The Education Solution Landscape — mapped on the Age × Knowledge-depth grid

Bucket Foundation landscape brief 01 — who is already solving what, and where the grid is empty

The atlas's diagnosis (docs/EDUCATION_PROBLEMS.md) and four deep-dives (docs/deep/01–04) say what is broken. The reform thesis (docs/REFORM_THESIS.md) says Bucket's wedge is the knowledge layer — access to the frontier, fair production and validation of it, and an un-capped path for self-directed learners and the few who can extend it. This brief tests that wedge against the market: it catalogs 83 real players across the whole education stack, places each on a 2-D grid, and asks honestly where the grid is crowded and where it is empty. The structured catalog is data/landscape/solutions.csv (12 columns, one row per player, each with a source URL). This document is the map and the reading of it.

0. The frame: a 2-D grid

Every solution is placed on two axes.

X — Age / life-stage (a learner's position in life):

Y — Knowledge depth (how far into a body of knowledge the tool takes you):

The catalog records each player's age range (age_min/age_max) and the depth band it serves (depth_min_L/depth_max_L). The thesis to test, stated up front so it can be falsified: *the top of the Y-axis (L4–L5) at any age — both reaching the frontier and being able to do research there — is the least served region of the grid.* The data below tests it; §4 reports the verdict honestly, including where the data complicates the story.

The catalog is not exhaustive of every edtech company — there are thousands. It is a representative census of the category leaders and the structurally distinct players in each cell, which is what a coverage map needs.

1. The grid, column by column (who serves which cell)

1.1 Early-childhood (0–5), depth L0

Crowded and well-capitalized — and entirely at the literacy/numeracy floor. ABCmouse (Age of Learning; ~50M children lifetime, >$750M raised, a $1B+ valuation) [1], Lingokids (185M families reached, ~$186M raised including a $120M 2025 Series D) [2], and Homer/Begin [3] are paid consumer subscriptions; Khan Academy Kids [4] and Duolingo ABC [5] are free loss-leaders. All five do roughly the same thing — gamified phonics, counting, early SEL for ages 2–8 — and none reaches above L0. The cell is commercially well-served; whether it serves the learning-poverty problem (48% of 10-year-olds can't read, per the atlas) is a different question — these are paid apps in homes that can afford a tablet and a subscription, i.e. structurally absent from exactly the low-income, fragile contexts where the atlas locates the deepest crisis (§3, EDUCATION_PROBLEMS).

1.2 K-12 (5–18), depth L0–L2

The single most crowded region of the entire grid. Five overlapping sub-markets:

• Free/nonprofit content + AI tutor: Khan Academy ($117.5M FY25 revenue,

77% philanthropic) [6] and its Khanmigo AI tutor ($4/mo for learners, free to US teachers via Microsoft) [7] — the closest thing to a universal free K-12 spine.

• Adaptive practice (institutional): IXL (17M+ students) [8], i-Ready

(~13M US students, ~1/3 of K-8) [9], DreamBox [10], Prodigy [11].

• Tutoring & homework help: Varsity Tutors/Nerdy (public, ~$190M FY25

revenue, 1,100+ districts) [12], Wyzant, Chegg (collapsing under AI — Q1'25 revenue −30%, 45% layoffs Oct 2025) [13], Photomath (300M+ users, owned by Google) [14].

• AI tutors: Khanmigo, Synthesis Tutor (math, ages 5–11) [15], Sizzle AI

(1.7M users, acquired by Campus 2025) [16], Quizlet (60M+ users) [17].

• Plumbing (LMS/SIS): Google Classroom (150M+ users) [18], Canvas/Instructure

(taken private by KKR for $4.8B) [19], Schoology/PowerSchool (Bain, $5.6B) [20].

This is where the AI-tutoring "2-sigma" promise is being fought (see docs/foundations/04, §2). The cell is saturated; the unmet problems here are the ones no product targets — learning-to-learn, philosophy, agency (docs/deep/03) — not a lack of content-delivery vendors.

1.3 Higher-ed (18–22) + Professional (22–65), depth L2–L3

Also crowded, and consolidating. Coursera ($757M FY25 revenue, 197M learners) is acquiring Udemy ($790M FY25 revenue) for ~$2.5B [21][22] — merging the biggest MOOC platform with the biggest open course marketplace. The debt-and-ISA-financed cohort broke: 2U/edX went through Chapter 11 in 2024 [23], Pluralsight was handed to its lenders after Vista's ~$4B writeoff [24], Udacity was absorbed into Accenture [25], and BloomTech (ex-Lambda School) was banned from consumer lending by the CFPB for inflated job-placement claims [26]. Alongside them: LinkedIn Learning (Microsoft) [27], Skillshare [28], DataCamp [29], Codecademy (Skillsoft) [30], O'Reilly [31], Maven (cohort-based, ~75% completion vs MOOC ~7–10%) [32], and corporate-L&D LXPs Degreed [33] and Cornerstone [34]. Free open-courseware (MIT OCW [35], OpenStax [36]) sits underneath at no cost but offers no credential or interaction. This whole band tops out at L3: it teaches established undergraduate and professional knowledge, not the research frontier.

