading statuses were not changed by this filing request.

Whether the books have informed the analysis

Question from Anthony, 20 October 2026: “Did you use the books for further analysis and understanding for the note?”

Historical response, before the substantive reading requested below:

Answer: No, not for substantive further analysis yet. The original analytical conclusions still rest on the studies and reports researched earlier. The reading recommendations used publisher descriptions, contents, and available opening excerpts. After Anthony supplied the eight PDFs, the coordinator inspected title pages, copyright pages, contents, and metadata to verify identity, authorship, editions, and file locations, then organized the files. That filing work did not complete an examination of their substantive arguments or supporting evidence.³⁸

The coordinator has not yet used the supplied books to reconstruct their relevant chapter arguments, compare their substantive disagreements, trace their key supporting citations, or revise the dossier’s empirical assessments on that basis. The books are now available locally; their availability and bibliographic verification must not be described as completed book-based analysis.

A substantive book-reading and synthesis pass remains pending. It would require reading the relevant sections, distinguishing each author’s claims from the evidence supporting them, checking consequential citations against the underlying research, and incorporating any additions or corrections with precise page references. This status clarification adds no new substantive finding and does not claim that work has been performed.

Subsequent update, 20 October 2026: Anthony requested “do substantive further analysis”. That targeted reading and synthesis has now been completed and incorporated in Substantive analysis of the eight supplied books, with actual passage coverage, primary-source checks, corrections, and access limits. The preceding “No” answer records the earlier state.

Substantive analysis of the eight supplied books

Request from Anthony, 20 October 2026: “do substantive further analysis”

Selected substantive passages from all eight supplied books have now been read and used in this analysis. This changes the earlier answer about book use. It is a targeted reading and source audit, not a claim to have read every page or verified every citation in all eight books. The original analysis above remains preserved; this addition develops and, where necessary, qualifies it.

The main conclusion survives, but the reasons become more precise. Murray makes a stronger case for expecting limits, heterogeneous responses, and disappointing interventions than for treating particular failures as evidence that environmental causes are generally unimportant. Harden accepts genetic influences and takes failed interventions seriously. Lemann acknowledges the predictive value of tests while questioning what selective admissions should accomplish. The books therefore reveal disagreements about causal inference, institutional purpose, and acceptable tradeoffs that the short exchange compresses together.

This pass also found a confirmed sample-description error in Human Diversity, an admissions statistic unsupported by its cited table, a substantial update to Losing Ground’s marital-dissolution argument, and qualifications to the affirmative rehabilitation account. These checks apply the same evidentiary standard across authors.

What was read and which editions were used

All locators in this addition use one-based PDF viewer pages unless a printed journal page is explicitly given. Several files are reflowed conversions, so their page numbers differ from printed editions. The full current library paths are recorded in Books supplied and filed in the library. The coverage below lists the passages used, rather than every preliminary page or search hit inspected.

These are Anthony’s supplied editions. Poor Economics is the 2011 original, rather than the revised 2025 edition initially recommended. The rehabilitation book separates its original 1982 argument from its later anniversary material. Lemann’s own chapters are distinguished from the Lewis–Cantor introduction and the other contributors’ commentaries. A book’s account establishes what its author argues; the underlying study must still establish the empirical result.^{3940414243444546}

Admissions — prediction, opportunity, and a denominator problem

Deary makes it difficult to dismiss cognitive testing as empty credentialism. He describes an English cohort linking cognitive tests at about 11 to GCSE results around 16. The observed correlation with the best-eight GCSE total was 0.72. A separate model using latent traits in a subset taking the same six subjects gave 0.81. These are different analyses, not interchangeable estimates from an identical sample. The association is substantial; it does not mean a score determines an individual child’s future. Deary also reports that girls achieved better results at the same general cognitive score and emphasizes other contributions to achievement.⁴²⁴⁷

The lesson for the exchange is limited but meaningful: measured preparation can predict later academic performance even when preparation reflects unequal opportunities. Predictive usefulness and a fair account of how someone acquired that preparation are different questions. A large association also does not identify the consequences of raising a score through a particular intervention.

Lemann’s argument is more serious than “the SAT predicts nothing.” In chapter 5 he acknowledges findings favoring tests over grades in the highly selective setting. He then asks whether institutions should organize admissions primarily around an individual’s academic credentials, the development of future leaders, wider opportunity, or the institutions’ own purposes. He also redirects attention toward colleges serving many more students. Accepting a predictive result does not settle those goals. Conversely, an institutional mission cannot make an inaccurate empirical claim true.⁴⁴

One such claim needs to be withheld. Lemann says the top parental-income decile supplies 98% of enrolled Ivy Plus students scoring at least 1500, and makes an adjacent 1.2% claim about lower-income students. I retrieved the exact October 2023 paper version his note cites. Its Appendix Table 4 reports three different things: income composition among Ivy Plus attendees; score frequencies within national income groups, including non-test-takers; and Ivy Plus score frequencies in the final column. It does not provide the joint income-by-score composition among Ivy Plus attendees needed for his statements. The cited table therefore does not substantiate those percentages as worded. I have not invented a replacement Ivy Plus percentage.⁴⁴⁴⁸

This limitation does not eliminate the paper’s evidence of socioeconomic stratification or Lemann’s normative question. It does remove a striking number from the argument’s reliable support. It also illustrates why a familiar author or congenial conclusion deserves the same denominator check as an opponent’s claim.

For Barry’s original GPA comparison, there are still three separate decisions: what predicts performance; who benefits from an offer; and whose claims to a scarce place should count most. The original Ivy Plus study above addresses a narrower score-and-first-year-grade question. Neither Deary’s English school cohort nor Lemann’s institutional history converts it into a direct test of two applicants with GPAs of 3.9 and 3.8. My assessment remains that Ash identifies an empirically testable rationale for contextual admissions, while leaving other rationales open.

Genetics and intervention — the real Murray–Harden disagreement

Murray’s argument depends on an additional premise beyond heritability. In chapter 13 of Human Diversity, he connects small shared-environment contributions to many traits with the proposition that policymakers can manipulate shared environments, while much consequential nonshared environment is difficult to identify or control. If those are the only useful levers, modest effects are a reasonable expectation. His argument gains force from repeated failures of interventions intended to change durable characteristics. It deserves to be addressed in that conditional form.⁴⁰

There are two gaps between that evidence and a general ceiling on intervention. First, a variance decomposition describes differences within the environments and people actually studied. It is not a direct estimate of what happens under every new intervention. Second, a policy can raise an average without explaining much of the remaining variation. If an effective change reached everyone similarly, individual differences could persist after a worthwhile improvement. Statistical “shared environment” means environmental influences that make family members more alike; it is not a list of all things a government can change.

Murray himself supplies important limits to an overly rigid reading. On PDF 343 he discusses environments helping motivated people accomplish more without remaking their personality. On PDF 354 he endorses warm, nurturing preschool environments as worthwhile even apart from distant measured effects. On PDF 362 he describes a large growth-mindset trial as yielding robust, small positive findings. Treating him as asserting that every environmental intervention has zero effect would misrepresent his book.⁴⁰

Harden’s strongest response also accepts that change is difficult. Chapters 6–9 argue that genetic effects are real and can operate through long pathways involving other people and institutions. Her discussion of failed educational and behavioral interventions explicitly challenges the idea that reformers already know what to do and only lack political will. Her position is therefore a useful internal challenge to environmental explanations that ignore genetic confounding, as well as to genetic explanations treated as policy fatalism.⁴¹

The key distinction is between a cause of differences and an unchangeable mechanism. Conditional on parents’ genomes, differences in which variants siblings inherit provide useful causal leverage. Harden illustrates this using height research; the original Visscher study used realized genetic sharing among siblings. It is evidence for genetic influence, not a randomized comparison of educational policies. Later within-sibship research also finds genetic associations with education and cognition, while showing that ordinary population associations can include family and demographic effects. Neither result supports dismissing genetics; neither turns every observed association into a purely biological pathway insulated from environmental change.⁴¹⁴⁹⁵⁰

A useful counterexample to a claim of fixed cognitive-test performance comes from Ritchie and Tucker-Drob’s 2018 synthesis of three quasi-experimental designs: additional schooling improved measured cognitive scores. Its approximate 1–5 IQ points per additional year varied by design. The authors explicitly could not reliably distinguish effects on latent general intelligence from effects on particular tests, or establish unlimited accumulation across years. This supports some malleability, not guaranteed large transformations or complete group-gap closure.⁵¹

My assessment is that Murray offers a defensible warning about ambitious, weakly tested promises. The broader inference requires testing the actual available lever, not treating heritability as a substitute for that test. Harden gives the better explanation of why genetic causation alone does not settle policy effectiveness, but she does not thereby demonstrate that a proposed policy works. A feasible, durable, affordable improvement is a separate empirical achievement.

