Feedback is information about the difference between what was intended and what actually happened, used to improve the next attempt.
In one line: feedback works when a learner knows the goal, sees credible evidence of the current result, understands the important gap, and changes the next action in a way that can be checked again.
A red cross on a page is information. A grade is information. “Good job” is information. A detailed teacher comment is information. But none of these automatically improves learning.
Feedback becomes educationally powerful only when it changes what the learner notices, understands or does next.
What Is Feedback?
Feedback is part of a control loop. There is an intended state, an observed state and some comparison between them. The comparison generates information that can guide correction.
Goal → attempt → evidence → comparison → correction → new attempt.
This structure appears everywhere. A musician hears that a note is sharp and adjusts. A driver notices the car drifting and corrects the steering. A student sees that a mathematics method works until the final substitution and repairs that step. A writer discovers that an argument contains claims but not enough evidence and changes the next draft.
1. Feedback Requires a Clear Target
Correction is difficult when the destination is vague. “Improve your writing” gives much less direction than “make the relationship between the claim and the evidence explicit.”
Good feedback therefore begins with a sufficiently clear success condition. What should a strong answer, performance or process look like? Which features matter most?
Without a target, comments become noise because the learner cannot tell which differences are important.
2. The Attempt Must Produce Evidence
Feedback cannot be more precise than the evidence beneath it. A teacher who sees only a final score knows less than a teacher who can inspect the student’s working, explanation, timing and error pattern.
The same applies to students. If a learner says, “I am bad at comprehension,” the claim is too large. Which questions fail? Literal retrieval? inference? vocabulary? textual support? time?
Better evidence creates a smaller and more repairable problem.
3. Diagnosis Finds the Useful Gap
Not every visible mistake is the real cause. A wrong answer may originate earlier in the chain.
A student loses a mathematics mark because the final answer is wrong. But perhaps the concept was correct, the method was correct and the actual failure was one arithmetic sign. Another student produces the same wrong answer because the chosen method was inappropriate from the start.
Both students need correction. They do not need the same correction.
Useful feedback therefore asks: Where is the earliest meaningful divergence from a successful route?
4. Feedback Must Be Interpretable
Expert comments can fail when they are too compressed for the learner. “Develop analysis” may be obvious to an experienced teacher but not to a student who does not yet know what analysis looks like.
Good feedback reduces ambiguity. It may point to a specific sentence, name the missing relationship, contrast the current answer with a stronger version, ask a discriminating question or model one part of the repair.
The goal is not maximum comment length. It is enough information for the learner to see what to change.
5. Timing Determines Whether Feedback Can Still Change Learning
Feedback received while a mental model is still active can be easier to connect to the original decision. But immediate feedback is not always necessary or desirable. Sometimes learners benefit from attempting correction themselves before seeing the answer.
The important principle is that feedback arrives while it can still influence a relevant next action. A beautifully marked assignment returned after the course has ended may contain excellent information with very little remaining leverage.
6. Feedback Should Point Toward Action
The learner needs a route from information to behaviour.
- Rewrite this paragraph with one claim per sentence.
- Redo these three questions and circle the point where your method changes.
- Explain the concept without using the textbook definition.
- Compare these two examples and state the rule that distinguishes them.
- Attempt the same skill tomorrow without the worked solution.
Action converts feedback from commentary into learning.
7. The Next Attempt Is Part of the Feedback
A correction is still a hypothesis until another performance tests it. The learner may understand the comment yet fail to apply it independently. Or a repair may solve the original question but collapse when the context changes.
This is why strong feedback loops include another attempt. Improvement needs a receipt.
Feedback is complete only when the changed action is observed again.
8. Feedback Also Changes Self-Knowledge
Students are constantly estimating what they know and how well they can perform. Those estimates can be too low, too high or surprisingly accurate.
Repeated prediction, attempt, measured result and update can improve calibration. The learner gradually becomes better at knowing which tasks are secure, which require support and which merely feel familiar.
That makes feedback useful even beyond the immediate task: it can improve the student’s model of their own capability.
The Whole Feedback Chain
Goal → attempt → evidence → identify the important gap → explain it clearly → choose a repair → attempt again → measure the change → update.
