How Feedback Changes Performance | Diagnose, Repair, Retest and Transfer

Quick Read

• Feedback changes performance only when it changes what the learner does on the next task.
• Begin with the first weak link, not the last visible error: task interpretation, reading, evidence, reasoning, language, timing or checking may fail at different points.
• Useful feedback identifies what happened, why it happened, what decision should change and how the repair will be tested.
• A corrected page is not proof of learning. Retest after delay, on changed material and with reduced support.
• Feedback should become progressively less directive as the learner gains diagnostic control.
• Repeated comments should be clustered into mechanisms so one repair can improve many future outputs.
• The goal is not permanent tutor correction; it is independent self-monitoring and transfer.

Feedback begins before the comment: goal, evidence, gap and next action

Feedback works only when four things are sufficiently clear: what successful performance looks like, what the learner has actually done, where the first useful gap lies, and what action is likely to change the next attempt. Without those four elements, a comment can be accurate and still produce almost no learning.

“Improve your vocabulary” names a broad area. It does not tell the learner what failed, which word choice mattered, why the current choice was weak or what to do next. “Good work” may be encouraging, but it does not tell the learner what process should be repeated. A page covered in corrections can show that many errors exist while leaving the student unable to decide which one matters most.

This is why SETC treats feedback as a control loop rather than a marking event:

Goal → attempt → evidence → diagnosis → repair → fresh attempt → retest → delayed retest → transfer → reduced support.

The loop ends only when the learner’s future performance changes.

Feedback and formative assessment are connected

Useful feedback depends on information about the learner’s current state. That means the teacher needs formative evidence: a response, explanation, draft, reading process, oral answer, annotation, plan, timing pattern or error sequence that reveals enough of the learner’s thinking to diagnose.

The Education Endowment Foundation’s current guidance on teacher feedback makes this relationship explicit. It recommends laying the foundations for effective feedback through high-quality initial teaching and formative assessment, because feedback needs a clear learning intention and an identified learning gap. See EEF: Teacher Feedback to Improve Pupil Learning.

This matters for English because many visible errors are downstream symptoms. If a student’s comprehension answer is vague, the problem may not be “poor expression”. The student may have misunderstood the question, selected weak evidence, inferred too far, failed to distinguish two referents or lost the passage model. Feedback should target the earliest useful cause, not simply polish the last sentence.

Feedback is information, not punishment

The learner needs to be able to inspect an error without interpreting every error as a verdict on intelligence or character. The teacher therefore separates performance from person.

Compare:

You are careless.

with:

You identified the correct evidence, but the final sentence changed “may contribute to” into “causes”. Preserve the source’s level of certainty on the next answer.

The second comment gives the learner a decision they can change.

Correction, feedback and instruction have different jobs

Correction identifies or fixes an error.

Feedback provides information that helps the learner regulate a future performance.

Instruction teaches or reteaches a mechanism the learner does not yet control.

Sometimes all three are needed. If a Primary 3 learner writes He go to school yesterday, correction can replace the form, feedback can identify the unstable time relationship, and instruction can rebuild how past time and verb form interact. Merely changing go to went improves the sentence but may not improve the learner.

Marks are evidence, not feedback by themselves

A score tells the learner something about performance level. It does not necessarily identify the mechanism producing the score. Two students can receive 14/20 for entirely different reasons.

Maren may lose marks because her composition has a weak causal structure. Iona may lose the same number because she misreads answer scope. Leonie may know the content but leave the last section incomplete because of timing.

The same mark should not trigger the same repair.

Grades can change how feedback is read

Once a learner sees a grade, attention may shift toward judgement, comparison or emotion. This does not mean grades should disappear. It means teachers should decide what the learner needs to do with the information next.

If the purpose is revision, feedback must remain usable after the score has been seen.

The feedback target should be visible

Learners act more effectively when they know what successful performance requires. A writing rubric, worked example, annotated model, checklist or teacher explanation can make the target visible.

But targets must be translated into observable decisions. “Develop ideas well” is still abstract. A learner may need to see that development can involve explanation, example, consequence, evidence, comparison, qualification or a causal mechanism.

Success criteria are not templates

Criteria clarify what quality requires. They should not force every learner into one identical product.

For narrative writing, “the story must contain dialogue in paragraph three” is a template. “The reader must be able to understand why the character’s decision changes the situation” is a performance criterion.

Feedback should reduce uncertainty

After receiving feedback, the learner should know more clearly:

• what went wrong;
• why it mattered;
• what part of the process produced it;
• what to do differently;
• how to practise the repair;
• how success will be tested.

If the learner remains unsure which action to take, feedback has not yet become operational.

Four levels of feedback

A useful way to diagnose comments is to distinguish the level at which they operate. The categories below overlap with influential feedback research, but SETC uses them as practical teaching layers rather than fixed labels.

1. Task-level feedback

This addresses the current answer or product.

The quotation does not support the claim.

You omitted the recipient’s final request.

The tense is inconsistent with the timeline.

Task feedback is useful when the learner needs to see whether the current performance is correct or complete. Its limitation is transfer. A student can fix this answer and repeat the same error tomorrow.

2. Process-level feedback

This addresses the method that produced the answer.

Before selecting the quotation, state the claim you need evidence for.

After reading the situational-writing prompt, separate role, audience, purpose and required content before drafting.

When the timeline shifts, rebuild the sequence before choosing the tense.

Process feedback is often more transferable because it changes a decision rule.

3. Self-regulation feedback

This helps the learner monitor and control the process independently.

What should you check before deciding this source proves causation?

Which paragraph is currently consuming more time than its mark value justifies?

How will you know that this vocabulary choice fits the register?

Self-regulation feedback begins transferring the teacher’s diagnostic questions into the learner’s own internal system.

4. Self-level praise or judgement

Comments such as You are brilliant, You are careless or You are a natural writer target the person more than the task or process.

Encouragement matters, but identity praise is weak instructional information. It can also create fragility if the learner begins protecting the identity rather than engaging with errors.

SETC therefore praises observable decisions:

You checked the pronoun reference before answering and avoided the ambiguity that caught you last week.

The learner knows what to repeat.

First repairable cause: the central SETC feedback principle

When performance fails, trace backward until you reach the earliest cause that can realistically be changed.

Suppose a Secondary 3 student writes a weak argumentative paragraph.

Surface symptoms:

• repetitive vocabulary;
• weak evidence;
• vague conclusion;
• awkward transitions.

A teacher could correct all four. But the first repairable cause may be earlier: the paragraph has no clear claim. Without a clear claim, evidence cannot be selected precisely, vocabulary has no conceptual target and the conclusion has nothing definite to close.

The first feedback action should therefore stabilise the claim.