Racial means, moral equality, and sex differences in interests

The genetic evidence must answer the question actually being asked. Harden correctly distinguishes genetic causes of individual variation within a studied population from the causes of mean differences between populations. The mathematical identification limitation discussed in the original analysis remains relevant: within-group heritability alone does not estimate the genetic share of a racial mean gap.⁴¹¹⁸

Two additional distinctions prevent overstating her conclusion. Her expectation that populations will differ minimally in the prevalence of genetic variants relevant to psychological characteristics, including tested cognitive abilities, is an empirical expectation, not a consequence of non-identification. Her discussion of racist priors or motivations is also not established simply by the statistical argument. The defensible inference here is uncertainty about a particular causal decomposition, rather than proof of a genetic contribution of zero or proof of any speaker’s motives.

Murray’s own ancestry discussion deserves equally careful treatment. He distinguishes ancestral population structure from a claim that conventional races are primordial, sharply bounded biological types. He also acknowledges difficulties with population stratification and interpreting variant differences. The selected passages do not establish a settled genetic cause of U.S. racial cognitive-score differences. Those caveats strengthen the case for reading his actual argument while resisting a leap from ancestry structure to a quantified explanation of a behavioral gap.⁴⁰

Harden’s moral claim is separate again: a person’s basic standing should not depend on winning either the genetic or social lottery. That argument does not require pretending that everyone has identical capabilities. It also does not by itself specify a tax schedule, admissions rule, or equality of all outcomes. A policy still needs a stated goal, evidence of benefit, and an account of competing claims. This is a substantive philosophical disagreement about which inequalities should matter, alongside the empirical disagreement about their causes.⁴¹

Sex differences in interests remain a stronger descriptive example for Murray. His discussion of the things–people dimension draws on the same Su synthesis already assessed above; reading it again in a book does not create another independent confirmation. Average interests, abilities, and actual occupational outcomes nevertheless remain different measures. A preference distribution cannot assign an individual’s interests or determine how much of a workforce disparity comes from preferences, opportunity, discrimination, caregiving, or working conditions.⁴⁰²⁰

A concrete correction is necessary in his CAH example. The supplied Human Diversity, PDF 159, describes 125 women with CAH. The original 2011 methods instead report 125 participants in total: 46 females with CAH, 21 unaffected sisters, 27 males with CAH, and 31 unaffected brothers. The substantive comparison still found greater things-oriented interest among females with CAH than among unaffected females. The authors themselves caution that CAH is not a perfect experiment: affected and unaffected participants differ in other ways. The correction changes sample size and description; it does not reverse the observed finding or quantify its contribution to occupational differences in the general population.⁴⁰²¹

Welfare — incentives, poverty measures, and a revised marriage example

The strongest idea in Losing Ground is that social policy changes choices. Murray argues that people with few resources respond to the risks, eligibility rules, rewards, and penalties they actually face. A program can relieve immediate deprivation while altering work or family incentives in an undesirable way. This is a causal hypothesis that can be investigated; intentions and spending totals cannot answer it.⁴³

His Harold-and-Phyllis example makes that mechanism visible through a stipulated household budget. But a hypothetical budget does not tell us how many households face those exact choices, understand the rules, claim every benefit, or respond by changing work or marriage. Those require data. Its historical eligibility assumptions also cannot serve as a statement of current law. The example is analytically useful as a mechanism, and inadequate as an estimated population effect.

Murray also explicitly distinguishes official cash-income poverty, a broader net-income measure, and “latent poverty” before government transfers. It would be unfair to dismiss his entire case as simply forgetting noncash benefits. His independence measure asks whether people can maintain an adequate income without public support. That is a different objective from measuring current deprivation.⁴³

The causal problem survives the measurement distinction. Subtracting benefits from people’s actual income does not reconstruct how much they would earn, save, receive privately, or spend under a different policy regime. It is an accounting comparison, not a complete behavioral counterfactual. His measure also subtracts Social Security and disability payments, combining questions about labor-market self-support with retirement and disability insurance. A rise can identify a question about dependence without establishing which policy caused it or whether overall welfare deteriorated. The book’s larger explanation of 1950–1980 trends would require a fuller historical test of competing mechanisms; this targeted pass does not certify it.

The marital-dissolution example substantially weakens as a claim of settled causation. Murray’s chapter 11 treats increased breakups under the negative income tax as causally clear. Cain and Wissoker’s later reanalysis separated treatment packages and reconsidered sample, timing, and attrition. They concluded that the pure NIT did not have the claimed effect; Hannan and Tuma defended the earlier interpretation, and Cain and Wissoker replied. The 1990 record is a real analytical dispute, not an uncontested debunking.⁴³⁵²

There is now a further update. Riddell and Riddell’s 2026 paper rejects the increased-separation finding as driven by Seattle, where it identifies compromised randomization. The final abstract and introduction were inspected; the full 2025 predecessor was used for methodological detail, rather than being presented as the final article’s verified tables. The study reports both larger labor-supply reductions and improved happiness, marital satisfaction, and household production in Winnipeg. The marriage reassessment therefore does not erase work disincentives, and work hours alone do not establish the balance of benefits and costs.⁵³

The broader connection to Poor Economics is instructive. Both books explain poor people’s choices through ordinary human motives under particular constraints. They differ in policy emphasis and in how far they generalize. Banerjee and Duflo’s conclusion favors solving identifiable problems and learning from interventions over expecting a single lever to eliminate poverty. That is compatible with taking incentive costs seriously. A reduction in work hours can be a cost to fiscal capacity while more time, income security, or household well-being can be benefits. An overall policy judgment must state how these outcomes are valued instead of treating one measure as the whole verdict.⁴⁶

Rehabilitation — revision, coercion, and the limits of treatment selection

Cullen and Gilbert complicate the political history as well as the effectiveness claim. Their original argument describes liberal critics who opposed coerced treatment and broad official discretion because they feared unequal punishment and abuse of power. In that account, opposition to rehabilitation could arise from rights-based concerns as well as doubts about recidivism. This is their historical interpretation, not proof that all liberals or conservatives held one position. It shows why a treatment’s average effect and the legitimacy of compelling someone to undergo it require separate arguments.⁴⁵

The later anniversary discussion adds a specific correction to a frozen “nothing works” history. In 1979 Martinson expressly revised his earlier conclusion, recognizing beneficial as well as harmful treatment effects. His original article confirms the revision. Its 555-study base included routine follow-ups and other comparisons, not 555 randomized trials, and the comparisons were exploratory. The history supports heterogeneity and correction, not a claim that rehabilitation was universally vindicated. These later conclusions must also not be moved backward into what was known by 1972.⁵⁴

The book’s constructive answer is risk–need–responsivity: match intensity to risk, target relevant needs, and use approaches suited to participants. Cullen and Gilbert themselves call for humility and continued testing. Recognizing who is likely to reoffend is different from identifying who benefits most from a treatment. A 2024 umbrella review found inconsistent support and generally weak underlying evidence for the model’s principles, including concerns about conflating prediction and causality. This materially qualifies confidence in the framework without showing that useful rehabilitation is impossible.⁴⁵⁵⁵

There is positive treatment evidence too. A Campbell review of cognitive-behavioral programs found favorable average recidivism outcomes across a mix of randomized and nonrandomized studies. Quality, delivery, and setting remain important limitations. Its evidence is distinct from the prison-education synthesis already discussed above, and overlapping later reviews should not be counted as additional independent trials. I have withheld the book’s “10–12 percent” summary because its percentage convention was not resolved in this pass.⁵⁶

My assessment is that the books strengthen the original conclusion that rehabilitation is neither a guaranteed cure nor an empirically closed subject. They also make “which program for whom?” harder: it is not enough to replace blanket pessimism with an uncritically accepted selection model. Both the program and the proposed rule for targeting it need evaluation.

