How Knowledge of Results Works | Tell the Learner What Happened Without Doing the Correction for Them

eduKateSG Learning Node Series · 0268

A learner makes an attempt. The teacher can respond in two fundamentally different ways. One is to say what happened: the answer was 7, the throw landed 18 centimetres left, the paragraph scored 3 out of 5 for evidence, the timing was 0.4 seconds slow. The other is to say how the performance was produced: the sign changed incorrectly, the release was late, the evidence was not linked to the claim, the sequence broke before the final step.

Motor-learning research gives the first category a specific name: knowledge of results, usually abbreviated KR. It is augmented information about the outcome of an attempt relative to a goal. Knowledge of performance, by contrast, describes aspects of the movement or process that produced that outcome.

The distinction sounds technical, but it is one of the cleanest ways to improve feedback design. Teachers often give process explanations when the learner first needs a result signal, or give scores when the learner has no idea what produced them. The useful question is: what information is missing from the learner’s own evidence right now?

Quick answer

Knowledge of results works by telling the learner how the attempt ended relative to a target without necessarily prescribing the correction. It can calibrate internal judgement, reveal the direction and size of error, and support adjustment across repeated attempts. But more KR is not automatically better. Frequency, delay, task difficulty, learner age, intrinsic feedback and dependence all change its usefulness.

The core loop is: attempt → estimate the result → receive KR → compare estimate with outcome → choose an adjustment → attempt again → gradually remove KR → test whether performance survives without it.

The owned reader job

This Learning Node owns the question: when is outcome information enough to drive learning, how much should be provided, and when does constant result feedback stop helping the learner become independent?

It sits beside How Feedback Works | Feedback Timing, which owns when feedback arrives; Feedback Bandwidth, which owns how much information can be used; and How Self-Controlled Feedback Works, which owns learner control over access. This article owns the outcome signal itself.

Knowledge of results is not the same as praise

“Good job” is social evaluation. “You were 4 centimetres short” is knowledge of results. “Your final answer is correct” is KR. “Your second paragraph contains two pieces of evidence but neither is explained” is closer to process information.

KR can be positive or negative, precise or coarse. A score, distance, time, success/failure signal or numerical error can all function as KR if they tell the learner about the achieved result.

Its power comes partly from objectivity. The learner can compare an internal prediction with an external outcome. That comparison helps answer two questions: Was I successful? and Can I tell when I am successful?

Intrinsic feedback comes first

Not every task needs externally supplied KR. Some tasks reveal their own result clearly. A basketball either enters the hoop or misses. A program passes the test or fails. A mathematical substitution either satisfies the equation or it does not. A musical note can be heard against a reference pitch.

In other tasks, the result is hidden, delayed or ambiguous. A learner may not know whether an essay argument is convincing, whether a clinical procedure met a quality threshold, or whether a pronunciation error is perceptible to a listener. External KR then adds information unavailable from the task alone.

Before adding a scoreboard, ask: what result can the learner already detect? Feedback that merely repeats obvious information can occupy attention without adding evidence.

KR can calibrate the learner’s internal estimate

Suppose a student estimates that an answer is probably 90% correct. The teacher reveals that it is wrong. That result is useful, but the deeper information is the mismatch between confidence and outcome. Repeated mismatches reveal poor calibration.

A strong feedback routine therefore asks for the learner’s estimate first. In physical tasks, the learner can predict whether the movement was long or short before seeing the measured error. In mathematics, the learner can state confidence before checking. In writing, the learner can score the paragraph against a rubric before seeing the teacher’s result.

KR then becomes a calibration instrument rather than a verdict.

Worked example: arithmetic fluency

A child completes a one-minute set of multiplication facts. The teacher reports: “32 correct, 3 incorrect.” That is KR. If the goal is fluent retrieval, the result immediately tells the learner whether performance is moving toward the criterion.

But the number alone does not explain the errors. If all three errors involve 7×8, 6×7 and 8×9, the learner may need targeted process or knowledge feedback. KR shows the size of the problem; diagnosis identifies the weak structure.

A useful sequence is therefore: result first, then only as much process feedback as the pattern justifies. Do not turn every score into a lecture.

Worked example: algebra

A student solves an equation and receives “x = 4 is incorrect.” That is low-resolution KR. If the learner can independently substitute 4 into the original equation and see that the two sides differ, the result signal may be enough to trigger self-correction.

If the learner cannot locate the error after a reasonable attempt, the teacher adds process information: “Your transformation from line two to line three changed the sign of the constant.”

The feedback ladder matters. If the teacher immediately supplies the corrected line every time, the learner never has to use the result signal to search their own reasoning.

Worked example: timed performance

A learner practises a piano passage with a target tempo. After each attempt, a metronome-linked system reports average timing drift. That is KR. The learner is not yet told which finger or transition caused the slowdown.