Do not repair downstream symptoms first

Downstream correction creates beautiful local patches while the generating problem remains active.

If a student repeatedly uses vague pronouns because paragraph structure is unstable, correcting every this and it can consume hours without repairing the organisational cause.

Failure chains

English performance often follows chains:

Question interpretation → evidence search → inference → answer scope → sentence construction → checking.

Or:

Prompt interpretation → idea selection → causal structure → paragraph plan → drafting → revision → editing.

Feedback should locate the earliest unstable link that explains the downstream loss.

One error can have several possible causes

A spelling mistake may come from weak phoneme–grapheme knowledge, morphology, visual memory, rushed editing or unfamiliar vocabulary. A vague answer may come from weak comprehension or weak expression.

Diagnosis requires evidence beyond the final product.

Feedback should be proportional to leverage

Not every error deserves equal teacher time.

High-leverage feedback changes a mechanism that affects many later performances. Low-leverage feedback repairs a one-off detail with little transfer value.

Examples of high-leverage targets:

• question-scope control;
• claim–evidence alignment;
• sentence reference;
• paragraph purpose;
• timing checkpoints;
• retrieval failure;
• register calibration;
• proofreading routine.

Severity is not the same as frequency

An error can be rare but severe. One misread command word can damage an entire essay. Another error can be frequent but low-cost, such as a repeated minor punctuation issue in an otherwise strong draft.

Prioritisation should consider:

• mark/performance cost;
• frequency;
• generalisability;
• position in the failure chain;
• ease of repair;
• future relevance.

Feedback load: the learner can only act on so much

A heavily annotated page may feel thorough to the teacher and impossible to the learner.

If twenty comments represent three underlying mechanisms, group them.

For example:

• five comments about unclear pronouns → one reference-control target;
• seven comments about weak paragraph links → one argument-progression target;
• eight comments about vocabulary → perhaps one register/precision target.

Priority stack

SETC uses a practical stack:

1. repair the highest-leverage mechanism;
2. repair any error that changes meaning seriously;
3. identify the next recurring pattern;
4. leave cosmetic polish until the architecture is stable.

Feedback budget

For one task, the teacher may deliberately limit the learner to one primary target and one secondary target. This is not ignoring errors. It is sequencing them.

Feedback timing: immediate and delayed feedback solve different problems

Immediate feedback is useful when an error would otherwise be rehearsed repeatedly, when the learner cannot continue productively or when the task is primarily instructional.

Delayed feedback can be useful when the learner needs to complete a process independently before evaluation, when interruption would break fluency or when retrieval after a delay is itself part of the learning.

The question is not “Which timing is best?” but “What learning process are we trying to protect?”

Immediate feedback in grammar practice

If a learner is practising a new sentence pattern and repeatedly produces the same form incorrectly, rapid correction can prevent reinforcement of the error.

Delayed feedback in timed writing

Interrupting every sentence in a timed essay destroys the performance condition. Let the learner complete the task, then diagnose the pattern.

Delayed retest

Even when feedback is immediate, proof of learning should often be delayed. A learner who corrects the sentence while the teacher’s explanation remains active may be demonstrating short-term support rather than independent memory.

IES practice guidance on organising study and instruction recommends spacing learning over time and reviewing key material after delays, alongside retrieval and deep explanatory questions. See IES: Organizing Instruction and Study to Improve Student Learning.

Feedback and retrieval

After feedback, the learner should retrieve the rule or decision rather than merely reread it.

Example:

Teacher: Your answer went beyond the evidence. What is the scope rule?

Learner: I should include only the inference the text supports, not a plausible extra cause.

Then the learner applies the rule to a new item.

The explanation becomes active knowledge.

Feedback and spacing

If a repair is checked only in the same lesson, familiarity can masquerade as mastery. Revisit it after a day, a week and later inside a mixed task where the learner is not told which rule is being tested.

Spacing protects against “I understood it when you told me” becoming the final standard.

Feedback and interleaving

Blocked correction makes the target obvious. If every question requires pronoun-reference analysis, the learner does not need to select the strategy.

Later practice should mix:

• pronoun reference;
• inference;
• main idea;
• tone;
• vocabulary;
• evidence scope.

The learner must decide which mechanism applies.

Feedback and explanation

Asking learners to explain why a correction works can reveal whether the rule is understood or merely copied. The IES practice guide supports deep explanatory questioning as a learning strategy.

But explanation should not become compulsory verbal performance after every minor error. Use it when the mechanism matters.

Feedback delivery: written, verbal, peer, self and automated feedback

The medium of feedback matters less than the quality of the information and what the learner does next. Written comments can preserve detail. Verbal feedback can be immediate and interactive. Peer feedback can expose alternative approaches. Self-feedback can build independence. Automated feedback can increase practice volume. Each medium also has characteristic failure modes.

Written feedback

Written feedback is useful when the learner needs a durable record, when the task is complex enough to revisit, or when examples need to be marked directly in the text.

Strong written feedback is:

• specific enough to locate the issue;
• brief enough to process;
• prioritised;
• linked to a future action;
• consistent in terminology.

A margin full of symbols that the student cannot interpret is not efficient feedback simply because the teacher worked hard.

Codes and symbols

Editing codes can speed marking, but only if the learner understands them and still has to think. If VT means verb tense and the learner must diagnose the correct tense, the code preserves cognitive work. If every code is accompanied by the correct answer, the learner may become dependent on correction.

Comment placement

Local comments are useful for local issues. End comments are useful for patterns. If the same problem appears six times, one end comment can group the evidence and set a repair task.

Verbal feedback

Verbal feedback can reveal misunderstanding immediately because the learner can respond.

Teacher: Why did you choose this evidence?

Learner: Because it was in the same paragraph.

The answer reveals the strategy. The teacher can repair evidence selection rather than sentence wording.

Verbal feedback disappears unless converted into a rule

Students often leave a conversation feeling they understood everything and later cannot remember the decision. End important verbal feedback with a retrieval step:

Tell me the rule you will use next time.

Or record one short action in the learner’s feedback log.

Conferencing

A short writing conference can be more diagnostic than extensive written marking because the teacher can ask the learner to explain choices.

Useful conference questions:

• What were you trying to make the reader understand here?
• Which sentence is carrying the main claim?
• Where did you choose this evidence?
• What were you uncertain about?
• What would you change first if you had five more minutes?

The learner’s explanation distinguishes knowledge gaps from execution failures.

Peer feedback

Peer feedback can create more feedback opportunities and make criteria visible because students evaluate work from the outside.

But peer feedback is only as good as the task design. “Swap essays and give feedback” can produce vague praise, incorrect corrections or social discomfort.