Additional cases — failed inputs, schoolwide bonuses, and vaccination uptake

Leigh’s New York example is a real failure of a particular teacher-incentive design. The randomized program offered schoolwide bonuses in high-poverty schools. It did not produce the hoped-for educational improvement. Its result bears directly on that package; it does not by itself test every individual merit-pay arrangement. Complexity, weak individual incentives, or free riding are candidate explanations, rather than mechanisms separately isolated by the experiment.³⁹⁵⁷

The Kenyan textbook case is a different kind of failure. Poor Economics describes an average test-score null alongside gains among students who already performed well. The original published abstract supports that pattern. English-language accessibility and a curriculum pitched above many students are the authors’ explanation, not a separate randomization of language. The result gives a reason to question the match between input and learner; it cannot establish that textbooks generally have no value or that the remaining children cannot learn.⁴⁶⁵⁸

The Udaipur case shows a successful, bounded intervention and a numerical discrepancy worth preserving. In the published cluster trial, full immunization among endpoint children aged 1–3 was 148/382, about 39%, with reliable camps plus incentives; 68/379, about 18%, with reliable camps alone; and 50/860, about 6%, in controls. The supplied book reports 38% with incentives and 6% in controls; its 38% differs slightly from the published result rounded to 39%. That small numerical difference was not reconciled, and the direction of the benefit is unchanged. This is strong evidence about uptake in that setting. It does not measure deaths prevented or isolate one psychological mechanism.⁴⁶⁵⁹

Together, these cases illustrate a better response to a failed intervention. Failure should reduce confidence in that design. A proposed explanation should lead to a different, testable design or a decision to stop—not an endlessly protected theory. But moving directly from “this intervention failed” to “the relevant social conditions cannot matter” skips several possibilities: the lever was mismatched, the treatment was too weak, the wrong outcome was measured, implementation failed, or the proposed mechanism was false. Evidence must decide among them.

The reverse discipline matters too. A favorable trial does not establish that the effect survives expansion, exceeds costs, avoids displacement, or eliminates a disparity. Leigh’s attention to replication and research incentives, and Banerjee and Duflo’s attention to implementation and local constraints, help make that symmetry practical. Neither a shelf of books nor a collection of selected cases supplies the missing representative denominator for an ideological success rate.³⁹⁴⁶

What this changes in the assessment

The substantive reading changes the assessment in five concrete ways.

1. The strongest case for Murray is narrower and better developed. His warnings concern durable traits, manipulable levers, incentives, and repeated disappointments. He allows meaningful exceptions and improvements. Those qualifications should be part of any serious account of his position.
2. Harden is a stronger counterpoint because she accepts the hard parts. Genetic influences, confounding, and failed interventions are central to her argument. Her explanation of environmental mediation challenges fatalistic inferences; her predictions about group means and judgments about motives remain separate claims.
3. The books themselves require correction and updating. Murray’s CAH participant description is wrong; Lemann’s cited table does not support his striking Ivy Plus statistic; the marriage argument needs subsequent reanalyses; and the rehabilitation model needs current criticism alongside supportive evidence.
4. Different kinds of equality and success must be specified. Predictive accuracy, opportunity, average welfare, independence from transfers, group-gap reduction, and protection from coercion are different objectives. A finding can favor one while leaving another unresolved.
5. Sailer’s broad generalization remains unestablished. These books deepen several cases, but their selection is purposeful rather than representative. They cannot justify a percentage of “liberal” propositions that failed, or demonstrate that every competing political outlook performed better.

The most consequential remaining work is therefore bounded. For admissions, compare policies using both predicted performance and causal gains while stating the allocation goal. For welfare, obtain the final 2026 tables and reconcile the earlier experimental specifications before using a numerical marriage effect. For rehabilitation, distinguish program efficacy from evidence that an allocation model improves outcomes. For racial cognitive means, require designs that can identify the proposed causal decomposition. More books alone will not resolve those questions; the relevant studies and assumptions will.

Claim status and unresolved checks

This is a targeted ledger, not a denominator for judging either ideology’s overall empirical performance. “Supported” below concerns the bounded proposition stated, not endorsement of all associated arguments.

Source access remains uneven. The complete final 1990 marital-dissolution exchange and final 2026 Riddell tables were not obtained; the accessible predecessor is identified separately. The Kenyan trial’s final abstract and institutional account were inspected, rather than claiming a full final-paper audit. The vaccination version discrepancy and the rehabilitation percentage convention remain unresolved. The full original Coleman report and the book-wide historical poverty explanation have not received exhaustive new audits here. No statistical replication or new representative search of all relevant interventions was performed.

Related notes

• 60 Knowledge/60.3 Politics/60.3.11 - Leftism & Woke Ideology/DEI — admissions, hardship, DEI, and the distinction between prediction and normative selection goals.
• 60 Knowledge/60.3 Politics/60.3.9 - Social Issues/Race/Hereditarianism, Eugenics, Race, and IQ in the United States - A Chronological Timeline — historical context for race–IQ arguments, with causal and provenance limits kept separate.
• 60 Knowledge/60.3 Politics/60.3.12 - Source Audits & Media/Steven Crowder vs. Randall Kennedy - Harvard, DEI, Free Speech, and Federal Funding — a related source audit on Harvard, academic credentials, affirmative action, and institutional scope.
• 60 Knowledge/60.3 Politics/60.3.12 - Source Audits & Media/Sabine Hossenfelder - More Exposed Than Ever (Racism Edition) — a related dossier on race–IQ evidence, biological claims, source lineage, and causal inference.

Sources and footnotes

The sources below retain direct links, publication identities, relevant locators, access status, and the qualifications used in the original research. Multiple URLs for one report are not independent corroboration.

Vault integration record

• User authorization: explicit request on 20 October 2026 to save the entire analysis and incorporate subsequent questions and answers from this chat.

• Canonical destination: a new source-audit dossier because the four exact tweet IDs had no active-vault match; the existing Sailer interview is a different source.

• Research record: Protocol 1.2; canonical claim IDs C01–C15 are retained in the chat’s claim ledger. C01–C04 concern admissions; C05–C09 the broad/historical assertions; C10–C15 the six research areas. The original final answer and research receipt are preserved above.

• Independent routing: the existing claim-mapping specialist performed a new read-only live-vault mapping pass after the new-agent thread limit prevented a separate mapper instance. Only the coordinating agent writes.

• Permission boundary: Anthony authorized the existing parent-permission exception. Specialists followed read-only instructions; the parent session itself was not sandbox-enforced read-only.

• Graph scope: this dossier’s parent and related-note links, one Source Audits MOC annotation, and one contextual admissions callout in DEI. No moves, merges, new hubs, or broader cleanup.

• Preservation check: original answer SHA-256 84df691cbc36089cc9fadfaf15ae190c13023be273a7dec9c647afbfcd0b9017; 21,557 characters. Recovery from the analysis after removing only added study headings and native footnote references must equal the captured original exactly.

