Knowledge Gatekeeping, the Open-Knowledge Counter-Movement, and What Actually Works

Bucket Foundation landscape brief 04 — who controls access to advanced knowledge, who is trying to pry the gates open, and which interventions the evidence actually supports

The atlas's diagnosis (docs/EDUCATION_PROBLEMS.md) measured the problem; the access analysis (docs/landscape/02-access-data-science.md) showed the cliff is down the depth axis, bought by income — a ~270× fall from undergraduate (37%) to the frontier (0.14%), widening from a sub-2× rich-poor gap at literacy to ~75× at the frontier. The solution landscape (docs/landscape/01-solution-landscape.md) showed the grid's top-right (L4–L5) is the thinnest occupied region. The reform thesis (docs/REFORM_THESIS.md) put Bucket's wedge there: the knowledge layer — access to the frontier, fair production/validation, an un-capped path for self-directed learners. This brief fills three gaps those documents named but did not work through: (1) the political economy of who controls the gates and how they profit; (2) a scorecard of the open-knowledge movement that has been trying to break the locks; and (3) the intervention-effectiveness evidence — the layer prior docs explicitly flagged as missing — with real effect sizes and an honest account of what has weak or null support. It ends where the data points: the highest-leverage places to expand access to advanced knowledge specifically, the L3→L4→L5 region.

Part 1 — The political economy of knowledge gatekeeping

The access cliff in brief 02 is not an accident of nature. At each depth level there is an institution that controls the gate and, in most cases, profits from controlling it. The structure is worth naming level by level.

1.1 The shape of the gates, by depth level

The most concentrated, most profitable, and least defensible gate sits at L4 — the publishers — so it gets the most scrutiny here. But note the asymmetry the solution landscape already flagged: teaching tools stop at L3, and the only thing that reaches L4–L5 in the credentialing column is the credential, not the learning. The gates at L4–L5 are where access to advanced knowledge is actually decided.

1.2 The academic-publishing oligopoly — real numbers

Roughly half of all peer-reviewed articles are published by five firms: Elsevier (RELX), Springer Nature, Wiley, Taylor & Francis, and SAGE. Elsevier alone controls ~25% of the global scientific-publishing market [1].

The profitability is the headline. RELX, Elsevier's parent, reported 2024 group revenue of £9.43 billion and adjusted operating profit of just under £3.2 billion (up 10%) [2]. The figure that matters for this brief is the scientific, technical & medical (STM) division — i.e. Elsevier's journals — which reported revenue of £3.05 billion at an adjusted operating margin of 38.4% [1][2]. A ~37–40% operating margin places a journal publisher alongside Apple, Google, and Coca-Cola, and well above the margins of most industries [1]. The STM segment is ~28% of RELX revenue but ~31% of its profit [1] — the journals are disproportionately the profit engine.

1.3 Why the margins are an anomaly: the inputs are donated

A 38% margin is normal for a software monopoly that builds its own product. It is an anomaly for a publisher, because in scholarly publishing the firm pays for almost none of the value it sells:

• The content is donated. Authors are not paid for articles; they

typically assign copyright to the publisher for free.

• The quality control is donated. Peer review is performed for free by

other researchers. One 2021 estimate put the global value of reviewers' unpaid time at >$1.5 billion (US-based reviewers alone) in 2020, on >100 million hours of work — equivalent to over 15,000 person-years — with China- and UK-based reviewers adding hundreds of millions more [12].

• The editing is largely donated. Most editorial boards are academics,

often unpaid or nominally paid.

• The customer is frequently the original funder. University libraries —

funded by the same public and the same grant overheads that paid for the research — buy back access via subscriptions.

This is the public-funds round-trip: governments fund the research, researchers donate the writing and reviewing, and the public (via libraries, or directly via paywalls) pays again to read it. The reform thesis's "<1% of people ever read primary research" is the downstream symptom; the round-trip is the mechanism.