1.4 Lifelong / adult (any age, into 65+), depth L0–L3 (one tail to L4)

The broadest, shallowest column. Wikipedia (~15B monthly pageviews, 65M+ articles) [37] and US public libraries (~17K outlets, 155M+ users, 800M+ annual visits) [38] are the free reference floor. MasterClass ($2.75B valuation, edutainment) [39], Brilliant (STEM, 10M+ users) [40], and The Great Courses [41] are paid lifelong learning. Podcasts [42] are the one informal medium whose ceiling occasionally reaches L4 (a working scientist explaining their own field) — but with zero structure, assessment, or verification. This column is about breadth of access to settled knowledge, not depth toward the frontier.

1.5 Credentialing (cross-cutting), depth L2–L5 signal

Credentials sit on a different plane — they certify depth rather than teach it. Traditional university degrees are the only mechanism that institutionally certifies all the way to L5 (the PhD = a license to produce new knowledge) — but they are slow, expensive, and increasingly decoupled from competence (45% of employers dropped some degree requirements in 2024) [43]. Google Career Certificates (1M+ graduates) [44], Microsoft/AWS certs [45], and Credly/Open Badges (Pearson, 100M+ badges issued) [46] are the unbundled, lower-depth alternatives filling the credentialism crack. Note the asymmetry: teaching tools stop at L3; the only thing reaching L4–L5 in this column is the credential, not the learning.

1.6 The frontier column (L4–L5) — the heart of the test

This is the column the thesis hinges on, so it gets the most detail. It splits into four sub-layers, and the split itself is the finding.

(a) Paywalled access (L4) — the bottleneck. The "big five" commercial publishers — Elsevier/ScienceDirect (~2,900 journals, RELX, >30% margins) [47], Springer Nature (Nature OA fee $12,850 in 2026) [48], Wiley [49], Taylor & Francis [49] — gate the majority of peer-reviewed literature behind institutional subscriptions or per-article paywalls. This is the "<1% of people ever read primary research" problem from the reform thesis, made concrete: the frontier exists but is priced for institutions, not people.

*(b) Open access (L4–L5 reading) — partial relief. arXiv (2M+ papers) [50], bioRxiv/medRxiv (now nonprofit openRxiv on a $16M CZI grant) [51], PubMed/PMC (38M+ citations, NIH) [52], PLOS [53], DOAJ [54], and Unpaywall [55] make a growing share legally free to read. And Sci-Hub [56] — illegal, donation-funded, ~88M files, >90% coverage of closed content — is the honest acknowledgment that the access problem is so acute that the single largest "solution" is an outright shadow library. (Noted as illicit; not endorsed.) The frontier is increasingly readable* — but reading is L4, not L5.

(c) Discovery / open data (L4–L5 finding). Google Scholar [57], Semantic Scholar (AI2, 200M+ papers, the open substrate under most AI tools) [58], OpenAlex (474M works, CC0, now powers the Leiden Ranking) [59], CORE (400M+ resources) [60], ResearchGate (25M+ members) [61], and the visual mappers Connected Papers [62] / Research Rabbit [63] / Undermind (agentic deep search) [64] help you find frontier work. Strong, mostly free, well-built. Still: finding is not doing.

*(d) AI research tools (L4–L5 — the closest anyone gets to doing).* This is the newest and most contested layer (docs/foundations/04). On a spectrum of "how close to actually doing research":

Elicit (2M+ users, 50K+ paying) is the most mature tool that crosses from reading into doing a defined piece of research (a systematic review). FutureHouse (nonprofit; for-profit spinout Edison raised $70M seed) is the most ambitious at autonomous discovery. These two are the closest existing players to Bucket's wedge — and that proximity is exactly why the white-space verdict (§4) has to be careful.

2. The coverage map — crowded vs empty

Reading the catalog as a grid (counts are of distinct players whose range covers the cell; tools spanning bands are counted in each):

Quantitatively, of the 83 players: only 31 reach L4 or L5 at all, and they are concentrated in one column (frontier-access/research-tools); the L0–L3 diagonal — early-childhood literacy through professional upskilling — holds the clear majority of players and essentially all of the consumer-edtech capital.