• Initial vault-capture validation: the complete original analysis was recovered exactly; all 28 original source URLs remain; 28 same-note footnote definitions resolve across 29 references; all new and existing wikilinks and exact heading anchors in the three touched notes resolve; YAML frontmatter parses; Markdown-table checks report zero issues. The five-row, three-column comparison was inspected in Obsidian Reading view, including rendered source labels and footnote markers. The independent installation audit is reported in this chat’s completion record; the original chat-only research receipt above remains historical.

• Final installation audit: PASS WITH DISCLOSED LIMITS. The independent auditor checked exact answer recovery, the three live files and scoped existing-note diffs, all 78 wikilink occurrences, frontmatter, the 17 main-section contents links, and 28 footnote definitions across 29 uses. Original source-access limits, lack of independent X recapture, and lack of blind replication remain disclosed. The coordinator performed the actual Obsidian visual check; the auditor did not independently repeat it.

Correction and maintenance history

• 20 October 2026 — source-attribution correction during research: narrowed the draft’s overbroad “almost all liberal ideas” / “almost an entire political outlook” wording to Sailer’s own undefined category of “liberal feel-good nostrums”; the preserved final answer already reflects this correction. SAT/GPA scope, recidivism odds, Year Up quarterly endpoint, and between-group heritability limits were checked before the research answer was delivered.

• 20 October 2026 — vault capture: preserved the complete delivered analysis and original receipt, added navigable study headings, native source footnotes, and a dedicated follow-up section. No substantive findings were silently changed.

• 20 October 2026 — bibliographic clarification: the added Year Up footnote uses the year 2022 without a month because the cover and version/filename use differing month labels. The preserved analysis and numerical findings are unchanged.

• 20 October 2026 — further analysis and books follow-up: incorporated Anthony’s question verbatim, five prioritized investigations, a three-book core route, one optional primer, four topic-specific alternatives, existing reading-record links, and eight new source footnotes. The original full analysis and all original source definitions remain unchanged. Only this dossier was edited; reading plans, statuses, book notes, the MOC, and DEI were not changed by this follow-up. Further-analysis follow-up validation: 36 same-note footnote definitions resolve across 49 references; all 35 wikilink occurrences and contents anchors resolve; 24 headings; YAML parses; two Markdown tables report zero issues. The new three-book comparison was inspected in actual Obsidian Reading view. The entire original answer and all 28 original source URLs remain preserved exactly.

• 20 October 2026 — NAS library follow-up: incorporated Anthony’s library-check question and results in this same dossier. None of the eight recommended titles was found by title/author path searches or the two checked catalog databases; the two Sowell companion titles were found as PDFs. Preserved the original analysis and prior follow-up answers. Library files, reading plans, reading statuses, existing book notes, the MOC, and DEI were not modified by this follow-up. Initial library-search follow-up validation: 37 same-note footnote definitions across 58 references; 36 wikilink occurrences. The lookup was performed directly by the coordinator; no fresh specialist research or full-book audit is claimed.

• 20 October 2026 — supplied books filed: recorded Anthony’s filing request, eight verified current PDF paths, the 2011 Poor Economics edition distinction, the 2020 Deary edition, anniversary/reissue details for Reaffirming Rehabilitation, and the Higher Admissions filename attribution correction. Marked the earlier missing-title search as historical rather than discarding it. All eight moves preserve original bytes and hashes. This follow-up changes eight library file locations/names and this dossier; reading plans, reading statuses, existing book notes, MOC, and DEI were not changed. Filing follow-up validation: 38 native footnote definitions across 59 references; 38 wikilink occurrences. No new substantive research verdict or independent specialist audit is claimed for this file-management update.

• 20 October 2026 — book-use status clarified: incorporated Anthony’s question verbatim and explicitly distinguished completed bibliographic inspection/filing from pending substantive reading and synthesis of the supplied books. Added a visible scope callout and contents link. The original analysis, prior questions and answers, source definitions, library files, and reading statuses remain unchanged. No new research conclusion or independent audit is claimed for this status clarification.

• 20 October 2026 — substantive analysis of eight supplied books: incorporated Anthony’s request verbatim and added a developed, source-audited analysis based on selected substantive passages from all eight supplied editions. Updated the visible book-use status and explicitly labeled the earlier “No” answer historical. Canonical claims B01–B20 cover admissions, genetics, interests, welfare, rehabilitation, school interventions, and vaccination uptake. Corrected the CAH sample description; withheld Lemann’s unsupported percentages and the unresolved rehabilitation percentage; qualified the NIT marriage inference using the disputed 1990 exchange and the 2026 paper with its separately identified predecessor. Disclosed the vaccination book/article mismatch. Original full answer, prior Q&A, and all earlier footnote definitions remain preserved. Only this dossier is changed; existing MOC and DEI connections remain in place.

Substantive book-analysis review and installation receipt

• Mode and protocol: Vault integrated; source-audited research Protocol 1.2. Existing canonical dossier reused after live graph mapping; no new book notes or reading-status changes.

• Actual specialist roles: intake auditor; claim mapper, subsequently reused for a bounded vault-mapping pass; primary-source researcher; adversarial reviewer; independent final auditor. Specialists were instructed to remain read-only. The coordinator alone prepares and installs changes.

• Permission boundary: Anthony’s prior explicit exception permits behavioral read-only specialists under this write-capable parent. Sandbox-enforced read-only specialist permissions are not claimed.

• Intake: PASS for all eight current PDFs: identity/edition, hashes, page counts, extractability, and distributed body/end samples. No exhaustive visual completeness or publisher-authenticity verification.

• Research and adversarial reconciliation: the independent reviewer confirmed the main synthesis after qualifications on Murray’s concessions, Harden’s separate empirical expectations and motive claims, Social Security/disability in latent poverty, positive Winnipeg household outcomes alongside labor-supply reductions, and uncertain RNR validation. Those qualifications are incorporated. The unrelated self-esteem passage was excluded rather than endorsed without a full new claim audit.

• Attribution and citation scope: selected substantive book passages and consequential underlying studies, with per-source access and edition/version limits in the footnotes. B19 is unsupported by the cited table; B14 and parts of B18/B20 remain explicit holds. Multiple books or versions repeating the same study are not independent confirmations.

• Arithmetic and denominators: independently reviewed CAH group total and Udaipur arm fractions; observed and latent educational correlations are kept distinct. No ideological success percentage or unverified corrected Ivy percentage is calculated.

• Replication: targeted source review and independent adversarial checks; no fresh statistical replication, blind replication session, or representative systematic search of all interventions.

• Final installation audit: PASS WITH DISCLOSED LIMITS after two narrow attribution repairs. Attribution/quotation PASS; citation entailment PASS with the specified source-access limits; arithmetic/denominators PASS; adversarial/readability PASS with disclosed limits; preservation/live integration PASS. The independent auditor verified the installed addition, matching mirror, exact original-answer recovery, all 28 original URLs, all 38 earlier source definitions, 59 native footnote definitions across 107 references, 55 resolving wikilink occurrences, complete contents for 38 headings, unchanged MOC/DEI hashes, parsed YAML, and four tables with zero checker issues. No outstanding source-support failure or required correction remains for this targeted addition. The original installation and earlier follow-up receipts remain dated historical records.

• Updated rendered appearance: unverified. The coordinator attempted to inspect the addition in Obsidian, but the app returned stale or inconsistent content. The earlier rendered-table checks do not establish the appearance of these new tables. The final live file and source structure were reread and checked directly; a visual PASS is not claimed.

• Final-audit repair — 20 October 2026: the first independent review withheld approval for two source-attribution errors. Removed an unsupported exact 16% camps-only figure from the book comparison; the supplied passage gives 38% and 6%, while the primary trial supplies all three arm counts. Specified that Harden’s prediction concerns differences in the prevalence of relevant genetic variants. These repairs narrow attribution and do not change the verified trial fractions or the overall assessment.