Across several attempts, the result reveals a pattern: drift begins only above a certain tempo. The teacher then introduces process feedback on the transition that fails under speed. Outcome data and process diagnosis work together.

The important distinction is that KR answers how close was the result? It does not automatically answer what caused the distance?

Why feedback after every attempt can become guidance

A central idea in the motor-learning literature is the guidance hypothesis. Frequent augmented feedback can improve performance during practice because it steers the learner toward the target. But if the learner begins to rely on that external signal, performance may deteriorate when feedback is removed.

Classic experiments found situations in which reduced or delayed KR produced worse-looking acquisition but better later retention. This is one reason learning scientists insist on separating performance during practice from learning demonstrated later.

The guidance hypothesis is not a universal law that “less feedback is always better.” Later studies show important exceptions, especially for complex tasks, children and learners with specific conditions. The principle is diagnostic: check whether the feedback source is replacing internal error detection.

More difficult tasks can need more feedback

A learner cannot learn from an error they cannot perceive. On a difficult unfamiliar task, intrinsic signals may be weak or misleading. Frequent KR can help establish the basic mapping between action and consequence.

A study of a complex ski-simulator task found benefits from frequent feedback relative to lower frequency in that training context. Other work with children has also shown that reduced feedback can harm aspects of parameter learning. These findings challenge the simplistic rule that feedback should always be faded as quickly as possible.

The better rule is: provide enough result information for the learner to build an accurate action–outcome model, then test whether that model can operate with less external support.

Children are not small adults

Feedback-frequency effects can differ by age. In a 2012 study of rapid arm movements, children who practised with reduced feedback showed larger temporal parameter errors in retention than children who received frequent feedback, while the pattern differed from adults.

Another 2012 study explicitly examined the interaction of task difficulty and KR frequency in children’s motor learning. Reviews in paediatric rehabilitation likewise describe heterogeneous findings and caution against universal schedules.

For teachers, the implication is straightforward: do not copy a feedback-frequency rule from adult laboratory research and apply it mechanically to young learners. Development changes perception, working memory, error detection and how much structure a learner can infer from sparse signals.

Some learners need KR because intrinsic signals are impaired

Clinical research provides an especially clear example. A study comparing people with Parkinson’s disease and controls found that controls retained better after lower-frequency KR, whereas the Parkinsonian group benefited more from 100% KR in the studied timing task.

The authors interpreted the difference in relation to reliance on external feedback and reduced proprioceptive acuity. Whether that specific explanation generalises is a separate question, but the design lesson is strong: the same feedback schedule can have opposite consequences for different learners.

Educational design should therefore ask what information channel is available to the learner before deciding how aggressively to fade external results.

KR can be immediate, delayed, summary or faded

“How often?” is only one feedback dimension. KR can also vary in timing and aggregation.

• Immediate KR: the result follows the attempt almost immediately.
• Delayed KR: the learner has time to estimate and reflect before receiving the result.
• Summary KR: results from several attempts are reviewed together.
• Average KR: feedback describes the average error pattern across a block rather than each individual attempt.
• Faded KR: feedback is frequent early and becomes less frequent later.
• Bandwidth KR: feedback is given only when error exceeds a defined tolerance.

Each schedule changes the learner’s information problem. Immediate KR can accelerate correction but may reduce self-evaluation time. Summary or averaged KR can reveal trends while preventing trial-by-trial dependence. Faded schedules try to combine early support with later independence.

Delay can force an internal estimate

If KR appears the instant an attempt ends, the learner may never ask what they thought happened. A short delay creates a window for internal error estimation.

In one classic study, delayed KR produced worse acquisition accuracy but better 24-hour retention under certain task conditions. The result supports the possibility that immediate KR can become a crutch when the learner already has usable intrinsic feedback.

In a classroom, the equivalent can be simple: do not reveal the answer the instant the student clicks submit. Ask the student to identify the most uncertain step first. Then reveal the result.

Fading should respond to competence, not the calendar

A fixed schedule such as “feedback every fifth trial” is easy to administer but may not match learning. One learner may still be lost when feedback disappears; another may already be independent.

A stronger fading rule is criterion-based. Reduce KR when the learner can accurately predict the result, maintain performance across several attempts, and recover from small errors without external information. Increase it again when task difficulty changes.

This turns feedback frequency into an adaptive variable rather than a ritual.

Result precision should match the decision

More precise KR is not always more useful. A novice learning to throw may need “short” versus “long,” not an error of 13.7 centimetres. A student revising an essay may need “the claim is not yet supported” before a 20-point analytic score.

Precision earns its cost when it changes the next decision. If two result values would lead to the same action, the extra decimal places are measurement theatre.

Conversely, coarse binary feedback can be inadequate when the direction and magnitude of error matter. “Wrong” does not tell a learner whether an estimate is close, wildly off or wrong for a systematic reason.

