Curriculum Research
Research · 15 November 2026 · K–12 + Toddler + Adult

Build your own curriculum without dependency hell.

29 case studies on how real families, publishers, and researchers sequence learning so no lesson ever quizzes a concept that was never taught — and exactly what to steal for a family curriculum that runs from age 1 to adult.

29case studies & systems, across 7 categories
2reasoning-branch sources: Theseus's Ship + Vsauce
15ranked stealable techniques for a small app
1formal solution to dependency hell — see below
The headline finding

Dependency hell has a solved, formal answer.

Knowledge-space theory (Case 18 — Doignon & Falmagne, 1985; built into ALEKS) treats a curriculum as a prerequisite graph. Every learner has a knowledge state — the set of things they can actually do. The outer fringe is the set of items whose prerequisites are all already mastered. The system only ever offers outer-fringe items.

Everything else in this research is either a cheaper approximation of that idea — placement tests, mastery gates, mixed review, narration, backward design — or a warning about where sequencing breaks. If the Compiler stores prerequisites as data and enforces the outer-fringe rule, dependency hell becomes structurally impossible, not something you audit for after the fact.

Three laws that show up everywhere

Independent systems, decades apart, keep arriving at the same three rules.

01

A correct answer is not mastery

Khan, Kumon, Bayesian Knowledge Tracing, and Orton-Gillingham all independently require re-demonstration — later, mixed with other skills, under time pressure, or in production (spelling it, narrating it). Never trust a single correct quiz response. (Cases 1, 14, 16, 19, 24)

02

Repeated failure is a prerequisite alarm

In ASSISTments data, students in the bottom 20% of prerequisite knowledge “wheel-spun” — 10+ failed attempts, never mastering — 50% of the time, vs 10% for the top 20%. The fix is to route down to prerequisites, not to motivate harder. (Case 19)

03

Under age 3, there is no testable curriculum

Head Start and NAEYC are explicit: rote teaching of discrete skills is developmentally inappropriate for infants and toddlers. The “curriculum” is a progression map plus an observation log. (Case 28)

The 29 case studies

Search by name or idea, filter by category, open any case for its full sequencing logic and readiness rule. Every case ends with the one technique worth stealing.

The steal list — ranked for a small family app

From the research synthesis, in priority order. The top four alone would fix most of the Compiler's dependency problem.

Apply it: the Compiler blueprint

What this means concretely for the family curriculum — Sofia (8), Emilee (toddler), the baby, and the adult strand for Maxx & Amber. Each move names the case it was stolen from.

Move 1 · Case 18

Make prerequisites data, enforce the outer fringe

Every lesson stores its prerequisites. A lesson — and especially its quiz — stays locked until every prerequisite is mastered. The 8-layer-stack failure mode (quiz accessible before the stack was ever discussed) can't happen by construction.

Move 2 · Case 29, Check 8

Audit what already exists, cheapest check first

For each existing module, find the first meaningful decision the learner must make, and verify every piece of knowledge that decision needs has already appeared. Modules that fail are exactly the dependency-hell modules. Then: backward design (22), curriculum mapping — “who teaches X?” (21), and the wheel-spinning signal (19).

Move 3 · Cases 14 + 16 + 19

Replace one-and-done quiz gates with real mastery

Two tiers — Proficient (100% in isolation) vs Mastered (right again later, mixed with other skills). Add fluency where it fits (accuracy + time), and treat 3 correct in a row as the cheap floor. 10 failed attempts routes down automatically.

Move 4 · Case 9

Mixed review in every session

Saxon's rule: every practice set samples a few items from all previously mastered skills — not a separate review week. Forgetting is a prerequisite failure too; this catches it within days.

Move 5 · Case 20

Decompose foundations with the counting ladder

Before single-digit addition: subitizing → comparison → rote counting → 1-to-1 correspondence → cardinality → hierarchical inclusion → conservation. Each step is observable in under a minute. Sofia's Daily Counter covers part of this; Emilee needs the toddler version below.

Subitizing→Comparison→Rote counting→1-to-1 correspondence→Cardinality→Hierarchical inclusion→Conservation→Addition
Move 6 · Cases 28 + 4 + 20

Toddler branch = observation, not lessons

For Emilee's age band: a parent-facing checklist by domain (relationships, motor, communication, cognition, self-regulation) and overlapping age bands, with play activities tagged to the skills they build. No quizzes, no child-facing tests, minimal screen assessment. Coverage gets audited from the log, not from testing the child.