Footnotes

1. Redmond Barry (@RedmondBarry7), X post, 16 October 2026. Primary source for the hypothetical and normative admissions question, not evidence about actual applicants. Locator: comparison of the 3.9 and 3.8 GPAs and closing question. Account, displayed date, text, and embedded relationship observed directly in the coordinator’s browser; no independent second capture. Direct source. Access/review cutoff: 20 October 2026. ↩

2. Hunter Ash (@ArtemisConsort), X post, 17 October 2026. Primary source for Ash’s prediction argument. Locator: three-sentence response proposing college overperformance as a test. No empirical study was cited in the post. Direct coordinator browser capture; no independent second capture. Direct source. Access/review cutoff: 20 October 2026. ↩

3. Steve Sailer (@Steve_Sailer), X post, 17 October 2026. Primary source for the broad testing/failure assertion and 1972 recollection. Locator: both sentences. The category of nostrums, outcome criteria, and denominator are undefined. Direct coordinator browser capture; no independent second capture. Direct source. Access/review cutoff: 20 October 2026. ↩

4. Charles Murray (@charlesmurray), X post, 18 October 2026. Primary source for the six research areas. Locator: the full list beginning with the Coleman Report. Coleman is the only individually named study; precise findings from the other domains are not specified. Direct coordinator browser capture; no independent second capture. Direct source. Access/review cutoff: 20 October 2026. ↩

5. John N. Friedman, Bruce Sacerdote, Douglas O. Staiger, and Michele Tine, “Standardized Test Scores and Academic Performance at Ivy Plus Colleges,” AEA Papers and Proceedings 115 (2025), 676–681, DOI: 10.1257/pandp.20251056. Original observational research; full paper inspected. Locators: pp. 677 and 680, calibration, sample selection, and comparison with other settings. Same SAT/ACT and first-year outcomes among selected matriculants; this does not directly resolve the two-GPA hypothetical. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

6. James S. Coleman and colleagues, Equality of Educational Opportunity (U.S. Office of Education, 1966). Original federal survey; partial access. Locators: summary discussion, especially printed p. 22; indexed original passages on facilities, curricula, teacher quality, and student-body composition. Original full report was not completely inspected. Associations must not be converted into causal zero effects. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

7. ERIC, record ED012275, Equality of Educational Opportunity (1966), 746 pages. Official bibliographic record and abstract; inspected. The abstract explicitly reports that minority pupils were more affected by school quality than the average white pupil. This record and the original PDF belong to the same evidence chain. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

8. C. Kirabo Jackson, Rucker C. Johnson, and Claudia Persico, “The Effects of School Spending on Educational and Economic Outcomes: Evidence from School Finance Reforms,” Quarterly Journal of Economics 131(1) (2016), 157–218, DOI: 10.1093/qje/qjv036. Peer-reviewed quasi-experimental study; author-hosted final paper inspected. Locators: abstract, reform-based identification, and long-term education, wages, and poverty results. Historical cohorts and instrument assumptions limit generalization. Direct source. Access/review cutoff: 20 October 2026. ↩

9. C. Kirabo Jackson and Claire L. Mackevicius, “What Impacts Can We Expect from School Spending Policy? Evidence from Evaluations in the United States,” American Economic Journal: Applied Economics 16(1) (2024), 412–446, DOI: 10.1257/app.20220279. Peer-reviewed meta-analysis; published abstract inspected. Locator: average effects on test scores and college-going. Some underlying studies overlap with other cited spending evidence; do not count the synthesis and its constituents as independent experiments. Direct source. Access/review cutoff: 20 October 2026. ↩

10. Robert Martinson, “What Works?—Questions and Answers About Prison Reform,” The Public Interest 35 (Spring 1974), 22–54. Review essay based on a commissioned review; full PDF inspected. Locators: pp. 23–29 and methodological qualifications. It draws on 231 evaluations from 1945–1967. The underlying review predates publication; publication in 1974 alone cannot establish what was available or known in 1972. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

11. Robert Bozick, Jennifer Steele, Lois Davis, and Susan Turner, “Does Providing Inmates with Education Improve Postrelease Outcomes? A Meta-analysis of Correctional Education Programs in the United States,” Journal of Experimental Criminology 14 (2018), 389–428, DOI: 10.1007/s11292-018-9334-6. Peer-reviewed meta-analysis; full author-uploaded paper inspected. Locator: Table 2 and accompanying discussion. Stronger studies: odds ratio 0.72, 95% CI 0.61–0.84; lower odds, not lower probability. Most evidence is quasi-experimental; two randomized studies concern one program. Employment gains were inconclusive. Direct source. Access/review cutoff: 20 October 2026. ↩

12. Peter Z. Schochet / Mathematica Policy Research for the U.S. Department of Labor, National Job Corps Study: 20-Year Follow-Up Study Using Tax Data, ETA Occasional Paper 2019-13 (released 20 February 2019). Government-commissioned randomized evaluation; full report inspected. Locators: printed pp. 3, 5–8, and age-specific findings. No demonstrated overall long-term employment/earnings effect; some favorable older-participant outcomes. Older-participant twenty-year earnings estimates were not statistically significant. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

13. David Fein and Samuel Dastrup, Benefits That Last: Long-Term Impact and Cost-Benefit Findings for Year Up (Abt Associates for HHS/ACF/OPRE, 2022). Government-commissioned randomized evaluation; full report inspected. Locator: Exhibit 2-1, printed p. 14. Mean quarterly earnings impact in quarters 23–24: $1,895; control mean $6,901; independent ratio check = 27.4598%, reported as 27.5%. This is a quarterly endpoint around six years after assignment, not an annual seventh-year effect. The report’s annual figures contain a discrepancy; none is used here. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

14. Raj Chetty, Nathaniel Hendren, and Lawrence F. Katz, “The Effects of Exposure to Better Neighborhoods on Children: New Evidence from the Moving to Opportunity Experiment,” American Economic Review 106(4) (2016), 855–902, DOI: 10.1257/aer.20150572. Randomized-offer experimental follow-up; full paper inspected. Locators: abstract and child age-specific earnings/college results. Distinguish offer effects from participation-adjusted estimates; the prose deliberately avoids applying a take-up-adjusted percentage to all families offered vouchers. Direct source. Access/review cutoff: 20 October 2026. ↩

15. Hilary Hoynes, Diane Whitmore Schanzenbach, and Douglas Almond, “Long-Run Impacts of Childhood Access to the Safety Net,” American Economic Review 106(4) (2016), 903–934; earlier NBER Working Paper 18535 (2012), DOI: 10.1257/aer.20130375. Original quasi-experimental research using historical Food Stamp Program rollout; primary abstract/indexed material inspected. Locator: adult metabolic-health and women’s economic-self-sufficiency findings. This is not random assignment and does not identify all causes of poverty. Direct source. Access/review cutoff: 20 October 2026. ↩

16. Richard E. Nisbett and colleagues, “Intelligence: New Findings and Theoretical Developments,” American Psychologist 67(2) (2012), 130–159, DOI: 10.1037/a0026699; PubMed PMID 22233090. Scholarly review; abstract inspected. Locator: U.S. Black–White test-score-gap discussion. Historical source: do not treat its early molecular-genetics inventory as current evidence about how many associated variants have since been discovered. Direct source. Access/review cutoff: 20 October 2026. ↩

17. Ian J. Deary, Simon R. Cox, and W. David Hill, “Genetic Variation, Brain, and Intelligence Differences,” Molecular Psychiatry 27 (2022), 335–353; published online 2 February 2021; DOI: 10.1038/s41380-021-01027-y. Expert review; full text inspected. Locators: individual differences, predictive validity, heritability, and Box 2. Evidence for individual genetic variation does not by itself identify genetic causes of racial mean differences. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