A Useful Metaphor: Feedback Is a Mirror With Directions
A mirror tells you what is currently visible. Useful feedback does one thing more: it helps you understand what to change.
A distorted mirror is dangerous because it produces the wrong correction. A mirror that arrives too late is less useful. A mirror that says only “better” or “worse” provides little guidance. The best mirror is accurate enough, timely enough and connected to a next move.
Feedback at Three Zoom Levels
Micro: the specific error
Which word, step, inference or decision needs correction?
Meso: the recurring pattern
Does the same type of failure appear across several questions, drafts or lessons?
Macro: the learner’s control system
Is the student becoming better at noticing errors, seeking evidence, selecting repairs and judging their own readiness?
How Feedback Fails
- No clear target: the learner cannot tell what success requires.
- Wrong diagnosis: the visible mistake is corrected while the earlier cause remains.
- Too much feedback: every flaw is marked and the learner cannot identify priorities.
- Too little information: ticks, grades or praise provide no repair route.
- Identity feedback: comments label the person rather than describe the performance.
- No action: the learner reads the comment and moves on.
- No second measurement: nobody checks whether the correction became independent capability.
How Feedback Is Repaired
Reduce the feedback to the smallest change with the highest expected value. Give the learner something observable to do. Then create another opportunity to use the same capability.
Instead of correcting ten things badly, it is often better to repair one important mechanism well enough that the learner can recognise and control it next time.
What Parents and Students Should Notice
- Can the student explain what the feedback means?
- Do they know exactly what they will do differently?
- Does the feedback identify a process rather than merely a score?
- Is there another attempt soon enough to test the repair?
- Are the same comments appearing repeatedly without behavioural change?
The best feedback is not the comment that sounds most impressive. It is the information that produces a better next performance.
Feedback Has Three Jobs: Direction, Position and Next Move
A high-resolution feedback system answers three different questions. Where am I going? clarifies the goal and success criteria. Where am I now? compares the current performance with that goal. What should I do next? turns the gap into a manageable action.
Feed up → feedback → feed forward → new performance.
If any part is missing, comments lose leverage. A learner can know the error without knowing the destination, know the destination without understanding the present gap, or understand both yet still lack a feasible next action.
Feedback Can Target Different Levels
Not all feedback is about the same thing. Useful feedback may address:
- Task: what is correct, missing or inaccurate in this particular answer.
- Process: how the learner approached the task and which method or reasoning step should change.
- Self-regulation: how the learner plans, monitors, checks, seeks evidence and decides when to change strategy.
- Person: broad judgements such as “clever”, “careless” or “good student”. These usually contain less actionable information and can contaminate performance with identity.
The correct level depends on the failure. A missing unit may need task feedback. Repeatedly choosing the wrong formula may require process feedback. Failing to check an answer even when uncertainty is obvious may require self-regulation feedback.
The Learner Is Part of the Feedback System
Teachers do not inject feedback directly into performance. The learner receives a signal, interprets it through prior knowledge and emotion, decides whether it is credible and useful, and then either changes behaviour or does not.
This explains why the same comment can help one student and confuse another. “Be more concise” is actionable only if the learner understands what unnecessary material looks like and knows what information must remain.
Feedback quality therefore has two sides: the quality of the information sent and the learner’s ability to receive and use it.
More Feedback Can Produce Less Learning
A page covered in corrections may be accurate but unusable. Working attention is limited, and learners need to decide which change matters first. When every flaw is highlighted equally, priority disappears.
High-quality feedback therefore manages information load. It may deliberately leave minor issues for later while targeting the smallest repair likely to unlock the largest improvement.
Maximum information is not the same as maximum useful feedback.
Timing Depends on What the Learner Can Do With Delay
Immediate correction can be valuable when an error is likely to be rehearsed repeatedly or when the learner lacks the knowledge required to self-correct. Delayed feedback can be useful when the learner benefits from first detecting, explaining or repairing the error independently.
The practical question is not “immediate or delayed?” as a universal rule. It is: what timing gives this learner the best chance to think, correct and later reproduce the improvement?
Feedback Needs an Opportunity to Be Used
If students receive comments and immediately move to unrelated work, the system has produced information but not necessarily learning. The learner needs time and a relevant task on which to act.