Constrain peer feedback to observable jobs

Better prompts:

• Underline the sentence you think is the paragraph’s main claim.
• Circle the evidence and draw an arrow to the claim it supports.
• Mark one place where you needed to reread.
• Write the question you think this paragraph answers.
• Identify one word whose register feels different from the rest.

The peer provides evidence. The writer makes the final decision.

Peer feedback trains readers as well as writers

When Iona identifies weak evidence in Maren’s paragraph, she practises claim–evidence alignment. That reading practice can later improve her own writing.

Peer authority has limits

Peers can report reader experience accurately even when they cannot diagnose the technical cause.

I lost the connection between these sentences is valuable evidence.

You must add a semicolon may be wrong.

Teach peers to distinguish observation from prescription.

Self-feedback

Self-feedback is the long-term goal: the learner can compare current performance with a target and choose the next action.

This does not mean the learner never needs expert feedback. It means more of the regulation moves inside the learner.

Self-checklists

Checklists work best when they represent known failure risks rather than every possible feature.

Leonie’s timed-writing checklist may contain:

• Did I answer the exact question?
• Is every paragraph doing a different job?
• Did I leave enough time for the final section?
• Did I overstate any evidence?

A fifty-item checklist becomes another task to fail.

Prediction before feedback

Ask the learner to predict where the work is strongest and weakest before showing teacher feedback.

Then compare.

If the learner consistently misses the same weakness, metacognitive calibration needs teaching.

Confidence ratings

For comprehension or grammar questions, learners can rate confidence before checking answers. High-confidence errors deserve special attention because the learner’s internal monitoring is inaccurate.

Automated feedback

Digital systems can provide immediate responses at scale. Useful automated feedback may:

• identify whether an answer matches a known pattern;
• provide hints;
• generate additional practice;
• track recurring errors;
• space retesting;
• visualise performance over time.

The system becomes dangerous when fluent output is mistaken for reliable diagnosis.

Grammar checkers

Grammar checkers can catch spelling, agreement and stylistic patterns, but they may misread intentional style, dialect variation or technical language.

Require the learner to classify suggested changes:

• actual error;
• style preference;
• national-standard variation;
• false positive;
• uncertain—verify.

This turns the tool into reference literacy.

AI feedback

Generative AI can comment on writing, simulate readers, propose revisions and generate practice. Its usefulness depends on preserving the learner’s cognitive work.

A weak AI workflow:

Improve my essay.

The system rewrites. The product improves. The learner may not.

A stronger workflow:

1. Ask the learner to diagnose first.
2. Ask AI to identify one possible weakness without rewriting.
3. Compare AI feedback with the learner’s diagnosis.
4. Verify factual or language claims where needed.
5. Learner revises independently.
6. Use a fresh task later without AI.

SETC’s Digital & AI English owner develops this tool-use system in depth.

AI can generate confident bad feedback

An AI system can misidentify a correct regional form as wrong, invent a grammar rule or recommend stylistic uniformity that damages voice. Feedback should be treated as a hypothesis until checked against the actual language system and target context.

The Corpus, Dictionary & Reference Literacy owner provides the verification layer.

Automated scoring

A score generated by software should not be assumed to measure the construct accurately. If the tool rewards sentence length or vocabulary rarity, students may learn to optimise the metric instead of writing quality.

Always ask what the score is based on.

Feedback conversations: agency matters

Feedback imposed on a passive learner can become teacher-owned improvement. The teacher notices, interprets and fixes; the learner waits.

A stronger conversation makes the learner participate in diagnosis.

Ask before telling

Instead of immediately saying, Your topic sentence is vague, ask:

What exactly is this paragraph claiming?

If the learner cannot answer, the diagnosis becomes visible to them.

Ask for evidence

Show me the sentence that supports this conclusion.

The learner’s search process reveals whether evidence control is stable.

Ask for a repair plan

What will you change first?

The answer shows whether the learner can prioritise.

Ask for a transfer statement

Where else could this same problem appear?

This moves the feedback from one answer toward a general rule.

Ask for the future trigger

What will remind you to use this rule next time?

A rule without a cue may remain inaccessible under pressure.

Feedback language should preserve learner agency

Compare:

Change this to “however”.

with:

The relationship here is contrast, but the current connector signals addition. Choose a connector that makes the contrast explicit.

The second version keeps the decision with the learner.

Directive feedback has a place

Not every error needs a discovery dialogue. If a learner misspells an irregular high-frequency word and no deeper mechanism is involved, direct correction may be efficient.

The art is matching the feedback mode to the learning value.

Feedback and emotional state

English performance can feel personal because writing and speaking expose thought publicly. Learners can interpret criticism as criticism of identity.

This does not mean feedback should become vague praise. It means the environment should make error usable.

Threat narrows attention

When a learner is highly anxious or defensive, they may focus on protecting self-image rather than processing the feedback. The teacher can reduce unnecessary threat by being specific, proportionate and forward-looking.

Specificity reduces global judgement

Your evidence is weak can feel global.

This example shows one individual case; the sentence claims the pattern is widespread. Either narrow the claim or add prevalence evidence.

The learner sees a solvable relationship.

Praise effort carefully

“You worked hard” is supportive but can be misleading if the method was inefficient. Praise effective strategy:

You reread only the paragraph containing the referent instead of restarting the passage. That saved time without losing accuracy.

Trust increases feedback bandwidth

When learners trust that feedback is intended to improve capability rather than humiliate them, they can tolerate more precise challenge.

Trust is built through consistency, fairness, accurate diagnosis and evidence that following feedback actually helps.

Feedback and autonomy

The final direction of feedback is withdrawal.

At first the tutor says:

Check the claim strength.

Later the tutor asks:

What should you check?

Later still the learner checks without prompting.

Eventually the learner can explain the issue to someone else.

That sequence shows feedback becoming internalised.

The prompt ladder

One practical fading sequence:

1. Teacher gives the correction.
2. Teacher identifies the error location.
3. Teacher names the error type.
4. Teacher asks a diagnostic question.
5. Teacher gives a general reminder before the task.
6. Learner uses a self-checklist.
7. Learner performs independently.

Move downward only when the learner can carry the decision.

Do not fade by calendar

Support should reduce based on evidence, not because “we have done this for three weeks”.

Feedback records should track decisions, not comments

A useful feedback log includes:

The log becomes a history of repaired mechanisms.

Feedback across English: the mechanism changes with the performance

Feedback should look different in reading, writing, grammar, vocabulary, speaking and listening because the underlying performances are different. “Be more careful” is weak everywhere because it does not identify the decision that needs to change.