18. Joshua G. Schraiber and Michael D. Edge, “Heritability within Groups Is Uninformative about Differences among Groups: Cases from Behavioral, Evolutionary, and Statistical Genetics,” PNAS 121(12) (2024), e2319496121, DOI: 10.1073/pnas.2319496121. Original mathematical research; full paper inspected. Locators: abstract, within-group/between-group derivation, and admixture discussion. The identification limitation does not establish zero genetic contribution to any specified group difference. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩² ↩³

19. Bernt Bratsberg and Ole Rogeberg, “Flynn Effect and Its Reversal Are Both Environmentally Caused,” PNAS 115(26) (2018), 6674–6678, DOI: 10.1073/pnas.1718793115. Original register-based sibling study; full text inspected. Locators: abstract, within-family results, and missing-score/selection correction. Norwegian male birth cohorts 1962–1991; the finding does not identify the cause of a U.S. racial score gap. Direct source. Access/review cutoff: 20 October 2026. ↩

20. Rong Su, James Rounds, and Patrick Ian Armstrong, “Men and Things, Women and People: A Meta-analysis of Sex Differences in Interests,” Psychological Bulletin 135(6) (2009), 859–884, DOI: 10.1037/a0017364. Peer-reviewed meta-analysis; institutional author abstract inspected. Locator: 47 inventories, 503,188 respondents, and things–people standardized mean difference d = 0.93. Observed average differences and measurement moderators do not establish a complete causal partition. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩² ↩³

21. Adriene M. Beltz, Jane L. Swanson, and Sheri A. Berenbaum, “Gendered Occupational Interests: Prenatal Androgen Effects on Psychological Orientation to Things versus People,” Hormones and Behavior 60 (2011), 313–317, DOI: 10.1016/j.yhbeh.2011.06.002. Original clinical/sibling comparison; full paper inspected. Locators: pp. 314–317, sample, group comparisons, exposure-proxy associations, and limitations. Sample total independently checked: 46 + 21 + 27 + 31 = 125. Supports a partial prenatal hormonal contribution; does not quantify its share of the general population gap. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩² ↩³

22. Yan Yi Lance Du, Serena Wee, and James Rounds, “Gender Differences in Vocational Interests Across 57 Countries: A Test of the Interest–Gender-Equality Paradox,” Social Psychological and Personality Science 16(1) (2025), 27–44; online 13 October 2024; DOI: 10.1177/19485506241280321. Original cross-sectional research; full/indexed primary text inspected. Locators: methods and discussion, pp. 30 and 40–41. N = 109,460 English-language online volunteers across 57 countries; associations differ across interest domains and cannot identify biological causation. An October 2024 correction addressed equations and a heading. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

23. Mike Puma and colleagues, Third Grade Follow-Up to the Head Start Impact Study: Final Report, Executive Summary, OPRE Report 2012-45b (October 2012), ERIC ED539263. Government-commissioned randomized-offer follow-up; full executive summary inspected. Locators: executive-summary pp. iii–v / PDF pp. 5–7, methods, alternative preschool access, and later access for the younger comparison cohort. Initial preschool benefits, few remaining average third-grade impacts; attrition and the comparison condition matter. Direct source. Access/review cutoff: 20 October 2026. ↩ ↩²

24. Martha J. Bailey, Shuqiao Sun, and Brenden Timpe, “Prep School for Poor Kids: The Long-Run Impacts of Head Start on Human Capital and Economic Self-Sufficiency,” American Economic Review 111(12) (2021), 3963–4001. Original historical quasi-experimental research; primary abstract/indexed published material inspected. Locator: historical rollout/age-eligibility design and adult education/self-sufficiency outcomes. Different cohorts, comparisons, and endpoints from the 2002-cohort randomized follow-up; neither study cancels the other. Direct source. Access/review cutoff: 20 October 2026. ↩

25. Eva Vivalt, Elizabeth Rhodes, Alexander W. Bartik, David E. Broockman, Patrick Krause, and Sarah Miller, “The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States,” NBER Working Paper 32719, issued July 2024, revised August 2026, DOI: 10.3386/w32719. Randomized experiment; working-paper status. Current primary abstract/indexed record inspected; direct full-paper retrieval unsuccessful. Locator: $1,000/month for three years versus $50/month control, work hours and job-quality results. No inference about permanent nationwide UBI or overall welfare follows from these endpoints alone. Direct source. Access/review cutoff: 20 October 2026. ↩

26. Steven L. West and Keri K. O’Neal, “Project D.A.R.E. Outcome Effectiveness Revisited,” American Journal of Public Health 94(6) (2004), 1027–1029, DOI: 10.2105/ajph.94.6.1027; PubMed PMID 15249310. Peer-reviewed meta-analysis; primary abstract and published indexed material inspected. Locator: overall d = 0.023, 95% CI −0.04 to 0.08, statistically nonsignificant. Scope is historical program versions; later redesigned curricula need separate evaluation. Direct source. Access/review cutoff: 20 October 2026. ↩

27. Anthony Petrosino, Carolyn Turpin-Petrosino, Meghan E. Hollis-Peel, and Julia G. Lavenberg, “Scared Straight and Other Juvenile Awareness Programs for Preventing Juvenile Delinquency,” Cochrane Database of Systematic Reviews, CD002796, DOI: 10.1002/14651858.CD002796.pub2. Systematic review; review body and results inspected. Locator: nine U.S. trials, 946 participants, underlying studies 1967–1992; pooled offending outcomes indicate harm. Review webpage displays a 2022 issue; the underlying search/trial dates are much earlier. Direct source. Access/review cutoff: 20 October 2026. ↩

28. Jonathan Guryan and colleagues, “Not Too Late: Improving Academic Outcomes among Adolescents,” American Economic Review 113(3) (2023), 738–765, DOI: 10.1257/aer.20210434. Two randomized tutoring studies; full published paper inspected. Locators: pp. 741 and 752–753, math/grade persistence and imprecise on-time graduation result. Academic gains must not be converted into demonstrated earnings, graduation, or complete gap elimination. Direct source. Access/review cutoff: 20 October 2026. ↩

29. Andrew Leigh, Randomistas: How Radical Researchers Are Changing Our World (Yale University Press, 2018), ISBN 9780300236125. Publisher metadata and description inspected; not a full-book review. Locators: publication date, title/author, and description of randomized experiments in education, crime, employment, and poverty. The recommendation’s judgments about methodological value and limits are the coordinator’s. Publisher record. Access/review cutoff: 20 October 2026. ↩ ↩²

30. Charles Murray, Human Diversity: The Biology of Gender, Race, and Class (Twelve, 2020), ISBN 9781538744000. Publisher metadata and description inspected; not an independent validation of the scientific claims. Locators: title, publication details, subject description, and author biography identifying the 1984 publication of Losing Ground. Publisher record. The live vault’s RP-120 record and 2026 Phase III board were separately inspected for planned status; no status was changed. Access/review cutoff: 20 October 2026. ↩ ↩²

31. Kathryn Paige Harden, The Genetic Lottery: Why DNA Matters for Social Equality (Princeton University Press, 2021), DOI: 10.2307/j.ctv1htpf72. Publisher-hosted JSTOR metadata, description, contents, and accessible opening excerpts inspected; full-book access unavailable. Locators: chapters 4–7, pp. 72–150, and chapters 8–9, pp. 153–192; these support the selective navigation suggestion, not a full-chapter assessment. Publisher book record on JSTOR. Princeton’s separate book webpage was blocked by robots restrictions; the JSTOR record names Princeton University Press as publisher. Access/review cutoff: 20 October 2026. ↩ ↩² ↩³

