The essay comes back covered in ink.
Grammar.
Vocabulary.
Evidence.
Paragraph structure.
Thesis.
Twenty-three comments.
The student fixes two commas.
Feedback bandwidth is the amount and complexity of corrective information a learner can interpret, prioritise and act on effectively before the feedback itself becomes another source of cognitive overload.
This is the second pillar beneath How Feedback Works. The master already states that more feedback can produce less learning. This article owns the design problem behind that sentence: how much correction can the receiver actually use now?
Quick Read
Feedback has a bandwidth limit because learners must hold the task, the success criterion, the current error, the proposed correction and the next action in working attention at the same time. When too many comments compete, priority disappears and learners often fix the easiest surface issues while leaving the load-bearing problem untouched. Good feedback therefore compresses, sequences and prioritises. It may target one structural error before sentence polish, separate task feedback from process feedback, use a short correction set followed by another attempt, and carry unresolved lower-priority issues into later drafts. Maximum diagnostic detail can be useful for the teacher; maximum learner-facing detail is not automatically useful for the learner.
performance → detect many gaps → rank by leverage → choose usable subset → explain clearly → learner acts → observe new attempt → release next feedback layer
Information and Usable Information Are Different Quantities
A teacher can correctly identify twelve weaknesses.
The learner may be able to work on only two meaningfully before the next attempt.
The remaining ten observations are still diagnostically valuable.
They simply may not belong in the same feedback packet.
diagnostic bandwidth can exceed action bandwidth.
Working Memory Is Part of the Feedback Channel
To use one feedback comment, the learner may need to coordinate:
- the original task;
- the criterion;
- their current response;
- the teacher’s diagnosis;
- the corrected representation;
- the edit or practice action.
Add ten unrelated corrections and the learner must continually switch standards and representations.
This is why a highly annotated page can be accurate yet hard to learn from.
Priority Is the Compression Layer
Not every error has equal leverage.
A weak thesis can make ten paragraph-level edits premature.
A wrong mathematical model can make arithmetic polish irrelevant.
A misconception can generate several downstream mistakes.
Feedback should identify the smallest repair likely to remove the largest amount of future error.
Load-Bearing Errors Come First
Essay:
argument does not answer the question.
Do not begin with comma placement.
Mathematics:
wrong equation represents the situation.
Do not begin with calculator rounding.
Science:
evidence and claim are disconnected.
Do not begin with replacing one adjective.
The ordering principle is causal:
repair the error whose correction changes the most downstream structure first.
Task, Process and Self-Regulation Feedback Should Not Be Mixed Carelessly
The master distinguishes feedback levels.
- Task: this answer is missing evidence.
- Process: your method of selecting evidence is too broad.
- Self-regulation: you submit before checking whether each claim has support.
All three can be true.
Delivering all three at once may produce a diffuse message.
Choose the level whose change is most likely to improve the next performance.
Surface Errors Are Tempting Because They Are Easy to Mark
Spelling and punctuation are visible.
Deep reasoning gaps are harder to explain.
A feedback sheet can therefore become dense with low-level corrections even when the main problem is conceptual.
Feedback bandwidth should be spent where it produces the largest change, not where marking is easiest.
Comment Density Can Hide the Signal
Every sentence has an annotation.
The learner cannot tell which three comments matter most.
Use visual or structural hierarchy:
- Priority 1: argument;
- Priority 2: evidence linkage;
- Later: sentence economy.
The feedback becomes a route rather than a wall of observations.
A Feedback Packet Should Fit the Next Attempt
If the next task is a ten-minute paragraph rewrite, give feedback that can be acted on in ten minutes.
If the next task is a full essay redraft, the packet can be larger.
If there is no next attempt, even beautifully prioritised feedback has limited instructional bandwidth because action has nowhere to go.
Serial Feedback Can Beat One Giant Batch
Draft 1:
fix thesis and paragraph logic.
Draft 2:
fix evidence integration.
Draft 3:
polish language.
The same total feedback can become more learnable when sequenced across performances.
But Serial Feedback Needs a Persistent Error Ledger
Deferring a minor issue does not mean forgetting it.
Teacher or learner can record:
- current priority;
- deferred issue;
- reason for deferral;
- future checkpoint.
This protects bandwidth without losing diagnostic memory.
The Learner’s Expertise Changes Bandwidth
Novice:
needs fewer, more explicit changes.
Developing learner:
can coordinate several linked comments.
Expert:
may use dense shorthand because the underlying standards are already internalised.
The same feedback density can be helpful to an expert and unusable to a novice.
Feedback Literacy Increases Effective Bandwidth
A learner who can:
- decode rubric language;
- prioritise comments;
- ask clarifying questions;
- plan revisions;
- compare drafts;
can use more complex feedback effectively.
The third sibling, Feedback Uptake, owns that reception and action capability.
Timing and Bandwidth Interact
Ten comments today may overwhelm.
Two comments today, three after the next draft and two after transfer practice may be usable.
The first sibling, Feedback Timing, owns when the correction arrives. Bandwidth owns how much arrives in that packet.
Automated Feedback Can Flood the Channel
Digital tools can generate feedback instantly on:
- grammar;
- style;
- readability;
- structure;
- citations;
- content;
- format.
The low cost of generation makes overload easier, not harder.
A useful AI or automated feedback system needs a governor:
detect broadly → prioritise narrowly → reveal more only after the learner acts.
Do Not Confuse Detailed Feedback With High-Quality Feedback
Detailed can mean:
- specific and useful;
- or verbose and unfocused.
High-quality feedback reduces the right uncertainty at the right level and points to an actionable repair.