Reading feedback: preserve the passage before repairing the answer

A wrong comprehension answer can come from at least six different places:

1. the reader misunderstood the sentence;
2. the reader lost the referent;
3. the reader misunderstood the question;
4. the reader selected weak evidence;
5. the reader made an unsupported inference;
6. the reader understood correctly but expressed the answer poorly.

The final written answer cannot reliably distinguish those causes. The teacher therefore needs diagnostic questions.

Reading diagnostic sequence

Ask:

• What does the question want?
• Show me the part of the text you used.
• What does that sentence mean in your own words?
• What conclusion are you drawing from it?
• Which word in the question limits the answer?

The learner’s responses reveal the failure point.

Worked case: inference

Iona answers:

The character left because she hated her family.

The passage only states that the character “packed quietly after another argument and left before dawn.”

Weak feedback:

Inference wrong.

Better feedback:

You noticed evidence that the relationship is strained, but “hated her family” is stronger than the text supports. Build the smallest conclusion that explains the evidence. What can you say with confidence?

The learner might revise:

She left because conflict at home had become difficult enough that she wanted to get away.

Then use a fresh passage where the learner must calibrate inference again.

Worked case: answer scope

The question asks, Why did the manager reject the first proposal?

The learner writes three reasons from different stages of the passage, including one that emerged only after the proposal was revised.

The problem is not reading accuracy. It is temporal scope.

Feedback:

All three reasons exist in the passage, but only one existed at the moment of the first rejection. Anchor the answer to the time named in the question.

This kind of feedback teaches task control.

Worked case: pronoun reference

If the learner misidentifies it or they, do not simply write the correct noun. Ask the learner to trace grammatical and semantic candidates.

Which two nouns could “it” refer to? Which one makes the next sentence coherent?

The learner practises reference resolution.

Reading feedback should not become answer revelation

If every difficult item ends with the teacher explaining the answer, the learner may improve understanding of the passage without improving independent question solving.

Use hints that preserve reasoning:

1. point to the relevant paragraph;
2. identify the relationship type;
3. ask for evidence;
4. only then supply the answer if necessary.

Writing feedback: repair architecture before polish

Writing contains several layers:

task → idea → structure → paragraph → sentence → word → punctuation.

Feedback should usually move from higher leverage to lower leverage.

Task interpretation first

If the essay answers the wrong question, sentence polishing is low value.

Maren writes a polished essay about whether technology is beneficial, but the prompt asks whether schools rely too heavily on technology.

Feedback should begin with scope:

Your essay evaluates technology generally. The prompt asks about degree of reliance in schools. Rewrite the thesis so every paragraph can be judged against that narrower question.

Idea feedback

“Add more ideas” is weak. Ask whether the existing ideas are distinct, relevant and developable.

A student may already have too many ideas and too little explanation.

Feedback can say:

You have four reasons, but two are versions of the same point. Combine them and use the space to explain the mechanism and consequence.

Structure feedback

Reverse-outline the draft. Give each paragraph a job label. If the labels do not form a progression, revise architecture before sentence style.

Paragraphs 2 and 3 both explain cost. Paragraph 4 introduces fairness too late. Move fairness earlier because it changes how the cost argument should be evaluated.

Paragraph feedback

A paragraph needs a coherent intellectual or narrative job.

Useful feedback identifies the missing relationship:

• claim has no evidence;
• evidence has no explanation;
• example has no consequence;
• comparison lacks a common criterion;
• narrative action has no causal connection;
• description has no viewpoint.

Sentence feedback

Sentence-level feedback should identify meaning relationships, not only surface form.

This sentence contains two causes and one result, but the punctuation makes all three look equal. Rebuild the hierarchy.

Vocabulary feedback in writing

Do not reward rarity for its own sake.

“Exorbitant” is grammatically correct here, but the sentence is describing emotional pressure rather than price or excessive cost. Choose the word whose meaning matches the relationship.

The repair is semantic precision.

Editing feedback

Editing should move toward self-detection. Mark one or two examples of a recurring pattern, then ask the learner to find the rest.

Teacher marks two subject–verb agreement errors.

Instruction:

There are three more examples of the same pattern. Find them before I show you.

The learner practises error detection.

Grammar feedback: repair the relationship, not the label

Grammar feedback is strongest when it explains what relationship failed.

Agreement

Instead of:

SVA.

Ask:

What is the head noun of the subject? Is it singular or plural?

The learner learns to parse the structure.

Tense

Instead of:

Wrong tense.

Ask:

Which event happened first? Which one is still ongoing? What time anchor does the sentence give the reader?

Timeline understanding produces the form.

Articles

Article errors often involve reference and countability. Feedback should identify whether the noun is being introduced, known, generic, specific, countable or uncountable.

Prepositions

Some preposition errors are collocational rather than rule-general. Use dictionaries and corpus evidence where needed.

The reference system can support feedback instead of forcing the teacher to invent a universal rule that does not exist.

Sentence fragments and run-ons

Teach clause boundaries. Ask the learner to identify each independent proposition and the intended relationship.

Pronouns

Feedback should preserve reference:

Who exactly is “they” here? The paragraph contains three possible groups.

Vocabulary feedback: move from knowing a word to controlling it

Vocabulary has multiple dimensions:

• meaning;
• form;
• pronunciation;
• word family;
• grammar;
• collocation;
• register;
• connotation;
• retrieval;
• transfer.

Feedback should name the dimension that failed.

Meaning error

You used “reluctant” as if it meant unable. It means unwilling or hesitant. Rewrite the sentence so the cause is unwillingness.

Collocation error

“Powerful rain” is understandable, but English strongly prefers “heavy rain”. Record the phrase, not just the adjective.

Register error

“Kids” is natural in conversation, but this formal report needs “children” or a more specific group label.

Word-family error

This position requires an adjective. You know “analysis”; retrieve the adjective from the family.

Retrieval failure

The learner recognises a word in exercises but never produces it independently. Feedback should not add another definition. Create retrieval cues and require use after delay.

Vocabulary feedback should not become thesaurus inflation

If a learner replaces every common word with a rarer synonym, feedback should restore purpose:

Which word is more precise, not which word is harder?

Speaking feedback: preserve fluency while improving control

Interrupting every spoken error can destroy interaction and increase monitoring load. Feedback timing depends on the task.

During fluency practice

Let the learner complete a meaningful stretch of speech unless misunderstanding prevents continuation. Note recurring patterns.

After the response

Choose one or two high-value targets:

• idea development;
• response structure;
• pronunciation affecting intelligibility;
• repeated grammar pattern;
• interactional repair;
• register.

Pronunciation feedback

Avoid “fix your accent”. Identify the feature:

The word stress on “analysis” moved, and the listener needed a second attempt. Mark the stress and practise it inside the phrase “data analysis”.