Move 7 · Case 1

Narration as the un-gameable check

For reading and content modules: “tell it back in your own words” — recorded audio for young kids. You cannot guess a retelling, and it doubles as comprehension + composition practice. Use it wherever a multiple-choice quiz currently stands in for understanding.

Move 8 · Cases 26 + 27 + 10

One family, not five curricula

Group/skill split: content subjects together at different depths (read-alouds, history, science); skill subjects on per-child tracks (math, reading, writing). Name levels by topic, not grade, so pace carries no stigma — and keep a separate gap-repair track so remediation never blocks the main sequence. Older kids teaching younger ones counts as their mastery demonstration.

Move 9 · New strand

The adult academic-discipline branch gets foundations too

The strand for Maxx & Amber follows the same law as everything else: no module without its prerequisites underneath it. It starts with the reasoning branch below — deliberately chosen because Theseus's Ship needs no jargon and Vsauce builds its own prerequisites (confirmation bias) inside the video. Argument skills first, then the paradoxes.

Move 10 · Case 25

Validate every arrow

The cautionary case: phonemic awareness with letters beat phonemic-awareness-first (reading effect d=0.67 vs 0.38), even though “sounds before letters” feels logically necessary. When real usage data contradicts the prerequisite graph, fix the graph — and cap prerequisite drills (PA benefit plateaued at ~10.2 hours).

Reasoning branch — the two named sources

Requested for the curriculum's reasoning strand. Both are first-lesson material precisely because their prerequisites are light.

R1

Theseus's Ship — the paradox of identity

Plutarch, Life of Theseus 23.1 (late 1st c. CE) · Hobbes's second-ship twist, De Corpore (1655)

Identity over timeFirst paradoxOlder kids → adults

If every plank of a ship is replaced one by one until no original material remains, is it still the same ship? Hobbes sharpens it: a custodian rebuilds a second ship from the discarded planks — now the continuously-repaired ship (same form and history, no original matter) and the reassembled ship (all original matter, broken history) both claim the identity. Identity is an equivalence relation, so both cannot be identical to the original — the learner must pick a criterion (material, formal/functional, spatiotemporal, narrative) and defend it.

Teaching sequence: (1) Plutarch's story + “when does it stop being the ship?” vote → (2) Hobbes's second ship — force the form-vs-matter choice → (3) the body parallel — apply it to yourself → (4) name the candidate criteria → (5) the payoff: no agreed answer after ~2,000 years, and that's the point.

Best classroom version: PLATO, “Identity Activity: What makes you, you?” — full lesson plan with warm-ups, discussion questions, and the “Mission to Mars” game. plato-philosophy.org · Scholarly: Stanford Encyclopedia of Philosophy, “Identity Over Time” & “Material Constitution” — plato.stanford.edu
Caveat from the research: the PLATO activity's “every cell replaced in seven years” line is a popular simplification — not literally true for all cell types (many neurons persist). Fine as a discussion prompt; don't teach it as fact.

R2

Vsauce — “The Future of Reasoning”

Michael Stevens, Vsauce · 2021 (Season 2021, Episode 2) · 30:02

What reasoning is forAge 12+ & adultsVideo

The arc: the extended mind (your mind lives partly in tools, notes, and other people) → hyperobjects (Timothy Morton — things too distributed in time and space to grasp whole, like climate change) → The Enigma of Reason (Mercier & Sperber): if reasoning evolved to find truth, why are we such biased, lazy reasoners? Their answer — reasoning evolved for social functions: justifying, convincing, coordinating. “Instead of using reasoning to come to conclusions, we use conclusions to come to reasons.” Confirmation bias is a feature when the group adjudicates. The future it proposes is a return to those roots: national deliberation days and lottocracy (decisions by randomly chosen citizens, like a jury — with ancient Athens as precedent). Closing frame: “The future of reason may be its past.”

Discussion prompts: Is reasoning for truth or for winning arguments? When has a group you were in reasoned better than any member alone? Would you trust a randomly chosen group of citizens more than elected officials — and why? Pairs with R1: Theseus asks what makes a thing the same over time; Vsauce asks what reasoning itself is for — and both end by returning the question to the learner.

Watch: youtube.com/watch?v=_ArVh3Cj9rw · Transcript/summary: glasp.co/youtube/_ArVh3Cj9rw (fetched 2026-11-15)
Open question: the exact 2021 publish day was not verified — only the year and the S2021 E02 episode listing. Runtime 30:02 per episode listings.

Five questions, answered

The cross-cutting questions the research was asked to settle.

How far down do you decompose before a topic is teachable?

