Beyond Reason · Interactive companion to “The Order of Learning”
The Order of Learning
A map of 1,591 things children learn from 4 to 13 says the order is right almost everywhere. What the best systems change is the time between the steps.
The topics with the most resting on them all open at 4 and 5
Every dot is a topic a child learns between 4 and 13, placed by the age it opens and its subject. Size is how many later topics depend on it. Tap a dot to light up everything it rests on (dark) and everything that rests on it (light).
Scroll sideways to see every age.
What is learned first is what everything else rests on
Two bars for each age. Gray: how many of the 1,591 topics first appear at that age. Blue: how much of the whole map depends on those topics, counting every later topic that cannot be learned without them, directly or through a chain.
Share of topics that first appear at this ageShare of the whole map that depends on them
Data table
From counting blocks at 4 to scatter graphs at 13: one chain of 25 math topics
The longest run of hard prerequisites in the map. Read top to bottom: each topic needs the one above it. Columns are ages. Tap a topic to see what a gap there would cost.
Math is a tall stack with little slack; the other subjects are shorter and looser
Each circle is a subject, sized by its number of topics. Across: how many years usually pass between a required prerequisite and the topic that needs it (further left, the tighter the sequence). Up: how tall the stack under a typical topic is, in steps (its longest chain of earlier topics; higher, the more a subject depends on getting the order right).
Data table
In math, most prerequisites are needed again within a year
Take every pair of topics where the first must be learned before the second (a hard link). Count the years between them. “Same year” means both are due in the same school year. Mathematics on the left, every other subject on the right.
Data table
Common Core and England put the same math in the same year
160 topics carry both a US grade and a single English school year. Each row is a topic area. The dot shows how much later, on average, the US teaches it (right) or England does (left).
Data table
Five systems, one order, and calendars up to four years apart
The age at which each national curriculum first requires a landmark topic. Where the markers spread, systems disagree about when an idea is ready; where they sit together, the sequence is close to universal.
Data table
More years of formal school before 15 do not show up in the PISA score
PISA is taken at 15. Each bar runs from the age formal reading and arithmetic instruction begins to the test, so its length is how many years of formal schooling a 15-year-old has had. On the right, the 2022 mathematics score against the OECD average.
The calendar starts early where the brain has no deadline, and runs fastest while working memory is still growing
Top: how the capacities that schoolwork leans on grow between 4 and 13, and which windows close. Bottom: how much the map asks of each year. The three shaded stretches are where the calendar and the curves look out of step.
Ready before asked, asked before needed
For each idea, the hollow mark is the earliest age a study finds children handling it in a light form, and the filled mark is the age the map first asks for it. An arrow means the evidence sets no deadline at all. Tap a row for the study and the map topics; the button takes you to the matching change in the reshuffle.
Applying the evidence moves 105 topics and adds no time between the steps
Every square is one of the 1,591 topics on a fixed board: rows are clusters, columns are the year a topic opens. Check a change and the squares slide along their row to the new year; nothing else shifts. The four changes are the only calendar changes the evidence supports. Together they move 7% of the map and open no gaps between steps, which is the point: the calendar is not where the outcomes are. Tap or hover a square for the topic and why it moved; the table below the chart lists every move and sorts.
Scroll sideways to see every age.
What moved
Rules, counts, and evidence
A prerequisite fades unless it comes back before it is needed
Moving topics between years cannot add time between steps. A school can add return visits: the prerequisite brought back, briefly, before the topic that needs it opens. Call each one a retrieval appointment. Set the gap and how often the prerequisite comes back, and watch what is left when the next topic opens.
What to do with this
The map is open, versioned, and wrong in places. So is this analysis. Here is what moves outcomes, ranked by the evidence in this piece; what to do with it tomorrow morning; the trial I would fund first; and what would change my mind.
The levers, ranked
- Bring prerequisites back before they are needed. Retrieval practice g = 0.50 across 222 classroom studies; spacing d = 0.54. The strongest effects in the file, and the map gives the schedule: 2,030 hard links (Figures 5 and 12).