Fluency feedback

Fluency does not mean speed. If the learner speaks rapidly but abandons sentence structure, feedback may target pacing and chunking.

Interaction feedback

You answered your prepared point but did not respond to the questioner’s distinction. Start by paraphrasing the question before extending your position.

Listening feedback: diagnose what disappeared

A wrong listening answer can come from:

• sound recognition;
• vocabulary;
• syntax;
• attention;
• note-taking overload;
• inference;
• failure to recover after missing one phrase.

Replay diagnostically

Do not replay the entire recording automatically. Locate the region where meaning was lost.

Ask:

• What did you hear?
• Which word or relationship was uncertain?
• Did you know the word in print?
• Did your notes pull attention away?
• Did missing one item damage the next two?

Transcript use

Transcripts are useful after the learner attempts to recover the audio. Compare heard vs written forms, then remove the transcript and listen again.

Recovery feedback

If the learner misses one detail and then mentally replays it for thirty seconds, the later failure is attention recovery.

Feedback:

Mark the missed item and re-enter immediately. Protect the next information.

Feedback under examination conditions

Exam feedback should preserve the difference between knowledge and performance.

A student may know the rule but fail under time. Another may complete comfortably but misunderstand the construct.

Post-paper analysis

Classify lost marks:

• knowledge gap;
• question interpretation;
• evidence selection;
• reasoning;
• expression;
• timing;
• checking;
• recovery.

Then build the next practice around the dominant mechanisms, not around repeating the entire paper immediately.

Score stability

Feedback should examine variance. If the student scores 75, 58, 76, 61 and 74 across comparable papers, the central problem may be reliability rather than maximum capability.

Look for states that predict collapse: one hard question, fatigue, late-paper timing, unfamiliar topic, poor question choice.

Recovery is feedback-worthy

Students are often told only what they should have known. They also need feedback on what to do when the paper goes wrong.

You spent nine minutes trying to repair one inference answer. The next three questions were then rushed. Set a two-minute containment rule and return later.

Feedback by developmental stage

Early and Primary English

Feedback should be concrete, immediate enough to support learning, and linked to visible actions. Avoid abstract labels the child cannot interpret.

Point to the word that tells you where the character is.

Read the sentence again and make your voice stop at the full stop.

Upper Primary / PSLE

Begin shifting toward self-questioning. Students need to manage answer scope, planning, editing and timed performance with progressively less prompting.

Secondary English

Feedback should increasingly target interpretation, evidence, genre, audience, argument, independence and transfer across unfamiliar tasks.

Advanced English / GP

Feedback becomes more epistemic: claim strength, evidence quality, assumptions, definitions, counterarguments, synthesis, knowledge selection and rhetorical calibration.

The learner should increasingly be able to challenge or refine the feedback with evidence.

How do we know feedback worked? The proof is in later performance

A corrected page is not proof of learning. A student can copy the teacher’s replacement sentence perfectly and still reproduce the original error in the next task. Feedback must therefore be evaluated through retest and transfer.

Four levels of evidence

SETC distinguishes four increasingly strong forms of evidence:

1. Correction: the current error is fixed.
2. Reproduction: the learner can perform the same operation again in a very similar task.
3. Selection: the learner can recognise when the operation is needed among mixed alternatives.
4. Transfer: the learner can adapt the operation to a new surface, genre, topic or time condition.

Feedback has limited educational value if it never moves beyond correction.

Immediate retest

After feedback, use a short fresh item that requires the same mechanism.

Example: Iona over-infers in a comprehension question. After explanation, she receives a different short passage with a new inference problem.

If she succeeds, the repair may be understood. It is not yet durable.

Delayed retest

Return after enough time for the teacher’s wording and exact example to become less available.

Delay matters because immediate performance can depend on working memory and short-term familiarity.

A delayed retest asks whether the learner can retrieve the decision later.

Mixed retest

Do not tell the learner which skill is being checked.

Mix several question types. The learner must recognise the relevant mechanism.

This is a stronger test than repeated blocked practice because real English tasks rarely announce, “Use the inference rule now.”

Unseen retest

Change the content. If the learner practised answer scope in a narrative, test it in an informational text. If they practised evidence qualification in an essay about technology, use a different topic.

Constraint retest

Add the real constraint: time, oral delivery, limited word count, no notes, exam conditions or audience shift.

A mechanism that works only in calm untimed practice is not yet performance-ready.

Cross-skill retest

Sometimes feedback should transfer across skills.

A vocabulary precision lesson from reading should later appear in writing.

A claim–evidence relationship learned in comprehension should appear in oral explanation.

A register distinction taught in situational writing should appear in speech or email.

The feedback evidence ladder

Feedback should produce replacement behaviour

“Do not make this mistake” leaves a vacuum. The learner needs a replacement action.

Examples:

• Instead of do not over-infer → state the evidence and build the smallest conclusion that explains it.
• Instead of do not rush → check the clock after Question 5 and move on if one answer exceeds two minutes.
• Instead of do not use vague vocabulary → name the exact relationship: harmful, inefficient, unequal, unstable, misleading.
• Instead of do not write run-ons → identify the independent clauses before choosing punctuation.

Replacement actions are executable.

Error logs should store rules, not shame

An error log should not become a museum of failure.

Each entry should include:

• error pattern;
• first repairable cause;
• replacement action;
• one model;
• retest date;
• current status.

Remove or archive items once stable so the active log remains useful.

Error frequency needs a denominator

If the learner makes three article errors in a 50-word paragraph and three article errors in a 1,500-word essay, the raw count is identical and the control level is not.

Track errors relative to opportunities where useful.

Feedback metrics

Useful measures can include:

• error rate per relevant opportunity;
• accuracy on unseen items;
• time to complete a task;
• number of prompts required;
• percentage of feedback points self-detected;
• stability across tasks;
• performance after delay;
• transfer across genres;
• recovery after difficulty.

The metric should match the mechanism.

Do not turn feedback into metric chasing

If students know they are rewarded for “using three advanced words”, they may insert unnatural vocabulary. If a tutor tracks only corrections per page, students may write simpler sentences to reduce error.

Measures can distort behaviour when they become the target.

Use metrics as evidence, not definitions of quality.

Feedback and score improvement

Marks matter, especially in examination preparation. But score change can lag behind mechanism change.

A student may improve evidence selection while still losing marks from timing. Another may gain several marks because one high-cost error is repaired.

Keep both levels visible:

mechanism evidence + whole-performance evidence.

Variance matters

Average score can hide instability.

Suppose a learner’s five paper scores are:

72, 48, 74, 51, 73.

The average is 63.6, but the educational story is not “mid-60s student”. The learner has high capability with repeated collapse states.