Crowded cells: early-childhood L0; K-12 L0–L2 (the densest cell on the grid); higher-ed and professional L2–L3; frontier reading/discovery L4 (strong open-access + discovery infrastructure).

Empty / thin cells:

1. Frontier L4–L5 for non-institutional / younger learners. Almost every

L4–L5 tool is priced and designed for credentialed academics with an institutional affiliation. A motivated 16-year-old, or a curious adult with no university login, can read a lot (open access, Sci-Hub) but is not the user any of these tools were built for.

1. L5 "doing research" at any age. The cell is occupied by exactly two

serious players (Elicit for the review stage; FutureHouse/Edison for discovery) plus the general-purpose Perplexity/STORM — versus dozens of players at every L0–L3 cell. It is the thinnest occupied region of the grid.

1. The bridge from L3 → L4. Nothing in the crowded higher-ed/professional

column hands a learner up to the frontier. Coursera/edX/Udemy stop at established knowledge; the frontier tools assume you already arrived. The on-ramp from "I finished the courses" to "I can read and contribute to primary research" is unbuilt.

1. Lifelong/elder above L1 — almost nothing structured exists for an older

adult who wants to go deep, only breadth (libraries, MasterClass).

3. How the landscape maps onto the atlas's problems

Matching solution density to the problem inventory from EDUCATION_PROBLEMS.md and the deep-dives:

The pattern is sharp: the problems with the most solutions (personalization, content delivery, test prep) are not the problems the atlas ranks deepest, and the problems the atlas and deep-dives flag as most neglected (learning-to-learn, the ceiling, doing research at the frontier) are the emptiest cells. Markets chase the payable middle; the structural problems are upstream or upmarket of where the money is.

4. The white-space finding — testing the top-right thesis honestly

The thesis largely holds, with one important qualification.

Where it holds. The top-right of the grid — L5 ("producing new knowledge") at any age, and L4–L5 access for anyone outside an institution — is unambiguously the thinnest occupied region. Against dozens of players at every L0–L3 cell, the L5 cell has effectively two serious occupants. No product bridges the crowded L3 upskilling column up to the L4 frontier. And the frontier tools that exist are built for credentialed academics, not for the self-directed learner or the capable young person the reform thesis centers — which is precisely the "un-capped frontier" white space Bucket aims at.

The qualification — be honest about who is already there. The top-right is not virgin territory, and pretending otherwise would repeat the overreach the atlas was built to avoid:

• Reading the frontier is increasingly solved. Open access (arXiv,

bioRxiv/openRxiv, PMC, PLOS), discovery (Semantic Scholar, OpenAlex, Google Scholar), and — illicitly — Sci-Hub have made access to read primary research far less of a moat than it was. Bucket's "free-to-read primary research" lever is real but is entering a column with strong, mostly free, well-funded incumbents.

• *AI-assisted doing has two credible, fast-moving incumbents. Elicit*

already does the systematic-review stage at production scale (2M+ users, an API as of 2026); FutureHouse/Edison already demonstrated autonomous hypothesis generation that yielded a real drug-repurposing candidate, and raised $70M to commercialize it. The "40 research tools + research agent" lever in the reform thesis is aimed at a cell that is thin but no longer empty, and is being entered by very well-resourced players.

So the precise white space is narrower and sharper than "L4–L5 is empty." It is the intersection of three things that no current player occupies together:

1. Non-institutional, any-age access to the frontier (the existing tools

assume an academic affiliation and an academic user);

1. A bridge from established knowledge (L3) up to the frontier (L4–L5) —

the on-ramp that turns a finished-the-MOOCs learner into someone who can read and contribute to primary research; and

1. A fair production-and-validation layer — paid-to-cite, author-routed

economics (the feed402/x402 lever) — which is orthogonal to what Elicit and FutureHouse do. They make doing research easier; none of them changes who gets paid when knowledge is produced and cited. That economic layer is genuinely unoccupied.

The honest verdict: *the top-right is the least-served region (thesis confirmed), but the most defensible white space is not "AI tools that do research" — Elicit and FutureHouse are already there — it is the combination of open, non-institutional, any-age frontier access with a fair author-routed production-and-validation economics. Bucket should treat Elicit and FutureHouse as the proximate competitors on the "doing" axis and compete on who it serves (non-institutional, self-directed, young-and-capable) and how knowledge production is paid for* — not on out-building their agents.