1.4 The paywall + APC squeeze (open access did not end the rent)

When open access (OA) threatened the subscription model, the industry did not lose the rent — it moved the rent to the front end via the article-processing charge (APC): instead of the reader paying, the author pays to publish. The economics:

• The global average APC is ~US$1,626, but real prices run from a few

hundred dollars to ~US$12,000 [3]. Nature's OA fee reached $12,850 (2026) [solution-landscape ref 48].

• Hybrid-journal APCs run roughly twice the level of pure-OA-journal

APCs [3] — and hybrid journals collect APCs while still selling subscriptions, the much-criticized "double dipping."

• Total APC spending nearly tripled from $910M (2019) to $2.54B (2023)

[3]. In 2023, the top APC earners were MDPI ($681.6M), Elsevier ($582.8M), and Springer Nature ($546.6M) [3].

The APC flip has a perverse equity effect that brief 02's income gradient predicts: a paywall excludes poor readers; an APC excludes poor authors. A researcher in a low-income institution who could previously at least publish for free now faces a four-figure charge to be read — so the gatekeeping simply migrated from the consumption side to the production side. (Important caveat in the other direction: ~60% of journals in the DOAJ charge no APC at all [3] — the APC problem is concentrated in the prestige and hybrid tiers, not universal.)

1.5 The non-publisher gates: prestige, credentials, licensure

Money is only part of the gate. The harder lock is prestige and credential lock-in:

• Journal-prestige lock-in (L5). Careers, tenure, and grants are

decided substantially by where you publish (impact factor, journal brand). This keeps researchers funneling free labor into the highest-rent journals even when cheaper or open venues exist — a coordination trap, not a pricing problem. Reform efforts (DORA — the San Francisco Declaration on Research Assessment) attack exactly this.

• University credentialing (L2–L5). The degree is the only mechanism

that institutionally certifies all the way to L5 (the PhD as a license to produce knowledge). The solution landscape noted 45% of employers dropped some degree requirements in 2024, but the degree still monopolizes high-end certification.

• Occupational licensure (L3). Licensing boards gate entry to

professions. The share of US workers needing a license rose from <5% in the 1950s to ~25% today [16]; Kleiner estimates licensing restrictions cost consumers ~$203 billion/year and ~2.85 million fewer jobs [16]. Licensure is a real consumer-protection tool and a rent-extraction and exclusion mechanism — both are true.

The political-economy summary: advanced knowledge is gated at four points — tuition (L2), credentials + licensure (L3), publisher paywalls/APCs (L4), and funding + affiliation + prestige (L5). The publisher gate is the most profitable and least defensible (38% margins on donated inputs selling publicly-funded work back to the public). The prestige/credential gates are softer but stickier, because they are coordination traps that no single actor can defect from alone.

Part 2 — The open-knowledge counter-movement: a scorecard

For thirty years a loose coalition has been trying to pry these gates open. It has changed the landscape — but unevenly, and with a recurring pattern: it solved access-to-read far better than it solved cost-shifting or the "last mile" of actually understanding and using what's now reachable. A balanced scorecard:

2.1 Open Access (the policy front) — partial win, with a cost-shift

• What happened: ~**50% of newly published literature is now OA in some

form [4] — a structural shift from a generation ago when nearly everything was paywalled. Funder mandates (Plan S** / cOAlition S, NIH/US-OSTP public-access policies) pushed this.

• What worked: OA demonstrably increases reach. The **OA citation/usage

advantage is robust, and — crucially — it is just as large when OA is mandated as when self-selected** [15], so mandates are not a watered-down lever. OA articles also sustain downloads far longer than paywalled ones [15].

• What didn't: Plan S's direct causal effect is murky — one

analysis found "little evidence Plan S made a significant difference," with gains often lagging comparison groups [4]. And the win came with the APC cost-shift (Part 1.4): access-to-read improved, but the rent moved to authors. cOAlition S itself ended financial support for "transformative agreements" after 2024, conceding they weren't flipping journals fast enough [4]. Diamond OA (no fee to author or reader, usually university/society-run) is the model that escapes the cost-shift, but it is under-funded and fragmented.

