Learning Mathematics: Procedures and the Six Abilities

Watching someone solve a problem feels like learning, and often is not. This page gives a five-step method for learning a mathematical procedure so that it lasts, and then six separable abilities that tell you whether you actually understand it. For how to keep what you learn, see Placement, Repair and Retention: Keeping Mathematics Alive.

Why watching is not learning

A smooth explanation produces fluency in the listener, a feeling of "I could do that". Research on learning distinguishes performance during study from learning that persists and transfers. Conditions that make study feel easy can reduce long-term retention, while some effortful conditions improve it (Soderstrom and Bjork, 2015). Established The practical rule: if you only watched or read, you do not yet know whether you can do it. Effort is not the same as competence either; struggling for an hour on a method you were never shown is not a virtue. The method below balances guidance and independence.

Five steps for learning a procedure

The sequence is example, guided completion, independent solution, varied practice, then delayed retrieval. It draws on Rosenshine's principles of instruction (small steps, models, guided practice, regular review) and on the worked-example literature, where studying examples then gradually removing support outperforms unguided problem solving for novices (Sweller and Cooper, 1985; Renkl and colleagues, 2002). Established for novices in well-structured domains.

1. Read one worked example and name the reason for each step. Cover the next line and predict it. For every step, state the operation and why it is allowed. If you cannot explain a step, repair that prerequisite first.
2. Complete a partially worked problem. Fill in the missing steps yourself, without copying. As you improve, ask for less support.
3. Solve a fresh, comparable problem unaided. Keep the attempt. If it fails, find the first incorrect or unjustified step and take only the smallest hint that lets you continue.
4. Change one feature. Alter a sign, a representation, an assumption or the wording, so the method must be understood and not merely copied.
5. Mix with older methods and retrieve later. Once the method works, mix it with earlier ones so you must choose a method, and schedule fresh problems after a delay.

For material you already know, skip the examples and start with an independent diagnostic problem. The benefit of examples fades as expertise grows; this is the expertise reversal effect (Kalyuga and colleagues, 2003). For genuinely new material, do not spend the whole session guessing at a method you were never taught. If several minutes produce no useful step, look at an example or a targeted hint, then return to independent work.

A thirty-minute session

When the skill is familiar, move the example time into independent practice. On assessment days, use the whole block for fresh work and correction. These timings are a practical template. Provisional The research supports the sequence of ideas, not a particular number of minutes.

A four-week cycle

A cycle organises practice; it does not promise that a unit will close in four weeks. If prerequisites need longer, carry them forward.

The six abilities

Understanding is not one thing. This wing tracks six abilities separately, which echo the strands of mathematical proficiency described in Adding It Up (Kilpatrick and colleagues, 2001). The six-way split is a teaching convenience, not a measured psychological structure.

Keep separate records. A fast calculator may be unable to prove; a fluent explainer may be unable to calculate independently. A single "mastery score" hides exactly these differences. Work toward accuracy and understanding before speed. Practical targets for each ability, and how to make them observable, are in Deliberate Practice: Choose Targets You Can Observe; the habit of turning study into evidence appears in The Weekly Loop: Read, Reconstruct, Make, Defend, Log. Algebra, as the usual place these abilities first show up, is treated in Algebra: The Skills That Carry Everything.

Using a tutor or AI as critic

A tutor, teacher or AI model is most useful when it critiques your attempt rather than producing the answer. Good requests: "Identify the first unjustified step and the rule it breaks. Give one hint and wait." "Give me a partly worked problem and then a fresh one without the answer." "State the strongest objection to my argument."

Do the final assessment alone, without notes or AI, and use qualified human feedback at major milestones. A model's agreement is not evidence that you are right. One randomized study in high school mathematics (Bastani and colleagues, 2025) found that students with unrestricted access to a general chatbot did better on practice problems but worse on a later exam without it, while a version designed to give hints rather than answers largely avoided that harm. Provisional One study in one setting, but it fits the broader performance-versus-learning finding. More on this tradeoff is in Technology, AI and Human Judgment.

Connections and caveats

Plato's Divided Line, outlined in the Ladder, separates reasoning that starts from assumed hypotheses from the highest understanding that grounds them (Republic VI, 509d-511e). It is a fair picture of the aim of justification: not just following a method, but knowing why it works. That is an interpretive analogy, not an empirical claim about learning.

Rosenshine's ten principles were distilled from classroom research on effective teachers, cognitive science and studies of cognitive supports; they are a synthesis, not a single experiment. Exact numbers such as 15 minutes of practice or a four-week cycle are conveniences. Adjust them to the material and keep the principles.

Try this

1. Choose one procedure you are learning (for example solving a linear equation). Write one worked example with a reason beside each step, then make a version with two steps blank and fill them in from memory the next day.
2. Take a solved problem and change one feature (a sign, a variable, a domain). Solve the new version and write what the change did to the method.
3. For one week, keep a six-column record, one column for each ability, with a fresh correct-over-attempted count in each. Note which column is weakest.

Further reading

• Barak Rosenshine, "Principles of Instruction", American Educator, 2012 (free online).
• Kilpatrick, Swafford and Findell, Adding It Up (National Academies Press, 2001).
• George Polya, How to Solve It (Princeton University Press, 1945).
• Soderstrom and Bjork, "Learning versus performance", Perspectives on Psychological Science, 2015.
• Bastani et al., "Generative AI without guardrails can harm learning", PNAS, 2025.