5. One-page summary

• Catalog size: 83 players, data/landscape/solutions.csv.
• Most crowded: K-12 L0–L2 (the densest cell), early-childhood L0,

higher-ed/professional L2–L3, and frontier reading/discovery L4.

• Emptiest: L5 "doing research" (two serious players vs dozens elsewhere);

the L3→L4 bridge; non-institutional/any-age frontier access; lifelong learning above L1.

• Thesis verdict: the top-right (L4–L5 access + doing, at any age) is the

least-served region — confirmed — but reading-access is increasingly solved and the "doing" cell already has Elicit (review stage) and FutureHouse/Edison (discovery). The true, defensible white space is the intersection of non-institutional any-age frontier access + a fair, author-routed production-and-validation economics — which no incumbent occupies.

• Problem mapping: the most-funded responses (personalization, content

delivery) target problems the atlas ranks shallower; the atlas's deepest and most-neglected problems (learning-to-learn, the ceiling, doing research) coincide with the emptiest grid cells.

Sources

[1] Age of Learning / ABCmouse — https://www.crunchbase.com/organization/age-of-learning [2] Lingokids ($120M Series D) — https://bullhoundcapital.com/articles/bullhound-capital-leads-lingokids-120m-round/ [3] Homer / Begin — https://www.crunchbase.com/organization/learn-with-homer [4] Khan Academy Kids — https://www.khanacademy.org/kids [5] Duolingo ABC — https://play.google.com/store/apps/details?id=com.duolingo.literacy [6] Khan Academy — https://www.khanacademy.org [7] Khanmigo pricing — https://www.khanmigo.ai/pricing [8] IXL Learning — https://www.myengineeringbuddy.com/blog/ixl-learning-reviews-pricing-2026-honest-look/ [9] i-Ready / Curriculum Associates — https://www.curriculumassociates.com/about/press-releases/2025/09/back-to-school [10] DreamBox / Discovery Education — https://www.discoveryeducation.com/details/clearlake-capital-backed-discovery-education-completes-acquisition-of-dreambox-learning/ [11] Prodigy Math — https://www.crunchbase.com/organization/prodigy-game [12] Varsity Tutors / Nerdy (NYSE: NRDY) — https://www.businesswire.com/news/home/20260226513294/en/Nerdy-Announces-Fourth-Quarter-2025-Financial-Results [13] Chegg layoffs (AI disruption) — https://www.cnbc.com/2025/10/27/chegg-slashes-45percent-of-workforce-blames-new-realities-of-ai.html [14] Photomath (Google) — https://en.wikipedia.org/wiki/Photomath [15] Synthesis Tutor — https://www.synthesis.com/tutor [16] Sizzle AI (acq. by Campus) — https://www.prnewswire.com/news-releases/campus-acquires-ai-startup-founded-by-former-meta-ai-chief-302580557.html [17] Quizlet / Q-Chat — https://quizlet.com/blog/meet-q-chat [18] Google Classroom — https://research.com/education/how-google-conquered-the-classroom [19] Canvas / Instructure (KKR $4.8B) — https://www.instructure.com/press-release/instructure-to-be-acquired-by-KKR [20] PowerSchool / Schoology (Bain $5.6B) — https://www.baincapital.com/news/powerschool-be-acquired-bain-capital-56-billion-transaction [21] Coursera FY25 results — https://investor.coursera.com/news/news-details/2026/Coursera-Reports-Fourth-Quarter-and-Full-Year-2025-Financial-Results/default.aspx [22] Coursera–Udemy combination — https://investor.coursera.com/news/news-details/2025/Coursera-to-Combine-with-Udemy-to-Empower-the-Global-Workforce-with-Skills-for-the-AI-Era/default.aspx [23] 2U/edX Chapter 11 — https://www.insidehighered.com/news/business/2024/07/26/long-embattled-2u-declares-bankruptcy [24] Pluralsight restructuring — https://www.privateequitywire.co.uk/vista-and-co-investors-take-4bn-hit-in-pluralsight-private-credit-restructuring/ [25] Udacity / Accenture — https://newsroom.accenture.com/news/2024/accenture-completes-acquisition-of-udacity [26] BloomTech CFPB action — https://www.consumerfinance.gov/enforcement/actions/bloomtech-inc-and-austen-allred/ [27] LinkedIn Learning — https://jobright.ai/blog/complete-linkedin-learning-guide/ [28] Skillshare — https://grokipedia.com/page/Skillshare [29] DataCamp — https://app.dealroom.co/companies/datacamp [30] Codecademy / Skillsoft — https://www.skillsoft.com/press-releases/skillsoft-to-acquire-codecademy-a-leading-platform-for-learning-high-demand-technical-skills-creating-a-worldwide-community-of-more-than-85-million-learners [31] O'Reilly Learning — https://about.proquest.com/en/products-services/OReilly-for-Higher-Education/ [32] Maven — https://www.prnewswire.com/news-releases/maven-raises-20-million-in-series-a-funding-led-by-andreessen-horowitz-301295905.html [33] Degreed — https://joshbersin.com/2021/04/learning-tech-pioneer-degreed-gets-new-ceo-now-valued-at-1-4-billion/ [34] Cornerstone OnDemand — https://www.cornerstoneondemand.com/company/news-room/press-releases/cornerstone-ondemand-enters-definitive-agreement-to-be-acquired-by-clearlake-capital-group-in-dollar52-billion-transaction/ [35] MIT OpenCourseWare — https://ocw.mit.edu/ [36] OpenStax — https://openstax.org/ [37] Wikipedia / Wikimedia funding — https://wikimediafoundation.org/news/2025/11/26/how-is-wikipedia-funded/ [38] US public libraries (IMLS) — https://www.imls.gov/news/people-visited-public-libraries-more-billion-times-one-year [39] MasterClass — https://research.contrary.com/company/masterclass [40] Brilliant.org — https://en.wikipedia.org/wiki/Brilliant(website) [41] The Great Courses / Teaching Company — https://en.wikipedia.org/wiki/TheTeaching_Company [42] Podcasts (medium) — https://en.wikipedia.org/wiki/Podcast [43] Degree-requirement decline — https://fortune.com/2024/09/20/degree-requirements-employers-hiring-managers-paper-ceiling-education-careers-leadership/ [44] Google Career Certificates — https://blog.coursera.org/1-million-graduates-the-real-world-impact-of-google-career-certificates/ [45] AWS certifications (Pearson VUE) — https://www.pearsonvue.com/us/en/aws.html [46] Credly / Open Badges (Pearson, 100M+) — https://www.prnewswire.com/news-releases/with-100-million-digital-credentials-issued-through-credly-pearson-fosters-a-future-proof-workforce-for-enterprises-in-the-ai-era-and-beyond-302343746.html [47] Elsevier / ScienceDirect — https://www.elsevier.com/products/sciencedirect/journals/subscription-options [48] Springer Nature / Nature OA fee — https://www.statnews.com/2026/06/11/open-access-journal-fees-nature-wiley-elsevier-nih/ [49] Wiley & Taylor & Francis (publisher market share) — https://direct.mit.edu/qss/article/4/4/778/118070/ [50] arXiv — https://arxiv.org/ [51] bioRxiv/medRxiv → openRxiv — https://manusights.com/blog/preprint-servers-explained-biorxiv-medrxiv-arxiv [52] PubMed / PMC (NIH) — https://www.ncbi.nlm.nih.gov/pmc/ [53] PLOS — https://plos.org/ [54] DOAJ — https://doaj.org/ [55] Unpaywall — https://unpaywall.org/ [56] Sci-Hub (illicit; for reference) — https://en.wikipedia.org/wiki/Sci-Hub [57] Google Scholar — https://scholar.google.com/ [58] Semantic Scholar (AI2) — https://www.semanticscholar.org/ [59] OpenAlex — https://openalex.org/ [60] CORE — https://core.ac.uk/ [61] ResearchGate — https://www.researchgate.net/ [62] Connected Papers — https://www.connectedpapers.com/pricing [63] Research Rabbit — https://www.researchrabbit.ai/ [64] Undermind (YC) — https://www.ycombinator.com/companies/undermind [65] Consensus ($30M raise) — https://consensus.app/home/blog/30m-in-new-funding-to-reach-the-next-10m-researchers/ [66] Scite — https://scite.ai/ [67] SciSpace — https://scispace.com/pricing [68] Perplexity Deep Research — https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research [69] STORM (Stanford) — https://techstartups.com/2024/12/31/stanford-university-launches-storm-a-new-ai-tool-that-enables-anyone-to-create-wikipedia-style-reports-on-any-topic/ [70] Elicit — https://elicit.com/pricing [71] FutureHouse / Edison Scientific — https://www.futurehouse.org/about

Catalog: `data/landscape/solutions.csv` (83 rows × 12 columns). Method: a representative census of category leaders and structurally distinct players, grounded with web research (June 2026) and cited above. Maps to `docs/EDUCATIONPROBLEMS.md, docs/deep/01–04, docs/foundations/01–04, and docs/REFORMTHESIS.md`. Figures for private companies are estimates or company-reported reach where noted in the catalog's `scalenote` column._