• Scorecard: B−. Half the literature is now free to read. But the

business model that funds it re-created the gate on the author side, and the strongest single policy push (Plan S) has weak measured causal effect.

2.2 Preprints (the velocity front) — clear win in two fields

• arXiv (2M+ papers) made physics/math/CS effectively open at the point

of production decades ago; bioRxiv/medRxiv (now nonprofit openRxiv, $16M CZI grant) did the same for the life sciences [solution-landscape refs 50–51]. Preprints route around the publisher gate entirely: the paper is free, immediately, before peer review.

• What worked: in their home fields, preprints are now the de facto

primary channel. What didn't: adoption is wildly uneven across disciplines (chemistry, much of medicine, the humanities lag), and a preprint is un-reviewed — the velocity gain trades against the quality signal peer review provides.

• Scorecard: A− in physics/CS/bio, incomplete elsewhere.

2.3 Shadow libraries (the civil-disobedience front) — factual note

Sci-Hub is the most-used answer to the access problem and an illegal one. Founded 2011 by Alexandra Elbakyan, it hosts ~88M papers with >90% coverage of paywalled content. Factually and legally: Elsevier won a 2017 US default judgment of $15M in damages and an injunction (which cost Sci-Hub its .org domain) [13]; further suits by Elsevier, Wiley, and ACS proceed in the Delhi High Court [13]; UK courts ordered ISP blocks [13]. The judgments are largely unenforceable — Elbakyan is outside US jurisdiction with no US assets [13]. The honest reading (echoing the solution landscape): the fact that the single largest "solution" to frontier access is an outright shadow library is itself the strongest indictment of the legal gate. Noted as illicit; not endorsed; included because omitting it would misrepresent the real access landscape.

2.4 Reference + encyclopedic commons — quiet, massive win

• Wikipedia (~15B monthly pageviews, 65M+ articles) is the most

successful open-knowledge project ever built — and it is the L0–L2 reference floor for the entire planet, donor-funded, no paywall, no APC. Its limit is exactly the atlas's depth axis: it summarizes settled knowledge brilliantly (L0–L2) and reaches the frontier (L4) only as citations, not as understanding.

• Scorecard: A for breadth, by-design silent above ~L2.

2.5 OER — Open Educational Resources — proven cost win, modest learning win

• OpenStax (Rice University, free peer-reviewed textbooks): **6M+

students, >$1.5 billion in cumulative textbook savings; one 2024 grant cohort alone served 422,000 students and saved ~$33.4M [14]. MIT OpenCourseWare** put a full curriculum online for free two decades ago.

• What worked: the cost result is unambiguous and large, and OER

shows improved end-of-course grades especially for Pell recipients and part-time students [14] — i.e. it helps exactly the income-gradient tail brief 02 identified. What didn't: OER addresses the price of established (L1–L2) content; it does not reach the frontier and does not by itself solve the "can the learner actually understand it" problem.

• Scorecard: A on cost/access for L1–L2; not a frontier lever.

2.6 Open infrastructure (the plumbing) — strong, under-celebrated win

OpenAlex (474M works, CC0, now powers the Leiden Ranking), ORCID (researcher IDs), ROR (institution IDs), Semantic Scholar (200M+ papers), Unpaywall, CORE, and Creative Commons licensing together built an open metadata + discovery substrate that did not exist a decade ago. This is the layer most AI research tools are quietly built on (brief 01 §1.6c). It is the clearest, most durable open-knowledge win because it is infrastructure, not a single product — and it is the layer Bucket's own research-atlas already sits on.

2.7 Citizen science (the participation front) — real but bounded

Zooniverse, eBird, Foldit and others showed the public can contribute to knowledge production (L5) at scale — but almost always in the data- collection / classification stage, not hypothesis generation or theory. It proved participation is possible; it did not open the interpretive frontier to non-specialists.