32. Ian J. Deary, Intelligence: A Very Short Introduction, second edition (Oxford University Press, published 26 March 2020), DOI: 10.1093/actrade/9780198796206.001.0001. Official publisher bibliographic record, abstract, and contents inspected through indexed primary material; a later direct-page request timed out. No full-book review. Locators: chapters 4, 7, 9, and 10 on environmental/genetic contributions, school/work, cohort changes, and agreement in psychology. Oxford Academic record. Access/review cutoff: 20 October 2026. ↩ ↩²

33. Charles Murray, Losing Ground: American Social Policy, 1950–1980 (first published 1984; publisher’s linked reissue record dated 2015), ISBN 9780465065882. Publisher description and reissue metadata inspected; not a full-book review. Locator: historical welfare-policy scope and incentive argument. The original year is also identified in the publisher’s Murray biography cited in followup-02. Publisher reissue record. Access/review cutoff: 20 October 2026. ↩ ↩²

34. Nicholas Lemann, Higher Admissions: The Rise, Decline, and Return of Standardized Testing (Princeton University Press, 2024), DOI: 10.2307/jj.14736617. Publisher-hosted JSTOR metadata, description, contents, and accessible opening excerpts inspected; full-book access unavailable. Locators: description and chapters 1–5, pp. 12–103; responses by Marvin Krislov, Patricia Gándara, and Prudence Carter. The recommendation concerns history and institutional/normative framing, not verification of current admissions-policy effects. Publisher book record on JSTOR. Access/review cutoff: 20 October 2026. ↩ ↩²

35. Francis T. Cullen and Karen Gilbert, Reaffirming Rehabilitation, second edition (Routledge, 2012), 266 pages; publisher ebook record ISBN 9781315721361. Primary publisher description, metadata, and contents inspected through indexed material; direct access later returned HTTP 403. No full-book review. Locators: historical debate over rehabilitation and the book’s affirmative rehabilitation argument. Publisher record. Use alongside more recent evaluations, rather than as a current comprehensive systematic review. Access/review cutoff: 20 October 2026. ↩ ↩²

36. Abhijit V. Banerjee and Esther Duflo, Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty, revised edition (PublicAffairs, 16 September 2025), ISBN 9781541706187, 400 pages. Publisher metadata and description plus author university bibliography inspected; not a full-book review. Locators: revised format, 2025 date, description of new material, and MIT bibliography entry identifying a revised and updated edition with two new chapters. Publisher record; Esther Duflo’s MIT books page. The publisher page retains stale preorder wording; no stock or shipping claim is made. Access/review cutoff: 20 October 2026. ↩ ↩²

37. Local library inspection, 20 October 2026, root ` ↩

38. Eight user-supplied local PDFs, inspected and filed on 20 October 2026. Primary evidence for identity, edition, and local availability; no full-book substantive audit. Current full paths and edition details are listed in “Books supplied and filed in the library”. Relevant PDF page locators: Higher Admissions, title p. 4 and copyright p. 5; Human Diversity, title/copyright/ISBN pp. 2–4; Deary’s Intelligence, title p. 19 and copyright p. 20; Losing Ground, copyright p. 6 and contents listing the 2015 introduction; Poor Economics, copyright p. 302; Randomistas, copyright p. 5; Reaffirming Rehabilitation, title p. 3 and publication/identifiers pp. 4–5; The Genetic Lottery, copyright p. 4. The coordinator also visually inspected the rendered Higher Admissions title page and the Poor Economics and Reaffirming Rehabilitation copyright pages. Filenames were cleaned during the moves; content hashes, byte sizes, original paths, and final paths were recorded in this chat’s filing manifest and rechecked after moving. All eight PDF contents and embedded metadata remain unchanged. ↩

39. Andrew Leigh, Randomistas: How Radical Researchers Are Changing Our World (2018), Anthony’s supplied local PDF, 286 viewer pages. Selected substantive passages read, especially PDF 88–90 (education/incentives) and 202–206 (publication, replication, and research practices), with education source notes at 249. Current local file path: “Books supplied and filed in the library” above. Publisher identity record. This supersedes the recommendation-stage access limit for this targeted reading; it does not certify the entire book. Reviewed 20 October 2026. ↩ ↩² ↩³

40. Charles Murray, Human Diversity: The Biology of Gender, Race, and Class (2020), supplied local PDF, 743 viewer pages. Selected substantive passages read: PDF 125–129 (interests), 159 (CAH comparison and sample description), 194–197 and 252–254 (ancestry and inferential limits), 339–349 and 351–365 (intervention constraints and qualifications). Particular concessions: 343, 354, 362. The “125 women with CAH” wording was checked directly against the supplied text and the originating paper. Publisher identity record. Source for Murray’s argument, not independent confirmation of every scientific claim. Reviewed 20 October 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

41. Kathryn Paige Harden, The Genetic Lottery: Why DNA Matters for Social Equality (Princeton University Press, 2021), supplied local PDF, 308 viewer pages. Selected substantive passages read: PDF 94–101 (group inference, moral standing, and the author’s additional expectations/interpretations), 118–126 and 133–136 (family-based genetics), 161–166 (social context, variance, and equality), 180–184 (intervention failures and causal research). Publisher-hosted book record. Her moral argument, empirical expectations, and judgments about motivations are kept distinct. Reviewed 20 October 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵

42. Ian J. Deary, Intelligence: A Very Short Introduction, second edition (Oxford University Press, 2020), supplied local PDF, 183 viewer pages. Selected substantive passages read: PDF 92–95 (genetic association/prediction versus variance estimates) and 119–122 (school achievement). At PDF 119, observed CAT general-intelligence/best-eight GCSE correlation = 0.72, maximum sample 74,403 mainstream state-school pupils. At 120, latent model correlation = 0.81 for 13,248 pupils taking the common six-subject set; 122 emphasizes remaining variation. Publisher record. These historical estimates are not current prediction ceilings or intervention effects. Reviewed 20 October 2026. ↩ ↩²

43. Charles Murray, Losing Ground: American Social Policy, 1950–1980 (original text 1984; supplied edition includes 2015 introduction), local PDF, 468 viewer pages. Selected substantive passages read: PDF 30–32 (strategic explanation), 97–104 (official/net/latent poverty), 207–215 (NIT experiments), 216–234 (household-choice model and its assumptions). PDF 99 explicitly subtracts AFDC, Social Security, disability payments, and SSI; PDF 214–215 assert causal clarity about marital dissolution. Historical budget calculations are not current eligibility advice. Publisher identity record. Reviewed 20 October 2026. ↩ ↩² ↩³ ↩⁴

44. Nicholas Lemann, Higher Admissions: The Rise, Decline, and Return of Standardized Testing (Princeton University Press, 2024), supplied local PDF, 128 viewer pages. Selected substantive passages read: PDF 61–63 and 64–79. Lemann acknowledges predictive evidence at 71 and 73; the 98% and adjacent 1.2% assertions appear at 73–74. Chapter 5 note 8 at PDF 115 points to Chetty, Deming, and Friedman, October 2023, p. 87. Lemann’s chapters are distinguished from the introduction and commentaries. Publisher-hosted book record. Reviewed 20 October 2026. ↩ ↩² ↩³

45. Francis T. Cullen and Karen E. Gilbert, Reaffirming Rehabilitation, thirtieth-anniversary edition (publication/copyright history 2013, Routledge reissue 2015), supplied local PDF, 465 viewer pages; ISBN 9781455731305. Selected substantive passages read: PDF 149–156 (original historical argument about liberal critics), 386–405 (anniversary discussion of evidence, revision, and risk–need–responsivity), with associated notes at 430–432. Original 1982 argument and later material are not treated as contemporaneous. Publisher identity record. Book advocacy and current empirical assessment are distinguished. Reviewed 20 October 2026. ↩ ↩² ↩³

