Built on how memory actually works.
Eight findings from cognitive psychology, replicated for decades. Every exercise type and every scheduling rule in Axo traces back to one of them. This is the answer to “why not just ask ChatGPT.”
Last updated November 9, 2026
A model, not a measurement. Each review resets the curve and flattens the next one.
- Read once
- With Keep Sharp
2 of 4 reviews, on days 1, 3
Table view
| Day | Read once | With Keep Sharp |
|---|---|---|
| 0 | 100% | 100% |
| 1 | 66% | 100% |
| 3 | 29% | 100% |
| 7 | 5% | 73% |
| 14 | 0% | 42% |
| 21 | 0% | 24% |
| 30 | 0% | 12% |
Ebbinghaus drew the first version of this curve in 1885. The shape has held up. The intervention that flattens it, review spaced just before you would forget, is the single best-supported finding in the science of learning, and it is the engine under Keep Sharp.
Eight principles. Eight places you’ll feel them.
- 01
Retrieval practice.
Pulling information out of memory strengthens it far more than rereading it. Roediger and Karpicke's 2006 studies, and hundreds since.
In Axo
Every Bite ends in questions, not a summary. Keep Sharp is all retrieval. There is no “review the text” screen without a question attached.
- 02
Spaced repetition.
Memory decays on a curve. Reviewing just before you forget resets the curve with a flatter slope.
In Axo
Each concept gets a memory score and a next-review date. Keep Sharp pulls what is due today, across all your Paths.
- 03
Desirable difficulty.
Making recall slightly harder improves long-term retention, even though it feels worse in the moment. Bjork's work.
In Axo
A question that comes back is asked another way than you first met it, and on review you type the answer from memory. Review sessions mix topics.
- 04
Interleaving.
Mixing related problem types beats blocking them, because you learn to tell them apart.
In Axo
Keep Sharp interleaves concepts from different units and Paths. Capstones mix every unit.
- 05
Chunking and cognitive load.
Working memory holds a handful of items. Well-sized chunks and no clutter let more get through. Miller, then Sweller.
In Axo
Bites are 5 to 7 minutes and carry five concepts, one per question. One idea per screen. No decoration on learning screens.
- 06
Elaboration and generation.
Explaining why, connecting new facts to known ones, and producing an answer yourself all deepen encoding.
In Axo
Scenario questions ask what you would say to a friend. Typed answers on review make you produce the word, not pick it. “Help me learn this” explains a question another way and drills it.
- 07
Dual coding.
Words plus matching audio or an image form two memory traces instead of one. Paivio.
In Axo
Every Bite is read aloud in Axo’s voice, words lit as they are said. Every Bite has a scene: Axo talks an idea through over a picture drawn in code, timed to his words.
- 08
Goal gradient and completion.
People speed up as they near a visible goal, and a defined end point drives follow-through far more than open-ended content.
In Axo
Paths are fixed length with a visible rail. A capstone and a badge at the end. Go Deeper creates a new Path instead of stretching the old one.
Gamification that helps, not the kind that harms.
Points, streaks and badges reliably bring people back. The failure mode is rewarding time spent instead of knowledge gained. So:
- XP is earned only for correct retrieval. Rereading earns nothing.
- A streak is kept by one Bite a day. Nothing about it is for sale, and the alert the night it would end names the Bite that saves it.
- No hearts, no energy, no gems. You never run out of the right to learn.
- The only board is your friends’, by XP today and this week. It has a top and no bottom: a friend with nothing yet has no place, not last place.
What we don’t do.
- Learning-styles quizzes. There is no evidence for them.
- Long passive video.
- Highlighting as a study method.
- Cramming modes. Keep Sharp brings an idea back just before you would forget it, not the night before a test.
One weekend of reading.
- Make It Stick, Brown, Roediger and McDaniel.
- Ultralearning, Scott Young.
- Duolingo's engineering posts on spaced repetition and difficulty modeling, Duolingo.
Feel it instead of reading about it.
The Bite about axolotls uses four of the eight. It takes three minutes.