2.8 The open-knowledge scorecard, in one table

The honest cross-cutting verdict: the movement won access-to-read (OA + preprints + shadow libraries) and won the plumbing (open infra + OER cost). It did not win the economics (APCs re-created the gate on the author side; prestige lock-in is untouched) and it did not win the last mile — access is not the same as the ability to understand and use what's now reachable. A 16-year-old can now legally read a huge share of primary research; almost nothing helps that 16-year-old understand it or get to it from where their education left them (the L3→L4 bridge brief 01 named as unbuilt). That last-mile gap is the hinge for Part 3 and the conclusion.

Part 3 — What actually works: the intervention-effectiveness evidence

Brief 01 §3 flagged this as the missing layer: knowing where solutions cluster is not the same as knowing which solutions work. Here is the strongest causal evidence, with effect sizes (in standard deviations, SD, the field's common unit) and an honest separation of what the evidence supports from what it doesn't.

A calibration note first: in education, an effect of ~0.10 SD is small, ~0.30 SD is substantial, and >0.40 SD is large and rare at scale. The EEF's months-of-progress unit is a useful translation (roughly: ~0.1 SD ≈ +1 month, ~0.3 SD ≈ +3–4 months).

3.1 The strongest developing-world lever: Teaching at the Right Level (TaRL)

Pratham's Teaching at the Right Level — group children by learning level rather than age/grade, focus on foundational reading and arithmetic, assess simply and continuously — is the best-evidenced intervention in the developing-world education literature. Across six randomized evaluations in seven Indian states (J-PAL / Banerjee, Duflo, et al.), it "has led to some of the largest effect sizes rigorously measured in the education literature" [5].

• Early "Balsakhi" version: 0.14 SD year 1, 0.28 SD year 2 [5].
• The intensive "Learning Camp" model: **~0.71 SD reading, ~0.69 SD

arithmetic vs control — roughly 1.85 years of learning gain from a 30-hour program** [5].

This directly attacks the atlas's deepest problem (48% learning poverty) and it is cheap — which is why it is the canonical "what works" case. The caveat the evidence itself insists on: it works "when implemented systematically" [5]; fidelity at government scale is the hard part, and results attenuate when implementation slips.

3.2 The most robust general lever: tutoring

Tutoring is the most consistently effective intervention across the entire evidence base. A comprehensive 2024 meta-analysis (AERJ) found a pooled effect of ~0.37 SD [6] — equivalent to moving a median student from the 50th to the ~66th percentile, or roughly 3–15 months of additional learning depending on grade [6]. Dosage is decisive: high-dosage tutoring is ~15–20× more effective than low-dosage [6].

The honest scaling caveat: the 0.37 SD comes largely from smaller, well-run studies. When scaled to programs serving 500–7,000 students, the average drops to ~0.25 SD [6] — still substantial, but the "2-sigma" (2.0 SD) Bloom promise that animates the AI-tutoring wave (brief 01 §1.2) is not what real tutoring delivers at scale. ~0.25–0.37 SD is the defensible number; 2.0 SD is aspirational.

3.3 Conditional cash transfers (CCTs) — work for enrollment, less for learning

CCTs (pay families to keep children enrolled/attending) are well-evidenced for access: meta-analysis across 15 developing countries finds primary enrollment effects of ~5 percentage points (a ~6% relative increase), larger at the secondary level [7]. More generous transfers → larger effects [7]. The crucial honesty: CCTs reliably get children into seats; their effect on learning is much weaker — they attack the access crisis, not the learning crisis, and the two are distinct (atlas §4). (The literature also flags publication bias in CCT effect estimates [7].)

3.4 The meta-evidence verandah: EEF, J-PAL, What Works Clearinghouse

Three clearinghouses aggregate this evidence and are the right external authorities to defer to:

• EEF (Education Endowment Foundation) — the Teaching & Learning

Toolkit, 33+ strands ranked by impact (in months), evidence strength, and cost [8][9].

• J-PAL — randomized-evaluation evidence, especially developing-world

[5][7].

• What Works Clearinghouse (US Dept of Education) — US program evidence

with strict design standards.