Sometimes that takes one sentence.
Feedback Bandwidth Should Follow Error Hierarchy
One practical hierarchy:
- wrong task understanding;
- wrong concept/model;
- wrong strategy/method;
- weak evidence/structure;
- execution error;
- surface polish.
Not every task uses this exact order.
The principle is to fix errors that invalidate later layers before polishing later layers.
Feedback Bandwidth Can Be Negotiated
Ask the learner:
- What are you trying to improve in this draft?
- Which comment is least clear?
- Which two changes can you act on now?
- What should we defer?
This makes feedback selection a shared control decision rather than one-way annotation.
A Bandwidth Receipt
After feedback, the learner should be able to state:
- the one or two main gaps;
- why they matter;
- what changes next;
- which issues are deliberately postponed;
- how the next attempt will show improvement.
If the learner cannot summarise the packet, the bandwidth may be too high or the feedback too ambiguous.
Retesting Determines Whether Bandwidth Was Actually Usable
The learner receives three priority comments.
Next attempt changes none of them.
Maybe the learner ignored the feedback.
Maybe the comments were not understood.
Maybe three was still too many.
The fourth sibling, Feedback Retesting, provides the receipt.
A Better Feedback-Bandwidth Model
rich diagnosis → error hierarchy → receiver capacity → select high-leverage subset → clear action packet → protected revision opportunity → new attempt → release/defer next layer → track recurring pattern
A 30-Lens Feedback Bandwidth Audit
- Task: what is the learner revising?
- Diagnostic count: how many issues were detected?
- Action count: how many can be acted on now?
- Working memory: how complex is each correction?
- Priority: which error has highest leverage?
- Hierarchy: structural or surface?
- Dependency: will fixing one issue remove others?
- Task level: answer-specific or whole-product?
- Process level: is method the real issue?
- Self-regulation: is checking/planning the issue?
- Receiver expertise: novice or expert?
- Rubric literacy: can learner decode the language?
- Comment density: is the page visually overloaded?
- Grouping: can related comments be collapsed?
- Examples: is one model more useful than five explanations?
- Actionability: can each priority become a next move?
- Timing: should feedback be distributed across drafts?
- Opportunity: how much revision time exists?
- Automation: is a tool generating unlimited low-priority comments?
- Deferral: which issues intentionally wait?
- Error ledger: are deferred issues remembered?
- Recurring mechanism: does one pattern explain many comments?
- Teacher shorthand: is compression understandable?
- Student summary: can learner restate priorities?
- Agency: can learner choose among lower-priority issues?
- Emotional load: does volume feel like identity judgement?
- Second attempt: can the packet be used immediately?
- Retest: did the prioritized repair survive?
- Efficiency: did less feedback produce more change?
- World return: did the selected correction set alter later performance, or merely document every imperfection?
Laboratory 1: Twenty Comments to Three
Take a marked piece of work with many corrections. Group them by underlying mechanism, then choose the three highest-leverage comments for the next attempt. Keep the rest in a diagnostic ledger.
Laboratory 2: Structural Before Surface
Mark one draft twice: once correcting every visible surface error, once targeting only the main structural problem. Compare which feedback produces the larger improvement in the next draft.
Laboratory 3: Progressive Reveal
Give one priority comment, let the learner revise, then reveal the next. Compare comprehension and uptake with a condition where all comments appear at once.
For Primary Readers
If someone tells you ten things to fix at the same time, you may forget most of them. Fix the biggest mistake first, try again, then take the next useful correction.
For Secondary Readers
Explain why a heavily annotated page can be less useful than a smaller prioritized feedback set and design a three-step correction sequence for one piece of work.
For Advanced Readers
Model feedback bandwidth as a receiver-limited communication channel in which diagnostic information must be compressed into an actionable subset under attention, expertise and revision-opportunity constraints. Optimal packets prioritise causal leverage rather than total error coverage, with deferred information retained for later release.
Common Misconceptions
- “More feedback means better teaching.” More information can exceed the learner’s action capacity.
- “Correct every error so nothing is missed.” Diagnosis can record every issue while learner-facing feedback remains prioritized.
- “Surface corrections should come first because they are easy.” Structural errors often invalidate surface polishing.
- “Short feedback is low quality.” A precise high-leverage instruction can outperform a long unfocused comment.
- “AI solves feedback capacity because it can generate more.” Unlimited generation increases the need for prioritisation.
Research Corridor
- Education Endowment Foundation — Teacher Feedback to Improve Pupil Learning — actionable feedback and opportunities to use it.
- EEF Teaching and Learning Toolkit — Feedback — strong average effects with important variation.
- Valerie Shute — Focus on Formative Feedback.
- eduKateSG — How Feedback Works.
Frequently Asked Questions
How much feedback should a teacher give?
Enough to identify the highest-leverage gap and give the learner a feasible next action. The right amount depends on learner expertise, task complexity and the next revision opportunity.
Should teachers ignore minor errors?
Not necessarily. Record them diagnostically, but defer some learner-facing corrections when higher-level repairs deserve the available attention first.
Why do students fix easy comments but ignore important ones?
Surface edits are often easier to understand and complete. Important structural comments may require more interpretation, modelling and protected revision time.
Final Thought: Feedback Should Fit Through the Door the Learner Can Actually Open
The best feedback packet is not the one that contains everything the teacher noticed. It is the one that carries the most important repair the learner can understand, act on and prove in the next performance.
FEEDBACK · FOUR PILLAR LEGS
Return to How Feedback Works, or continue through Feedback Timing, Feedback Uptake and Feedback Retesting. Return to the How X Works Hub.