- Check mastery before moving on, in mathematics only. The 114 topics on the longest chain, in the fraction cluster, and at the chokepoints. Singapore writes this premise into its syllabus; the only AI gain that survived a week sat inside a mastery check (Figure 11, dark blue).
- Give fractions the years 8–10 and teach them concretely. Fraction knowledge at 10 uniquely predicts algebra at 16; 71% of the math a child meets from 10 rests on it (Figures 1 and 3).
- Interleave the practice. Mixed problems beat blocked ones 61% to 38% on a delayed test; 9.7% of textbook problems are interleaved. Not in the map; in the file.
- Reclaim the formal years before 7. No evidence of a lasting gain from starting reading or arithmetic drills at 4 and 5, and guided play does at least as well: a cost lever, not an outcome lever (Figures 9 and 10).
- Reorder the calendar. The four changes the evidence supports move 7% of topics and open no gaps. Approximately zero (Figure 11).
Outside this corpus, and therefore not sized here: teacher mathematical knowledge, textbook quality, tracking, and the rich-poor gap, which in the United States is the widest of the Western five.
If you are a parent
- At the next parent-teacher conference, ask “which prerequisite is missing?” instead of “what’s the grade?” If your child is between 8 and 10 and the answer is fractions, that is the whole conversation.
- Ask about last month’s math, not just this week’s. Bringing an old step back, briefly and more than once, has the most evidence behind it of anything on this page: 222 studies, 48,478 students.
- Don’t buy a head start. Children taught to read at 5 are level with those taught at 7 by age 11, and guided play teaches early math at least as well as drill. Skip the flashcards.
If you teach
- Open each new topic with two minutes of recall on what it rests on. The map names those topics for every lesson; Figure 1 will show you them.
- Check mastery before moving on at the 114 math topics marked in Figure 11. In science, history, and the rest, let interest set the order; their graphs allow it.
- Watch for math anxiety in first and second grade. It is measurable at 6 and 7, it feeds on low scores, and the fraction years start at 8.
If you run a system
- Stop arguing about the order. It is the same in Tallinn, Helsinki, Singapore, and San Antonio. Fund what goes between the steps: retrieval, mastery checks, and play before 7.
- Fund one trial first: scheduled retrieval on the hard links in grades 3–5 math, against business as usual, with fraction knowledge at 10 as the outcome. Siegler supplies the prediction; the map supplies the schedule.
- Put AI inside the mastery check, as feedback at the moment of error. That is where a 6,997-student Tennessee trial found the only gain still there a week later (grades 6–8, suggestive). Below 11, the one study we found, Rori in Ghana (0.37 SD), covered grades 3–9 and was not split by age.
- Publish your standards as a graph, with reasons and a changelog. Marble did it in a summer. You’ve had since 2010.
What I don’t know
- Whether the 666 same-year links are a fact about learning or about the map’s one-year grain. Singapore’s syllabus orders topics within the year; the taxonomy does not.
- What happens after 13. The map is K–8, and the longest chain’s last steps are the ones with the least data.
- What a mastery check costs in time, and who pays it: the child who waits, or the class that waits for the child.
What would change my mind
- A randomized trial in which an earlier formal start holds its lead past age 11. The comparisons we have are not randomized, and none shows one [36][37].
- A spaced-retrieval schedule laid on a mathematics prerequisite chain that fails to improve the dependent topic. The classroom evidence for spacing and retrieval is strong; the link-by-link version is my inference [59][60].
- A K–5 trial of a student-facing AI tutor with a positive effect that is still there a term later, split by grade. We found none.
- A corrected map that moves the fraction cluster or shortens the longest chain. Every number here comes from Marble’s age tags, and 50 links still have a prerequisite tagged later than the topic that needs it.
The order is a graph. The calendar is a guess. The gaps are the game.
Read the essay · Flip the deck · All 126 sources · Get the map on GitHub · Send a correction