Their consensus on high-leverage, cost-effective levers: feedback, metacognition/self-regulation, mastery learning, peer tutoring, and reading comprehension strategies consistently rank high-impact / low-cost in the EEF Toolkit [9]. Notably, metacognition and self-regulation (≈ +7 months in the EEF Toolkit) is exactly the "learning-to-learn" capacity the reform thesis (deep/03) flagged as near-absent in actual products — a rare case where the strongest evidence and the emptiest market cell coincide.

3.5 What does NOT work (or works weakly) — the honest column

A balanced what-works document must say what the evidence fails to support. Popular interventions with weak, null, or poor cost-effectiveness:

• **Learning styles ("meshing" instruction to a visual/auditory/kinesthetic

preference): NULL. Rigorously tested and consistently disproven; effect sizes "very low and non-significant"; classed as a neuromyth** with no replicable evidence [10]. It remains widely believed by teachers — a cautionary tale about how long debunked ideas persist.

• Class-size reduction: real but small and expensive. EEF: ~+1 month

on average, and small reductions (30→25) are "unlikely to be cost- effective relative to other strategies" [8]. Modest benefit, poor cost-per-outcome, larger only for the youngest children [8].

• Technology-in-the-classroom: mixed / context-dependent. OECD: even

where computers are used, the impact on performance is "mixed, at best"; meta-analyses give small positive average effects that flip on implementation — collaborative/formative use helps, drill-and-monitor use often doesn't; intelligent-tutoring-system effects range from ~0.44 SD to no difference [11]. Technology is an amplifier of pedagogy, not a substitute — a direct caution for any AI-tutoring claim.

• Teaching assistants (untargeted): near-zero; but **targeted, trained

TAs delivering structured small-group/1:1 interventions are effective** (EEF) — the deployment, not the headcount, is what matters.

3.6 The what-works evidence, in one table

The what-works through-line: the levers with the strongest, most cost- effective evidence are structured, targeted, and human-mediated — teaching at the right level, tutoring at sufficient dosage, building metacognition, tight feedback. The levers that disappoint are the ones that add resources without changing the interaction (smaller classes, tech for its own sake) or that rest on folk theory (learning styles). For Bucket this is a discipline: any intervention should be judged against ~0.3 SD as the bar for "substantial," not against the 2-sigma marketing of the AI-tutoring field.

Part 4 — Conclusion: the highest-leverage points to expand access to advanced knowledge (the L3→L4→L5 region)

Combine the three parts. Who controls the gates: publishers at L4 (38% margins on donated, publicly-funded inputs), credentials + licensure at L3, funding + affiliation + prestige at L5. What the open movement has won: access-to-read (OA, preprints, shadow libraries) and the open plumbing (OpenAlex/ORCID/OER) — but not the economics (APC cost-shift, prestige lock-in) and not the last mile (access ≠ comprehension; the L3→L4 bridge is unbuilt). What works: structured, targeted, human/AI-mediated interaction (TaRL, tutoring, metacognition) at ~0.3 SD; not resources- without-interaction or folk theory.

The highest-leverage points to expand access to advanced knowledge specifically — the L3→L4→L5 region the whole atlas converges on — are the intersections where a won open-access asset meets an unsolved gate meets a what-works mechanism:

1. The L3→L4 comprehension bridge (the last mile) — highest leverage.

Access-to-read at L4 is largely won; comprehension of it is not, and no product bridges the crowded L3 upskilling column up to the frontier (brief 01 §2). The strongest what-works levers — tutoring (0.37 SD) and metacognition (+7 months) — are precisely the mechanisms that could carry a motivated L3 learner into reading primary research, applied to a target (the frontier) no current tutoring product aims at. This is the single point where the emptiest market cell, a won access asset (OA + open infra), and the best-evidenced mechanism all coincide.