Feedback should identify what differentiates the strong and weak performances.

State variables

Possible variables:

• fatigue;
• question order;
• topic familiarity;
• one early error;
• time allocation;
• confidence collapse;
• unfamiliar format;
• overchecking.

Feedback can target performance regulation as well as knowledge.

Feedback and confidence calibration

Confidence should increasingly match accuracy.

Track:

• high confidence + correct;
• low confidence + correct;
• high confidence + wrong;
• low confidence + wrong.

High-confidence errors are dangerous because the learner is unlikely to self-correct.

Low-confidence correct answers can indicate fragile knowledge or underconfidence.

Feedback on metacognition

Metacognitive feedback helps learners judge their own state.

Instead of only:

This answer is wrong.

Add:

You rated this 95% confident. What evidence made you so certain? Which signal should reduce your confidence next time?

Feedback and retrieval practice

Retrieval after feedback should focus on the decision rule.

Ask the learner to retrieve:

• the rule;
• an example;
• a counterexample;
• the trigger for using it.

Then apply it.

Feedback and spacing schedule

A practical schedule may be:

• same lesson: immediate fresh item;
• next lesson: mixed item;
• one week: unseen application;
• later unit: authentic transfer;

The exact interval depends on age, frequency of lessons and importance of the skill.

Feedback and interleaving

After a mechanism is understood, mix it with neighbouring mechanisms that learners confuse.

For reading:

• literal retrieval;
• reference;
• inference;
• tone;
• summary;
• evaluation.

For grammar:

• tense;
• agreement;
• pronoun reference;
• sentence boundary;
• preposition/collocation.

Selection becomes part of mastery.

Feedback and variation

Change the surface while preserving the mechanism.

If the learner practises claim–evidence alignment only on environmental essays, they may associate the process with the topic. Move to education, technology, health or a fictional school policy.

Feedback and fading

Track prompt level as data.

Progress can occur even before accuracy changes completely if the learner needs less external support.

Feedback and transfer distance

Near transfer asks the learner to apply the same mechanism in a similar context. Farther transfer changes more dimensions.

Example sequence for paragraph evidence:

1. same genre, new topic;
2. new essay question;
3. comprehension explanation;
4. oral argument;
5. academic source synthesis.

The mechanism is increasingly portable.

Feedback can fail through overfitting

If the teacher gives one memorable sentence frame and the learner reproduces it everywhere, feedback has improved local performance while reducing adaptability.

Teach the function behind the frame.

Model answers and feedback

Model answers are useful when treated as evidence of decisions.

Ask:

• What problem is this sentence solving?
• Why is this evidence sufficient?
• Where does the writer qualify?
• What could change in another prompt?

Do not ask learners to copy the model and call the result transfer.

Feedback and exemplar comparison

Comparing two samples can make quality differences visible.

Use contrast pairs:

• same idea, different evidence;
• same evidence, different explanation;
• same grammar, different register;
• same paragraph, different organisation.

The learner identifies the feature causing the difference.

Feedback and self-explanation

Ask learners to explain their revision:

I changed “proves” to “suggests” because the source is correlational.

This is stronger evidence than simply seeing the corrected word.

Feedback and teaching decisions

Teacher feedback should also update teaching.

If ten learners make the same error, the problem may lie in instruction, materials or task design rather than ten independent failures.

Feedback is information for the teacher too.

Class-level patterns

Track whether a pattern is:

• individual;
• group-wide;
• task-specific;
• instruction-induced;
• newly emerging under increased difficulty.

Feedback on feedback

Ask learners:

• Which comment helped you change something?
• Which comment did you not understand?
• Which correction keeps returning?
• Which feedback do you now apply without help?

If students repeatedly cannot interpret a feedback code, the teacher should change the feedback system.

The feedback release test

A feedback target is ready to leave active monitoring when the learner can:

• identify the risk independently;
• choose the correct repair;
• apply it on unseen material;
• retain it after delay;
• perform under realistic constraints;
• recover if the old error reappears;
• explain the rule in their own words.

Release does not mean the learner will never make the error again. It means the learner owns enough of the repair system to manage recurrence.

Feedback as a programme: organise change across weeks, not pages

A strong feedback system does not restart from zero every lesson. It accumulates evidence. The teacher knows which mechanisms are stable, which are improving, which have relapsed, which appear only under pressure and which no longer need active attention.

This turns feedback from a marking habit into a learning programme.

The feedback register

For each learner, maintain a small active register rather than an endless comment archive.

Useful fields:

• current primary target;
• secondary target;
• first repairable cause;
• replacement action;
• prompt level currently required;
• last successful retest;
• next delayed retest;
• transfer contexts already passed;
• current confidence/stability.

The register should be short enough to influence teaching decisions in real time.

Active vs archived targets

When a target stabilises, move it to archive. If it relapses later, reactivate it with the new context.

This prevents the learner from being defined forever by old weaknesses.

Weekly feedback cycle

A practical weekly rhythm might be:

Lesson 1: diagnose and isolate

Use current work to identify the target. Teach or repair the mechanism. Give immediate fresh practice.

Lesson 2: retrieve and vary

Do not reteach automatically. Ask the learner to retrieve the rule, then apply it to a different surface.

Lesson 3: interleave

Mix the target with neighbouring skills so the learner must select it.

Lesson 4: authentic performance

Return the mechanism to a full reading, writing, oral or examination task.

Later: delayed transfer

Check again when the learner no longer expects the target.

The exact schedule can compress or stretch depending on lesson frequency.

Feedback sprints

Some weaknesses benefit from short concentrated sprints.

Example: Secondary 2 sentence reference.

For two weeks:

• diagnose ambiguous pronouns;
• repair with explicit noun tracking;
• edit mixed paragraphs;
• apply in own writing;
• retest after delay.

Then release the target into ordinary work.

Do not keep the entire curriculum trapped inside one weakness.

Feedback spirals

Important mechanisms return at increasing complexity.

Primary answer scope becomes PSLE precision, Secondary evidence boundaries and GP source synthesis. The feedback language changes, but the core relationship remains.

A federated English system should preserve those longitudinal connections.

Three-student tutorials: feedback visibility without tutor monopoly

SETC’s three-student model creates a distinctive feedback environment. The group is small enough for individual reasoning to remain visible and large enough for learners to compare different approaches.

Three answers expose three mechanisms

Give Maren, Iona and Leonie the same question.

Maren gives a sophisticated but overlong answer.

Iona gives a precise answer but misses one inference.

Leonie gives the correct answer quickly but cannot explain the evidence.

The tutor can now make the differences instructional.

Students see that “correct/incorrect” hides important performance dimensions.