Until each step is (a) observable in a single behavior, (b) teachable in one short activity, and (c) checkable in under a minute — the counting ladder in Move 5 is the worked example. Singapore Math adds a second axis: decompose by representation (build it → draw it → symbolize it). Knowledge-space theory gives the stopping rule: decompose until the outer fringe is never empty — there's always a next learnable step. Over-decomposition is a real cost too (Case 25's ~10-hour plateau).

Mastery-based or age-based advancement?

Mastery-based — but the evidence is more modest than the headline. Bloom's +1σ for mastery learning shrinks to ~0.35–0.66 SD in later meta-analyses of tutoring systems, ~0.59 for Bloom's own program, and g≈0.92 (adjusted) in a 2026 STEM meta-analysis; the EEF rates it “moderate impact for very low cost.” All 50 US states now have competency-based policies built on “evidence of mastery, not seat time.” Under ~age 5, “mastery” means observed behavior in play and routines, not test gates — and age is only a loose expectation window, never a promotion criterion. Effect sizes vary widely by outcome measure (teacher-made tests ~0.5 vs standardized tests ~0.08 in Kulik et al. 1990), so treat any single number with suspicion.

What's the practical audit method?

Four methods, cheapest first: (1) the first-meaningful-decision check (Case 29, Check 8); (2) backward design — the transfer task's demands become the prerequisite list (Case 22); (3) curriculum mapping — “who teaches X?” for every skill (Case 21); (4) the empirical audit — if ~10–20%+ of learners stall at 10+ failed attempts, suspect a missing prerequisite (Case 19). Discipline: only load-bearing prerequisites, a diagnostic trace for each uncertain one, never convert the audit to a comforting score — a hard gap remains a hard gap — and re-run dependent checks after any change.

How do the best systems handle a learner who aces a quiz on a topic never taught?

They treat one correct answer as weak evidence, five ways: model guessing explicitly (BKT's Guess/Slip parameters, ~0.95 knowing threshold); require delayed, mixed re-demonstration (Khan's Mastered tier); require fluency, not just accuracy (Kumon's time + mistake limits); require production, not recognition (spell it — OG; narrate it — Mason); and check prerequisite consistency (KST/ALEKS — an advanced item passed while its prerequisites fail is an infeasible state: evidence of guessing, so probe downward). Genuine prior learning is fine — the assessment grants credit wherever it was learned. And note the causal direction: gaming is usually a symptom of missing prerequisites, so a spike in guessing is itself an audit signal.

What does a real curriculum for ages 1–3 look like?

Not lessons. An organized developmental scope and sequence across five domains — Approaches to Learning; Social & Emotional; Language & Literacy; Cognition; Perceptual/Motor/Physical — in overlapping bands (birth–9, 8–18, 16–36 months), delivered through play and trusted adults. First skills: attachment and object permanence; walking, climbing, scribbling, stacking; single words → two-word combinations → pretend play; imitation, cause-and-effect, sorting, subitizing to 3. Observed, never tested: anecdotal records, checklists, portfolios, documentation inside routines. (Cases 4, 20, 28; NAEYC DAP.)

Open questions & caveats — stated plainly

  • No named audit case report was found — beyond the curriculum-mapping examples (“no one is teaching students how to write a research paper in high school”; a dinosaur unit duplicated in kindergarten and grade 2), no documented case of a specific family or school auditing a curriculum and naming its missing prerequisites surfaced. The 16-point audit (Case 29) is a published method, not a case report.
  • Vendor-adjacent statistics — do not quote unchecked: the claims that St. Mary's Episcopal School (Atlanta) cut redundancy 40% after mapping, and a Learning Policy Institute 2023 figure that 68% of elementary schools lack vertical alignment in grades 3–5 math, come secondhand from a marketing-context source.
  • Orton-Gillingham evidence caveat: a 2021 meta-analysis (Stevens et al.) found OG's effects positive but not statistically significant on foundational skills (ES 0.32, p=.24). The stealable part is its sequence discipline, not a guaranteed outcome.
  • Mastery-learning effect sizes vary widely by outcome measure — see the Q&A above. The report states the range rather than a single number.
  • The If Machine (Worley) Theseus chapter was cited secondhand; verify against the book before quoting its wording.
  • Verification level: all facts are web-index verified on 2026-11-15 (search results and/or full page-text fetch). Six primary pages were fetched in full: AmblesideOnline, Khan Academy Help Center, Aurora Institute, Head Start ELOF, the TenMultigure audit, NAEYC DAP — plus PLATO and the Glasp transcript for the reasoning branch. Nothing was verified in a live browser; no prices or account state were at issue.