Do not confuse KR with grades

Grades are a form of result information, but many grades are too delayed, too aggregated and too detached from a repeat attempt to function as effective KR for learning.

A grade of 72% at the end of a unit tells the learner something about the overall result. It may not tell them which capability to adjust tomorrow, and there may be no tomorrow for that task. KR becomes instructionally powerful when it sits inside a loop where the learner can act again.

This is why small low-stakes result signals during practice can be more educational than a large high-stakes result delivered after the learning window has closed.

Cross-domain comparison: navigation

Imagine navigating toward a location with a compass. A result signal such as “you are 300 metres east of the target” does not tell you exactly how to walk, but it constrains the correction. Repeated location checks help calibrate direction.

If a navigation app gives turn-by-turn instructions every few seconds, you may reach the destination while learning little about the route. This is the guidance problem in another domain: successful assisted performance can coexist with weak independent control.

KR is like a position fix. It should help the learner update a model, not eliminate the need for a model.

A practical KR design sequence

1. Define the target result precisely enough to measure.
2. Identify what result information the task already reveals intrinsically.
3. Ask the learner to predict or estimate the result before external KR.
4. Choose the lowest resolution that can change the next decision.
5. Provide KR often enough that the learner can map actions to consequences.
6. Do not automatically add process explanations when the result signal is sufficient for self-correction.
7. Increase process feedback when repeated KR reveals a stable error the learner cannot diagnose.
8. Introduce delayed or reduced-feedback attempts once the learner’s internal estimates become accurate.
9. Retest without KR to detect dependence.
10. Use changed conditions to test whether the learned mapping transfers.
11. Adjust frequency for age, expertise, task complexity and sensory or cognitive needs.
12. Return to more frequent KR if the task changes so much that intrinsic error signals become unreliable.

Common failure modes

• The score-only trap: the learner sees a result but cannot infer what caused it.
• The explanation-first trap: the teacher gives a full correction before the learner uses the result to search their own performance.
• The every-trial dependency: KR becomes necessary for confidence and adjustment.
• The less-is-always-better myth: feedback is faded despite the learner lacking usable intrinsic signals.
• The more-is-always-better myth: constant KR is treated as harmless because practice scores improve.
• The precision theatre problem: highly detailed numbers do not change the learner’s next decision.
• The no-prediction problem: KR arrives without an internal estimate, so calibration cannot be measured.
• The grade masquerade: a delayed aggregate grade is assumed to function like actionable practice feedback.
• The group-average rule: one feedback schedule is imposed on children, adults, novices and impaired learners alike.
• The no-removal test: nobody checks whether performance survives when KR disappears.

What the evidence says carefully

Decades of motor-learning research show that augmented result feedback can be powerful, but its effect is conditional. Classic guidance studies demonstrate that frequent KR can improve practice while weakening later independence in some tasks. Other studies show frequent feedback can help when tasks are complex or when learners need more external information. Paediatric and clinical research makes the heterogeneity even clearer.

The scientifically defensible position is therefore not a percentage such as “give feedback on 50% of trials.” It is a design principle: match result information to the learner’s ability to perceive error, then reduce external guidance only when internal control can carry the task.

For school learning, most direct KR experiments come from motor tasks, so effect sizes should not be imported uncritically. But the information architecture transfers cleanly: outcome feedback, process feedback, internal estimation and independence are separate variables that can be designed and tested.

Research and further reading

• Lee, White & Carnahan (1990), On the role of knowledge of results in motor learning: exploring the guidance hypothesis.
• Motor learning as a function of KR schedule and characteristics of task-intrinsic feedback — delayed versus immediate KR and later retention.
• Frequent feedback enhances complex motor skill learning — evidence that high frequency can help under complex-task conditions.
• Movement pattern and parameter learning in children: effects of feedback frequency.
• Interaction of feedback frequency and task difficulty in children's motor skill learning.
• The relationship between knowledge of results and motor learning in Parkinsonian patients.
• Feedback protocol of faded knowledge of results is effective for prolonging motor learning retention.
• Effectiveness of different extrinsic feedback forms on motor learning in children with cerebral palsy: a systematic review.

The return

A result is not a correction. It is evidence. Its educational value comes from what the learner can do with that evidence.

Sometimes the learner needs precise KR after every attempt because the task gives almost no usable intrinsic signal. Sometimes the learner needs fewer result updates so they are forced to read their own performance. Sometimes the result is enough. Sometimes it only reveals that a deeper process diagnosis is needed.

Knowledge of results works best when it helps the learner build an internal map of action and consequence—and then proves that the map still works when the external result signal is no longer constantly supplied.

eduKateSG Learning Node Series · 0268 · Previous: 0267 — Attentional Focus Instructions · Explore the How X Works Hub.