Our Learning Engine
Our curriculum defines the learning objectives, and roughly orders them. Day to day, the learning engine sifts through thousands of available exercises and finds the right mix of review, recombination, and new material.
Our open-source learning engine is not rigidly programmed, but instead allows a number of different interests to inform its decisions about the next card. The competing interests here mostly take the form of several tried and true pedagogical principles: interleaving, spaced retrieval practice, progressive scaffolding, and the like.
These interests are often in tension, and occasionally in direct conflict with one another. Sometimes this means that one interest wins out over the others, but the more common case is that the engine finds exercises that dovetail to serve many purposes at once: win-win situations that respect all the techniques.
The Techniques
Curriculum
Opening new things in a useful order, and only when they're ready
- Aims for
- The most useful new thing your child is ready for.
- In tension with
- Everyone else on this list, who hold new things back until the old ones are solid. Spaced Repetition most of all: every review is a slot not spent on something new.
We lead with letters and phonetic rules that are common and "well-behaved" (consistent in their usage). We leave the zebras and their xylophones for a little later.
The curriculum is where everything new comes from: a new letter or team, or a new combination of ones already in rotation. Among what it has opened, it weights the most useful highest.
It also holds the door. An exercise that needs something your child hasn't been introduced to yet is all but ruled out, however good it looks to the others.
The full order is laid out on the curriculum page.
Spaced Repetition (SRS)
Mechanizing long-term retention and growth
- Aims for
- Yesterday's material back, just before it slips.
- In tension with
- Curriculum: every review is a slot not spent on something new.
The problem: Learned concepts and information can be forgotten if not revisited at the right intervals.
The solution: Our spaced repetition system automatically brings back exercises at optimal intervals, right before forgetting would occur.
This transforms short-term knowledge into permanent mastery, ensuring your child retains what they learn for the long haul.
Interference Mitigation
Separating confusable concepts to prevent mix-ups
- Aims for
- Look-alikes and sound-alikes kept apart until each is solid.
- In tension with
- Progressive Overload, which eventually wants them together.
Similar-looking letters and sounds can blur together without proper spacing. We strategically separate the introduction of confusable concepts.
Example 1: Learning b and d together is just awful! The symbols are mirror images.
Example 2: Learning d and t together is challenging for a different reason. The sounds are quite similar (different mainly in how much resistance your tongue gives before letting go), and not well distinguished by younger ears.
By minimizing interference, we help each concept establish itself clearly before introducing similar ones.
Progressive Overload
Gradually increasing challenge as skills develop
- Aims for
- Exercises a little harder than comfortable.
- In tension with
- Interference Mitigation early on, and Procedural Exercise Generation, which wants one new thing at a time.
Like strength training, reading skills grow through carefully calibrated challenge.
We continuously ramp up difficulty as your child's abilities develop, always keeping them in the "sweet spot" where learning happens: not too easy, not too hard.
As skills solidify, we relax the guardrails. Letters that were once kept apart (like b and d) can now practice together through minimal-pair exercises. Procedurally generated exercises can include more unknowns at once. The interference risks that required careful separation early on become opportunities for strengthening discrimination and flexible application.
This maintains engagement and ensures steady progress without frustration or boredom.
Procedural Exercise Generation
Building new concepts from solid foundations
- Aims for
- Exactly one new thing per exercise, built from parts your child already owns.
- In tension with
- Progressive Overload, which would happily stack up more.
When introducing new concepts, we use solidly known building blocks to create exercises.
Example: If your child knows m better than k, we'll prefer the word same over sake when introducing the long-vowel "magic e" pattern (CVCe).
This ensures each new lesson introduces exactly one new thing, rather than overwhelming with multiple unfamiliar elements.
Balancing Tensions: a worked example
Switch the techniques on one at a time to see their respective influence. With all of them on, the winner is rarely any one technique's first choice: it's the exercise that dovetails.
Next up: Who said that?, “fin”.
- Who said that?fin
due for review
- Pick the wordmap
No strong feelings either way
- Fill the gapbib
No strong feelings either way
- Spell itsat
too easy: well below where your child is
- Spell itstamp
two blends in five letters: far above where your child is
- Spell itdad
dlooks likeb, andbis still new - Spell itship
Rejected: needs
sh, which the curriculum hasn't introduced yet
The numbers above are made up for illustration. To watch the real engine choose, with every technique's score on screen, try our technical demo.
We use our ongoing experience with learners to continually test our assumptions and get the right mix of influences.
How the engine knows
None of the above works unless the engine has an accurate picture of what your child knows. Two choices keep that picture honest.
Rejecting Self-Report
Active demonstration replaces mindless dismissal
Throughout the course, understanding is actively demonstrated by learners. No "I got it!" buttons or easy outs.
Children can't mindlessly click through content. Exercises require authentic engagement and correct responses.
This ensures our system knows what they actually know, not just what they claim to know, keeping our personalization accurate.
We test spelling, and infer reading
The app never asks a child to read aloud to it
Almost every exercise runs from sound to spelling: your child hears a word and builds it from letters. A tap on a big letter key is unambiguous; a toddler's spoken guess is not.
Spelling is also the harder direction. A child who can reliably build ship can read it, too.
Reading aloud, with its gentle corrections, is your part. You're a far better judge of it than any microphone, and it happens naturally in shared books.
[One line: the order these techniques produce, on a timeline: when a typical learner meets each lesson and each fact.]
Our curriculumThe quickest way to see all of this working is to try it.
View a sample lesson