This can be a redraft, corrected explanation, parallel problem, oral re-explanation, delayed retrieval task or another performance using the same underlying capability. The second attempt is not an optional extra; it is the receipt that tells us whether the feedback changed independent performance.
Feedback-Seeking Is a Higher-Level Capability
As learners mature, they should not depend entirely on adults to notice every gap. They can learn to ask better questions: Which part is weakest? What standard am I missing? Is the problem my knowledge, method, explanation or checking? Can you show me one contrasting example? What should I try again without help?
This changes feedback from something done to the learner into part of the learner’s own control system.
Feedback Can Fail Even When the Comment Is Correct
- Credibility failure: the learner does not trust the source or sees the standard as arbitrary.
- Comprehension failure: the comment uses expert shorthand the learner cannot decode.
- Priority failure: too many corrections obscure the important one.
- Agency failure: the learner is told the answer but never asked to perform the repair.
- Identity threat: feedback is interpreted as a judgement of worth rather than information about performance.
- Transfer failure: the correction works only on the original example.
- Receipt failure: there is no later task to verify that capability changed.
A High-Resolution Feedback Audit
- Goal: Does the learner know what successful performance requires?
- Evidence: Is the feedback based on enough observable performance?
- Diagnosis: Does it locate the earliest useful gap rather than the loudest symptom?
- Level: Is the feedback about task, process or self-regulation rather than global identity?
- Priority: Is the most important repair visible?
- Interpretability: Can this learner understand the language of the comment?
- Timing: Does it arrive while a relevant correction is still possible?
- Action: Is there something concrete the learner can do?
- Opportunity: Is there protected time to act on it?
- Transfer: Can the correction survive a changed example?
- Receipt: Is a later performance measured?
- Calibration: Does the learner’s self-model become more accurate?
Evidence Boundaries: Feedback Is Powerful but Variable
The Education Endowment Foundation’s feedback evidence synthesis reports strong average effects across a large evidence base, while also noting wide variation and studies in which feedback has negative effects. Its Teacher Feedback to Improve Pupil Learning guidance emphasises high-quality initial teaching, clear learning intentions, assessment of learning gaps, actionable feedback and opportunities for pupils to use the information.
The useful conclusion is not “give more feedback”. It is: design a feedback loop in which the right information reaches a learner who can act on it, at a time when another performance can test whether the change worked.
Connect Feedback to the Wider eduKateSG Mechanism Estate
- How Assessment Works — where the evidence comes from and what it justifies.
- How Revision Works — how correction is revisited until it survives delay and transfer.
- How Confidence Works — how feedback updates beliefs about capability.
- How Decision-Making Works — how information becomes a selected next action.
- How Responsibility Works — how learners gradually carry more of the checking and repair loop themselves.
Causal Gateway Handoff
- How the World Works — place feedback inside the wider causal map.
- How Control Systems Work — see feedback formalised as desired state, measurement, correction and stability.
- How Signal Systems Work — inspect whether the returning evidence is detectable, timely and trustworthy.
- How Engineering Works — follow test evidence into redesign, verification and another world-return cycle.
Continue Through eduKateSG
- How Assessment Works
- How Revision Works
- How Feedback Improves Primary English Writing and Comprehension
- How Teaching Works | The Complete Loop
Evidence and Further Reading
The Education Endowment Foundation’s systematic review of feedback approaches examines how different forms of feedback affect student attainment. The important practical lesson is not that every comment improves learning, but that feedback is a powerful learning tool whose design and use matter.
Frequently Asked Questions
Is praise feedback?
It can be, but praise often carries less information about how to improve than task- or process-focused feedback. “Your evidence directly supports the claim here” is more informative than “excellent work.”
Should every mistake be corrected immediately?
No. Sometimes learners benefit from detecting or correcting an error themselves. The right timing depends on the task, the risk of reinforcing a misconception and the learner’s current ability to self-correct.
Why do students keep making the same mistake after receiving feedback?
Understanding a comment is not the same as changing a habit or building a skill. The learner may need guided practice, repeated retrieval, a different explanation or another opportunity to apply the correction independently.
Final compression: Feedback works when accurate evidence reveals a meaningful gap, the learner understands the correction, changes the next action, and a later performance confirms that the repair actually worked.