1. **The author-routed production economics (the unwon gate) — most

defensible white space.* The open movement decisively failed to fix the economics: APCs re-created the gate on the author side and prestige lock-in is untouched. This is orthogonal to what Elicit/FutureHouse do (they make doing research easier; none changes who gets paid when knowledge is cited). A paid-to-cite, author-routed layer (feed402/x402 in the reform thesis) attacks the one gate the counter-movement left fully standing — and it sits on the won* open-infra substrate (OpenAlex/ORCID/ROR) rather than fighting it.

1. Non-institutional, any-age frontier access (the excluded user). Every

L4–L5 tool assumes an academic affiliation. The OA citation/usage advantage shows the content gate is opening; the user gate (who the tools are built for) is not. Building the L3→L4 bridge and the author-routed economics for the non-institutional, self-directed, and young-and-capable learner is the user-side white space that brief 01 §4 identified — and it is complementary to, not competitive with, the open-access wins this brief scored.

What this conclusion deliberately does not claim (consistent with REFORM_THESIS §3): none of this touches the learning crisis (48% can't read at 10), the access/financing crisis in low-income states, or the floor-level emergency — those are state-capacity problems, and TaRL/CCTs are the proven levers there, owned by states and NGOs, not a knowledge-layer foundation. Bucket's defensible, evidence-aligned contribution is the top of the depth axis: open the L4 reading gate further, build the L3→L4 comprehension bridge with the what-works mechanisms (tutoring + metacognition), and fix the one economic gate the open movement never closed (author-routed production/validation) — for the users the incumbents exclude.

