for five thousand years the world widened reading and never widened producing.
Bucket Foundation exists to reform education. Reform that is not grounded in evidence is opinion — so we built the evidence first. The education-atlas holds 78,326 observations across 219 countries; its flagship synthesis, the Knowledge-Access Gradient, is the founding research below. Every number on this page traces to it.
a broad shallow base and a thin, gated, unequal peak.
three crises, stacked
of 10-year-olds worldwide cannot read a simple text — 86.5% in Sub-Saharan Africa. These are children in school. The learning crisis is the deepest.
51.2M primary-age + 61.2M lower-secondary-age children out of school. The access crisis shrank but moved up a level.
of GDP is the world's education spend — below the agreed 4% floor, with 92 of ~200 countries beneath it. The most actionable lever.
the access cliff · depth, not age, is the binding constraint
Lay the world on a six-rung ladder of knowledge depth and access falls off a cliff. World-average access by depth — a ~270× fall from undergraduate to the frontier, the rich-poor gap widening from under 2× at literacy to ~75× at the frontier:
| depth | world access |
|---|---|
| L0 — basic literacy | 82.5% |
| L1 — K-12 / secondary | 62.4% |
| L2 — undergraduate | 37.4% |
| L3 — graduate / professional | 8.1% |
| L4 — read the research frontier | 0.14% |
| L5 — produce new knowledge | 0.06% |
of humanity only ever consumes knowledge; ~0.14% ever reaches the place where it is produced. The single most dramatic number in the corpus.
Every prior knowledge technology — writing, the printing press, mass schooling, the internet, open access — widened access to CONSUME knowledge. NONE widened access to PRODUCE it.
The mechanism is one empty cell. Order the channels people learn through by reach: the ones that scale (informal/open courses, self-directed study) all ceiling out below the frontier; the one that reliably reaches production — apprenticeship, the lab — cannot scale, because the supply of mentors is bounded by the existing producer population. A channel that is both scalable and production-reaching has been empty for all of recorded history. That empty cell is the consume-versus-produce gap. AI is the first technology in the entire arc — and the first modality — for which that verdict is not yet written. The number to watch is whether the produce-access rate ever moves.
reform the layer underneath the schooling.
The highest-leverage region the research names is the L3→L4→L5 zone — the comprehension bridge from established knowledge to the frontier, an author-routed production economics, and non-institutional, any-age frontier access. That is the prize the modality data calls the first channel that scales all the way to production. Bucket’s levers map onto exactly that region:
Free-to-read foundations — axioms, real math, laws, primary derivations — across seven branches. Attacks the paywall/credential gate that keeps <1% of people ever reading primary research, and the un-capped ceiling for those who could go furthest.
read the canon →An AI amplifier for self-directed learning at the frontier — instruments that let a motivated person produce research, not only consume it. The bridge from 'finished the courses' to 'can read and contribute to primary research' that no current product aims at.
open the research hub →Paid-to-cite, routed to the author — not the publisher. Attacks the one gate the open-access movement left fully standing: academic publishers earning ~38% margins on donated labor, selling publicly funded work back to the public.
the cite-forever license →Spaced-repetition mastery over the canon's seven branches — and a first-class 'Learning to learn' branch that TEACHES the highest-leverage missing skill the research names (metacognition, retrieval, spacing) as content you master, where 84% of students reread and 72% wrongly believe massing beats spacing. Most systems only USE this science; Bucket also teaches it.
open the academy →Non-institutional, any-age access to the full boundaries of knowledge, plus an open path to extend them — for the small number of people who can take a model and an axiom and reach a layer of reality nobody has reached before.
contribute →one continuous climb, no gap between the rungs.
The levers above are not five separate products — they are rungs on one ladder. The same L0→L5 depth scale that produced the access cliff is the path out of it: Academy mastery carries a learner up the consume side (L1–L2), the canon bridges to the frontier (L3–L4), and the research tools and the research agent are the produce-side terminus (L4–L5). The point is that there is no gap: “I learned mitochondrial bioenergetics” → “read the canon on it” → “use the research agent to do frontier work on it.” That continuity is Bucket’s answer to the empty scalable-and-production-reaching cell.
- L0basic literacyconsume82.5% world access
basic literacy / numeracy — read a simple text, do basic arithmetic
The ladder begins here; Bucket’s climb starts at the next rung.
- L1K-12 / secondaryconsume62.4% world access
K-12 / secondary schooling — the shared foundations everyone is meant to reach
- L2undergraduateconsume37.4% world access
undergraduate — systematic command of a field's established core
- L3graduate / proffrontier8.1% world access
graduate / professional — reading the field's own literature, not textbooks about it
- L4frontier (read primary research)frontier0.14% world access
frontier — reaching primary research: being a researcher, reading the primary literature
- L5producing new knowledgeproduce0.06% world access
producing new knowledge — adding to the frontier: publishing, deriving what was not there before
Bucket reforms the knowledge layer — how knowledge is accessed, produced, validated, and paid for. It is a foundation, not a ministry of education. It does not fix the schooling-logistics layer: K-12 funding inequity, the teacher pay penalty, the student-debt overhang, completion gaps, the access and financing crisis in low-income states, or the floor-level learning-poverty emergency. Those are state-capacity problems, and the proven levers there — Teaching at the Right Level, conditional cash transfers, the financing floor — are owned by states and NGOs.
Open knowledge is necessary but not sufficient: a reachable frontier is useless to a learner never taught to direct their own learning, and irrelevant to a child who is not in school at all. And reforming the knowledge layer is itself a value choice, not a verdict the data dictates — the research is explicit that there is no single cross-civilizational answer to what education is for. Bucket’s open-knowledge thesis is named in the atlas as one defensible option among contested aims, and we state it as one.
We are also honest about what is built versus what is scaffolded. The canon, the research tools, the academy, and the author-routed economics exist today. The autonomous research agent — the piece that would carry a self-directed learner across the production gate — and the turning flywheel that connects them are still being built. Bucket is a wager on the scalable-production channel that has never existed. AI is the first serious candidate to fill it; whether it does is the open question this whole mission is organized around, not a result we claim.
Every figure on this page is drawn from the education-atlas — the three crises (48.3% learning poverty, 51.2M + 61.2M out of school, 3.6% of GDP), the access cliff (L0 82.5% → L4 0.14%, ~270×), and the consume-vs-produce gap (99.86% / 0.14%) from THE-KNOWLEDGE-ACCESS-GRADIENT.md and EDUCATION_PROBLEMS.md; the levers and the honest boundary from REFORM_THESIS.md. Atlas sources: World Bank EdStats, UNESCO UIS, OECD PISA 2022, Our World in Data, OpenAlex.