Peer contrast

Ask:

Which answer would you trust most and why?

Which one is safest under exam conditions?

Which one would transfer best if the passage changed?

Peer contrast develops evaluative judgement.

Public reasoning, private dignity

The group can compare strategies without turning one student into an example of failure.

Use anonymised versions where sensitive.

Rotating expertise

Maren may be strongest at idea structure, Iona at evidence precision and Leonie at timing. The tutor can deliberately rotate moments where each student provides a useful observation.

This builds agency while preventing fixed ability labels.

Teacher visibility

Three students let the tutor hear explanations, watch time decisions and inspect drafts closely enough to distinguish the first repairable cause.

The advantage is not the number three itself. It is diagnostic resolution.

Feedback and tutor planning

A tutor should enter a lesson knowing:

• what was last repaired;
• what needs delayed retest;
• what whole-task performance is due;
• which scaffold should be reduced;
• which new content can be introduced safely.

This makes every lesson part of a longitudinal model.

Feedback and lesson notes

Lesson notes should record decisions, not narratives of activity.

Weak note:

Did comprehension, grammar and composition.

Useful note:

Comprehension: evidence selection now stable on literal questions; inference scope still overextended when background knowledge is strong. Composition: paragraph claims clearer after reverse-outline prompt; next lesson remove prompt and test new topic.

The second note supports future teaching.

Feedback and curriculum planning

If several targets depend on one prerequisite, repair the prerequisite first.

Vocabulary imprecision, weak summary and vague writing may all reflect inadequate distinctions among key concepts.

Do not create three separate programmes if one mechanism explains them.

Feedback and homework

Homework should test the repair rather than merely add volume.

After a feedback lesson:

• one short retrieval task;
• one varied application;
• one delayed task;
• perhaps one authentic transfer.

Ten pages of repetitive practice can hide whether the learner can select the method.

Feedback and parents

Parents often receive marks and comments without knowing what to do with them.

Useful parent communication explains:

• what pattern is being observed;
• what the tutor is repairing;
• what evidence will show improvement;
• what support at home is useful;
• what parents should avoid taking over.

Parent support should not become second marking

If the tutor is teaching the learner to self-edit, a parent correcting every sentence before submission can remove the evidence the tutor needs.

Parents can instead ask:

What are you checking today?

What feedback target are you working on?

How will you know you fixed it?

When marks fall

Do not respond with random extra practice. Ask which part of the performance chain changed.

The learner-state route Marks Are Falling begins with diagnosis.

When marks are stable

Feedback can shift from repair toward robustness: new genres, harder texts, faster timing, more independence.

The Keep Them Steady route develops stability.

Feedback and school teachers

Classroom feedback has constraints that individual tutoring does not: many students, limited marking time, curriculum deadlines and shared assessments.

A strong system can still preserve diagnostic principles.

Whole-class feedback

When many learners share patterns, give class-level feedback:

• three common strengths;
• three high-leverage issues;
• anonymous examples;
• repair tasks;
• individual exceptions.

This can reduce repetitive written comments while preserving instructional value.

Live feedback

While students work, circulate and intervene at points where a small prompt prevents a large failure chain.

Mini-lessons from evidence

If formative evidence shows widespread confusion, pause and teach the mechanism rather than writing the same correction thirty times.

Feedback policies should not reward marking volume

A policy that requires a fixed number of written comments can encourage visible labour rather than learning.

The EEF feedback guidance explicitly emphasises principles over assumptions about written versus verbal methods and notes that feedback carries opportunity costs. A sustainable system should focus teacher time where it changes learning most. See EEF feedback guidance.

Feedback and workload

Feedback has a cost for both teacher and learner.

Teacher time spent correcting one low-leverage detail is time unavailable for planning, diagnosis or instruction.

Learner attention spent processing twenty comments is attention unavailable for practising the main repair.

Efficiency is therefore an educational issue, not merely administrative convenience.

Feedback batching

Group similar patterns before responding.

If five grammar errors share one cause, teach the cause once and use the five errors as examples.

Feedback templates

Reusable prompts can improve efficiency if they remain diagnostic:

• What is the exact claim?
• Which evidence supports it?
• What does this pronoun refer to?
• Which word sets the answer scope?
• What changes if the audience changes?

Templates should ask reusable questions rather than produce reusable answers.

Feedback libraries

A mature teaching system can maintain a library of:

• diagnostic prompts;
• contrast examples;
• mini-practice items;
• retest items;
• transfer tasks;
• common failure chains.

This reduces preparation cost without reducing personalisation.

Feedback and curriculum data

Across many students, anonymised pattern data can reveal curriculum weaknesses.

If 70% of learners repeatedly misread comparison questions, the programme may need stronger comparison instruction.

Do not use learner data only to rank students. Use it to improve teaching design.

Feedback and fairness

Learners should understand the criteria used to judge work. Hidden expectations make feedback difficult to act on.

Fairness also requires distinguishing language proficiency from unrelated factors where possible.

An oral assessment intended to measure reasoning should not silently become an accent-prestige test.

Feedback and multilingual learners

Do not assume every language difference is caused by another language. But cross-linguistic comparison can help identify patterns.

If a learner already understands the concept in another language, feedback may focus on English expression rather than reteaching the concept.

The Multilingual English & Mediation owner provides the broader system.

Feedback and neurodiversity / learner variability

Different learners can need different representations, pacing, memory supports or response modes. The feedback target should remain intellectually clear while the delivery adapts.

Do not confuse a support need with reduced expectations about eventual independence.

Feedback and motivation

Motivation improves when learners can see a credible path from current state to improvement.

Feedback should therefore make progress visible:

Last month you needed me to point out every scope error. Today you found three of four yourself before checking.

This is evidence-based encouragement.

Feedback and boredom

Once a mechanism is stable, repeated correction can become demotivating and unnecessary. Increase variation or complexity instead.

Feedback and challenge

Difficulty should rise when the learner can maintain quality with reduced support.

Challenge can increase through:

• denser texts;
• less familiar topics;
• greater ambiguity;
• shorter time;
• new genre;
• mixed skills;
• less scaffolding.

The programme-level question

At the end of a month, ask:

Which learner decisions now happen correctly without us?

That is a more meaningful feedback outcome than the number of pages marked.

Worked diagnostic cases: the same symptom can need different feedback

Feedback becomes precise when the teacher resists the temptation to name the subject instead of the cause. “Weak comprehension”, “poor grammar” or “bad writing” are categories. Diagnosis asks what decision failed.

Case 1: two students both write vague comprehension answers

Student A understands the passage but writes:

He was upset because of what happened.

When asked orally, the student gives the exact reason immediately.