Sources

[1] Elsevier STM margin (38.4%) & market share — Journalology, "Elsevier: 2025 in review" — https://newsletter.journalology.com/p/elsevier-2025-in-review ; PublishingState, "How Much Profit Did Elsevier Make in 2024?" — https://publishingstate.com/how-much-profit-did-elsevier-make-in-2024/2025/ [2] RELX FY2024 results (£9.43B revenue, ~£3.2B profit, +10%) — Research Professional News — https://www.researchprofessionalnews.com/rr-news-world-2025-2-elsevier-parent-company-reports-10-rise-in-profit-to-3-2bn/ ; RELX 6-K (SEC) — https://www.sec.gov/Archives/edgar/data/0000929869/000092986924000056/tmb-20240725xex99d1.htm [3] APC economics (avg ~$1,626; $910M→$2.54B 2019–2023; MDPI/Elsevier/Springer top earners; ~60% DOAJ journals charge no APC) — OPUS Project, "Estimating Global APCs Paid to Six Publishers 2019–2023" — https://opusproject.eu/openscience-news/estimating-global-article-processing-charges-paid-to-six-publishers-for-open-access-2019-2023/ ; Wikipedia, "Article processing charge" — https://en.wikipedia.org/wiki/Articleprocessingcharge [4] OA ~50% of new literature; Plan S weak measured effect; transformative-agreement support ended after 2024 — Science/AAAS, "A mixed review for Plan S" — https://www.science.org/content/article/mixed-review-plan-s-s-drive-make-papers-open-access ; cOAlition S — https://www.coalition-s.org/coalition-s-confirms-the-end-of-its-financial-support-for-open-access-publishing-under-transformative-arrangements-after-2024/ [5] TaRL effect sizes (0.14/0.28 SD Balsakhi; ~0.71/0.69 SD Learning Camps; 1.85 yrs/30 hrs; 6 RCTs) — J-PAL, "Teaching at the Right Level to improve learning" — https://www.povertyactionlab.org/case-study/teaching-right-level-improve-learning ; J-PAL evidence page — https://www.povertyactionlab.org/evidence-effect/teaching-at-the-right-level [6] Tutoring meta-analysis (0.37 SD pooled; 0.25 at scale; high-dosage 15–20× low) — AERJ 2024 via TutorBase summary — https://tutorbase.com/statistics/tutoring-effectiveness ; NBER w27476 (96-RCT review) — https://www.nber.org/system/files/workingpapers/w27476/w27476.pdf ; EdWorkingPapers tutoring meta-analysis (Oct 2024) — https://edworkingpapers.com/sites/default/files/Tutoring%20Meta-Analysis%20Oct%202024unblinded.pdf [7] CCT enrollment effects (~5 pp primary, larger at secondary; publication bias) — García & Saavedra, "Educational Impacts and Cost-Effectiveness of CCTs: A Meta-Analysis," Review of Educational Research 2017 — https://journals.sagepub.com/doi/10.3102/0034654317723008 ; RAND WR921-1 — https://www.rand.org/pubs/workingpapers/WR921-1.html [8] EEF — class-size reduction (~+1 month, not cost-effective at small reductions) — https://educationendowmentfoundation.org.uk/education-evidence/teaching-learning-toolkit/reducing-class-size [9] EEF Teaching & Learning Toolkit (33 strands by impact/cost/evidence; metacognition ≈ +7 months) — https://educationendowmentfoundation.org.uk/education-evidence/teaching-learning-toolkit ; structural-learning summary — https://www.structural-learning.com/post/eef-teaching-learning-toolkit-guide [10] Learning styles null / neuromyth — Frontiers meta-analysis of the matching hypothesis — https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2024.1428732/full ; APA, "Toward a Deeper Understanding of the Learning Style Myth" — https://www.apa.org/pubs/journals/releases/edu-edu0000366.pdf [11] Ed-tech mixed evidence — OECD, "The impact of digital technologies on students' learning" (2025) — https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/09/the-impact-of-digital-technologies-on-students-learning14095366/9997e7b3-en.pdf ; OECD 2015 (Schleicher, "mixed at best"); Ran et al. meta-analysis (JCAL 2022) — https://onlinelibrary.wiley.com/doi/abs/10.1111/jcal.12611 [12] Peer-review unpaid labor (>$1.5B US reviewers, >100M hours, 2020) — Aczel, Szaszi & Holcombe, "A billion-dollar donation," Research Integrity and Peer Review (2021) — https://link.springer.com/article/10.1186/s41073-021-00118-2 [13] Sci-Hub legal status ($15M 2017 US judgment, injunction, Delhi HC suit, UK ISP blocks, unenforceable) — Wikipedia, "Sci-Hub" — https://en.wikipedia.org/wiki/Sci-Hub ; C&EN, "Lawsuits progress against Sci-Hub" — https://cen.acs.org/articles/95/i27/Lawsuits-progress-against-Sci-Hub.html [14] OER / OpenStax (6M+ students, >$1.5B saved; 422K students / $33.4M in one 2024 cohort; grade gains for Pell/part-time) — AVDF, "OpenStax Textbooks Transforms Access and Affordability" — https://www.avdf.org/news/openstax-textbooks-transforms-access-and-affordability-in-higher-ed/ [15] OA citation/usage advantage; mandate = self-selected — Gargouri et al., "Self-Selected or Mandated, Open Access Increases Citation Impact," PLOS ONE 2010 — https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0013636 ; Wang et al., "The open access advantage," Scientometrics 2015 — https://link.springer.com/article/10.1007/s11192-015-1547-0 [16] Occupational licensing (<5% 1950s → ~25% today; ~$203B/yr, ~2.85M fewer jobs — Kleiner) — White House CEA, "Occupational Licensing: A Framework for Policymakers" (2015) — https://obamawhitehouse.archives.gov/sites/default/files/docs/licensingreportfinal_nonembargo.pdf ; Brookings (Kleiner) — https://www.brookings.edu/articles/occupational-licensing-and-the-american-worker/

Method: grounded with web research (June 2026) and cited above; complements rather than repeats `docs/landscape/01-solution-landscape.md` (the player census), `docs/landscape/02-access-data-science.md` (the access measurement), `docs/EDUCATIONPROBLEMS.md (the diagnosis), and docs/REFORMTHESIS.md` (the wedge). Effect sizes are reported in standard deviations (SD) or the EEF months-of-progress unit; figures for private firms are company-reported or third-party estimates as noted. Sci-Hub is documented factually as illicit and is not endorsed.