Deliberate Practice: Choose Targets You Can Observe

"Study networking" is not a goal; it is a mood. You cannot tell when you have finished it, and you cannot tell whether you got better. This page gives a way to turn a vague intention into a target you can observe, and a loop for working toward it.

Why vague goals fail

A vague goal gives you no way to detect failure, so the work drifts toward whatever feels comfortable: watching, rereading, collecting resources. An observable target names a behaviour that someone could check.

The pattern: a verb you can watch (explain, return, ship, revert, cite, choose) plus a condition under which you would say "not yet".

The loop

1. Attempt the target without help.
2. Inspect the result against a source, a trace, a test or a reviewer.
3. Locate the first consequential error, the earliest mistake that made later ones inevitable.
4. Repair only that. Fixing five things at once hides which fix mattered.
5. Vary one condition (change an input, a constraint, a tool) and try again.
6. Return later to see whether the skill survived.

This is a plain restatement of the deliberate-practice idea in Ericsson, Krampe and Tesch-Romer (1993): well-defined tasks, feedback, and repetition with correction. Established as a description of how experts in structured domains train. Provisional as to how much it explains performance: a meta-analysis by Macnamara and colleagues (2014) found that deliberate practice accounted for a modest share of performance differences in several fields, and less in professions than in games or music. Practise this way because it works better than rereading, not because it guarantees mastery.

Retrieval and spacing

Two findings are among the best supported in learning research.

• Retrieval beats rereading. Roediger and Karpicke (2006) found that students who tested themselves remembered more after a week than students who spent the same time restudying. Dunlosky and colleagues (2013) rated practice testing and distributed practice as the two highest-utility techniques among those reviewed. Established
• Spacing helps, and the best gap depends on how long you need to remember. Cepeda and colleagues (2008) found the optimal gap grows with the retention interval. Revisiting a case "after about a week and about a month" is therefore a sensible rule of thumb, not a measured optimum. Provisional

In practice: before opening your notes, explain yesterday's hardest idea aloud or on paper. Then check. Keep no more than three open weaknesses at a time; a longer list is a backlog you will not clear, and it hides which gap matters.

The evidence ladder

Not all evidence of learning is equal. Climb one rung at a time.

Where the evidence comes from also matters. Three evidence modes: a paper or course, a lab, and real work. A lab proves only its own scope and versions: "this worked on these versions with this configuration" is a complete and honest sentence; "this works" is not. When an AI tool wrote part of the work, add one more distinction: code you reviewed (you read it and can explain it) versus code you merely accepted. Only reviewed work counts as yours.

A status record for each project

Keep one short record per active project, and nothing more:

• Outcome and boundary
• Environment and versions
• Status: not started, active, demonstrated in scope, or needs retest
• Acceptance checks
• Evidence mode and location
• Observed result and limits
• Next exact action

The last line is the one that makes tomorrow's start easy.

Session shapes and keeping the habit

A study session works better with a shape. One example for an hour of study: five minutes of retrieval, thirty on the new material, twenty applying a changed case, five recording where you stopped. A review session can invert this: fifteen minutes on an independent attempt, twenty-five on repair, ten on a changed retest, ten on reflection.

Have a 20-minute fallback for bad days: locate your place, study one piece, explain one changed example, note where you stopped. On a short week, shrink the scope; do not stack missed sessions onto later days. Hold to one active course per track. A library of courses is a shelf, not a queue. Adding a second active course means finishing or dropping the first.

Friday review

Ask four questions, and write short answers.

1. What can I now do that I could not do last week?
2. What did I predict wrongly?
3. What is the one gap blocking progress?
4. What is the first action next time?

For how the same loop applies to reading and argument, see The Weekly Loop: Read, Reconstruct, Make, Defend, Log. For mathematics, see Learning Mathematics: Procedures and the Six Abilities and Placement, Repair and Retention: Keeping Mathematics Alive. Physical training uses the same idea with a body as the instrument: Strength Training for Busy Lives: The Recovery-Based Cycle progresses a target only when the previous one was met. A musician's practice method is the oldest example in the library: slow the passage, find the first wrong note, fix that, change the tempo, return tomorrow. Troubleshooting, covered in Evidence-Based Troubleshooting: Separating Explanations, is this loop run on a failing system. The broader ideal behind all of it is in The Philosopher-Engineer: Study and Workshop Together, and the wing orientation is in Engineering and Career: Start Here.

Try this

1. Take one vague goal from your own list and rewrite it as an observable target with a "not yet" condition.
2. Run the loop once on a small skill. Write down the first consequential error you found, not the last.
3. Schedule a retrieval attempt for one week from now, and one for a month from now. Before each, write what you expect to remember.

Further reading

• Ericsson and Pool, Peak: Secrets from the New Science of Expertise (2016), for the case for deliberate practice; read Macnamara et al. (2014) alongside it.
• Brown, Roediger and McDaniel, Make It Stick: The Science of Successful Learning (Harvard University Press, 2014).
• Dunlosky et al. (2013), cited above, a free and readable review of study techniques.