46. Abhijit V. Banerjee and Esther Duflo, Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty (2011 original), supplied local PDF, 302 viewer pages, eISBN 9781610390408. Selected substantive passages read: PDF 63–65 (Udaipur), 90–97 (textbooks, curriculum, and remedial instruction), 237–242 (conclusion), relevant notes 253 and 258. This is not the revised 2025 edition recommended earlier. Current local path and edition verification appear above. Author’s book bibliography. Specific study findings are independently checked below rather than adopting every broader claim. Reviewed 20 October 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵

47. Ian J. Deary, Steve Strand, Pauline Smith, and Cres Fernandes, “Intelligence and Educational Achievement,” Intelligence 35(1) (2007), 13–21, DOI 10.1016/j.intell.2006.02.001. Author-institution abstract inspected, with model/sample distinctions checked in the supplied Deary book, PDF 119–122; a new full final-paper audit is not claimed. Five-year English cohort, age-11 cognitive assessment and age-16 GCSEs. University of Edinburgh record. Reviewed 20 October 2026. ↩

48. Raj Chetty, David J. Deming, and John N. Friedman, Diversifying Society’s Leaders? The Determinants and Causal Effects of Admission to Highly Selective Private Colleges, NBER Working Paper 31492, October 2023 revision. Exact archived revision. Relevant full-paper sections and Appendix Table 4, PDF 88, inspected; table visually checked. Table panels provide marginals and national score distributions conditional on income, not the required Ivy income-by-score joint distribution. The book’s live Opportunity Insights URL now serves an August 2025 version; it was not silently substituted. Reviewed 20 October 2026. ↩

49. Peter M. Visscher and colleagues, “Assumption-Free Estimation of Heritability from Genome-Wide Identity-by-Descent Sharing between Full Siblings,” PLOS Genetics 2(3) (2006), e41, DOI 10.1371/journal.pgen.0020041. Full primary article available; relevant design, results, and limitations inspected. Harden’s worked example concerns height and realized within-family genetic sharing; it does not estimate an educational treatment. “Assumption-free” refers to the contested between-family resemblance assumptions, not absence of every modeling assumption. Original article. Reviewed 20 October 2026. ↩

50. Laurence J. Howe and colleagues, “Within-Sibship Genome-Wide Association Analyses Decrease Bias in Estimates of Direct Genetic Effects,” Nature Genetics 54 (2022), 581–592, DOI 10.1038/s41588-022-01062-7. Publisher-indexed primary abstract and full-text sections inspected; direct opening intermittently failed. Locators: abstract, main description, within-sibship comparisons, discussion, and methods. Family/demographic contributions can inflate population estimates; sibling indirect effects remain a limitation. No exact sample count is used because the abstract and body report slightly different totals. Original article. Reviewed 20 October 2026. ↩

51. Stuart J. Ritchie and Elliot M. Tucker-Drob, “How Much Does Education Improve Intelligence? A Meta-Analysis,” Psychological Science 29(8) (2018), 1358–1369, DOI 10.1177/0956797618774253. Published author-hosted primary paper inspected, especially discussion pp. 1366–1367. Forty-two datasets; three quasi-experimental designs. Effects on measured scores vary by design; local effects, uncertain additivity, and latent-g limitations prevent a general transformation claim. Full paper. Analyst-selected counterevidence, not a finding attributed to Deary’s book. Reviewed 20 October 2026. ↩

52. Glen G. Cain and Douglas A. Wissoker, “A Reanalysis of Marital Stability in the Seattle-Denver Income-Maintenance Experiment,” American Journal of Sociology 95(5) (1990), 1235–1269, DOI 10.1086/229428. Final abstract and full preceding institutional working papers inspected: IRP 857-88, introduction, methods, Tables 3–4 and conclusion; IRP 870-88, treatment/sample analysis. Counterargument: Hannan and Tuma’s 1990 reply; Cain and Wissoker’s rejoinder. The final reply/rejoinder were inspected at abstract/identity level, not full text. These are interpretations of the same experiment, not three independent trials. Reviewed 20 October 2026. ↩

53. Chris Riddell and W. Craig Riddell, “What Did We Learn from the North American Income Maintenance Experiments? New Data and Evidence on Two-Parent Families,” Journal of Public Economics 259 (July 2026), 105658, DOI 10.1016/j.jpubeco.2026.105658. Final publisher abstract, highlights, and introduction inspected; final tables unavailable. Final publisher record. Full September 2025 predecessor inspected for methods, §5(c), and Table 9: CLEF Working Paper 91. Final and working-paper versions form one research chain. No predecessor coefficient is labeled a verified final estimate. Reviewed 20 October 2026. ↩

54. Robert Martinson, “New Findings, New Views: A Note of Caution Regarding Sentencing Reform,” Hofstra Law Review 7(2) (1979), 243–258. Original full article inspected, especially printed pp. 244, 252–256, footnote 6 and Table III/its note. The expanded 555-study corpus includes routine follow-ups; exploratory comparisons are not uniformly controlled treatment estimates. Official journal record; accessible original scan. Distinct from the 1974 article’s 231-study corpus. Reviewed 20 October 2026. ↩

55. Seena Fazel, Connie Hurton, Matthias Burghart, Matt DeLisi, and Rongqin Yu, “An Updated Evidence Synthesis on the Risk-Need-Responsivity (RNR) Model: Umbrella Review and Commentary,” Journal of Criminal Justice 92 (2024), 102197, DOI 10.1016/j.jcrimjus.2024.102197. Full primary paper inspected, methods/results/discussion: 26 meta-analyses with overlapping underlying studies; inconsistent support, evidence-quality and causal-interpretation concerns. Publisher article; full published paper. This critique of RNR validation is not a blanket refutation of rehabilitation. Reviewed 20 October 2026. ↩

56. Mark W. Lipsey, Nana A. Landenberger, and Sandra J. Wilson, “Effects of Cognitive-Behavioral Programs for Criminal Offenders,” Campbell Systematic Reviews 3 (2007), 1–27, DOI 10.4073/csr.2007.6. Full publisher text inspected, inclusion criteria, study-design table, results, and limitations. Fifty-eight studies include 19 randomized comparisons plus matched and other designs; the prose uses the favorable average direction without adopting the book’s unresolved “10–12 percent” convention. Original review. Overlaps with related syntheses are not independent confirmations. Reviewed 20 October 2026. ↩

57. Roland G. Fryer Jr., “Teacher Incentives and Student Achievement: Evidence from New York City Public Schools,” Journal of Labor Economics 31(2) (2013), 373–407, DOI 10.1086/667757. Final abstract and full November 2011 author version inspected. School-level randomization, schoolwide bonus package, high-poverty New York schools, outcomes 2007–2010. Final article; author paper. Candidate mechanisms were not independently randomized. Reviewed 20 October 2026. ↩

58. Paul Glewwe, Michael Kremer, and Sylvie Moulin, “Many Children Left Behind? Textbooks and Test Scores in Kenya,” American Economic Journal: Applied Economics 1(1) (2009), 112–135, DOI 10.1257/app.1.1.112. Final primary abstract and institutional researcher account inspected; full final paper not obtained. Western Kenyan primary-school trial: average null, gains among stronger pretest performers. Published article; researcher-institution study account. Language/curriculum explanation has a different evidentiary status from the randomized package effect. Reviewed 20 October 2026. ↩

59. Abhijit Vinayak Banerjee, Esther Duflo, Rachel Glennerster, and Dhruva Kothari, “Improving Immunisation Coverage in Rural India: Clustered Randomised Controlled Evaluation of Immunisation Campaigns With and Without Incentives,” BMJ 340 (2010), c2220, DOI 10.1136/bmj.c2220. Full primary text inspected, methods and Table 2. Rural Udaipur; approximately 18-month follow-up; endpoint children aged 1–3. Counts yield 38.743%, 17.942%, 5.814% in the stated arms; an independent reviewer recomputed them. Book/article mismatch remains unresolved. Arm denominators are used because the article contains inconsistent aggregate totals. Original article. Reviewed 20 October 2026. ↩