First repairable cause: expression precision.

Feedback:

You understood the reason. Your written answer hides it inside “what happened”. Name the event explicitly.

Student B gives the same vague answer but cannot explain the event orally.

First repairable cause: comprehension model.

Feedback:

Before writing, reconstruct what changed between the two paragraphs. Which event triggered the reaction?

Same answer. Different feedback.

Case 2: two students both make tense errors

Student A knows the timeline but forgets an irregular past-tense form.

Repair: lexical retrieval of the verb form.

Student B uses several correct verb forms but shifts between past and present because the narrative timeline is unstable.

Repair: temporal representation and editing cue.

Giving both students “revise past tense” wastes information.

Case 3: composition contains weak vocabulary

Maren’s composition uses ordinary words: nice, bad, sad, happy, big.

One possible cause is limited vocabulary. Another is limited conceptual differentiation.

Ask Maren what exactly was bad.

If she says, I don’t know, just bad, the deeper target is conceptual precision.

If she says, It was unfair because only one group had access, but still writes bad, the concept exists and lexical retrieval is weak.

The first learner needs distinctions; the second needs vocabulary activation.

Case 4: a student cannot finish essays

Leonie repeatedly runs out of time.

Possible causes:

• planning too long;
• writing introductions too carefully;
• slow handwriting;
• overchecking every sentence;
• weak idea retrieval;
• getting trapped by one paragraph;
• poor time awareness.

Collect timing evidence. Do not prescribe “write faster” until the bottleneck is known.

Case 5: learner performs well in tuition and poorly at school

This is a classic transfer problem.

Check whether tuition tasks contain hidden cues:

• teacher tells the learner which method to use;
• questions are grouped by skill;
• examples resemble models closely;
• teacher rescues quickly;
• timing is generous;
• vocabulary is pre-taught;
• feedback arrives before the full task is complete.

Feedback should move toward cue removal, not more explanation.

Case 6: student repeats the same teacher comment for months

The margin repeatedly says elaborate.

Ask what elaboration means operationally.

If the learner cannot answer, the feedback label never became a mechanism.

Teach several elaboration moves:

• explain why;
• give an example;
• show consequence;
• compare;
• qualify;
• connect evidence to claim.

Then retest without the word elaborate.

Case 7: student edits successfully only after the teacher points to the line

The learner possesses repair knowledge but not detection.

Feedback should shift from correction to scanning strategy.

Example:

Today I will not mark the sentence. There are four reference errors in the draft. Use your pronoun-check routine to find them.

The target is self-detection.

Case 8: student asks “Is this correct?” after every sentence

The problem may be linguistic insecurity rather than language knowledge.

Feedback strategy:

1. ask the learner for their own judgement;
2. ask for the rule or evidence;
3. confirm only after reasoning;
4. reduce confirmation frequency over time.

The goal is calibrated independence.

Case 9: strong student resists feedback

A high-performing learner can interpret feedback as a threat to identity.

Use evidence, not status:

Your argument is strong. The issue is that the source supports correlation and the sentence claims causation. Let’s check the wording against the evidence.

The feedback preserves achievement while insisting on precision.

Case 10: weak student receives only encouragement

Kindness without usable information can become abandonment.

Keep trying—you’ll get there may support emotion but not learning.

Add one clear route:

You found the correct paragraph. Next, we are working on turning the evidence into a one-sentence answer without adding background knowledge.

Case 11: student’s oral response is fluent but empty

The learner speaks confidently for one minute without developing a claim.

Feedback should not target fluency. It should target content architecture:

After your answer, I can hear your position but not the reason. Use position → reason → example → implication.

Case 12: hesitant speaker has strong ideas

The learner gives excellent written reasoning but pauses heavily in oral tasks.

Feedback should target retrieval speed and spoken chunking, not idea generation.

Practise recurring oral frames under gradually reduced preparation.

Case 13: reading speed is slow

Possible causes:

• decoding;
• vocabulary;
• syntactic parsing;
• over-annotation;
• rereading habit;
• lack of text structure knowledge;
• fear of missing detail.

Feedback should follow the bottleneck.

Case 14: student memorises model essays

The product can look strong until the prompt changes.

Do not respond by giving more models.

Ask the learner to extract reusable decisions:

• how the introduction frames the problem;
• how evidence is integrated;
• how paragraphs progress;
• how the writer qualifies.

Then remove the model and use a new topic.

Case 15: vocabulary list scores are high but writing stays ordinary

The learner has recognition but weak production transfer.

Feedback:

You know the words when cued. Now we need retrieval by meaning. Before drafting, identify the distinction you need, then retrieve the word without seeing the list.

Case 16: student “knows grammar” but errors reappear under time

The learner can explain the rule calmly but cannot monitor it while composing.

This is an automaticity/integration problem.

Use timed short writing with one target, then gradually increase content load.

Case 17: student’s summary copies phrases

Possible causes:

• weak understanding;
• fear of meaning change;
• limited paraphrase resources;
• poor hierarchy;
• belief that synonyms equal paraphrase.

Ask the learner to explain the source orally with the text closed. If the meaning cannot be reconstructed, paraphrase feedback is premature.

Case 18: learner gets better after feedback then relapses

Relapse is information.

Ask:

• Was the retest delayed?
• Was the skill mixed with others?
• Was support removed?
• Did the learner recognise the trigger?
• Did complexity increase?

The repair may need stronger retrieval or broader variation rather than more initial explanation.

Case 19: student receives contradictory feedback from two teachers

First classify the disagreement:

• different style preferences;
• different target standards;
• different interpretations of the task;
• one factual error;
• genuine judgement difference.

Use rubrics, dictionaries, corpora or official specifications to resolve what can be resolved. Preserve legitimate judgement differences where they remain.

Case 20: AI and teacher feedback disagree

Do not choose by confidence or fluency.

Identify the claim and verify against the language system, source evidence or official criterion.

This is where SETC’s Reference Literacy owner becomes part of feedback.

The diagnosis rule

Before giving a major feedback intervention, the tutor should be able to complete this sentence:

The learner is losing performance because ________, and the next task will test whether changing ________ improves ________.

If the blanks cannot be filled clearly, more evidence may be needed before more feedback.

What research can and cannot tell us about feedback

Feedback has a strong research base, but “feedback works” is too broad to guide teaching. Different studies examine different ages, subjects, delivery methods, outcome measures and time periods. Effects vary because the quality of feedback and the conditions around it vary.

The useful conclusion is not that every comment helps. It is that feedback can support substantial learning when it is connected to clear goals, accurate diagnosis, actionable next steps and opportunities to respond.

EEF: feedback quality over medium

The Education Endowment Foundation’s guidance on teacher feedback emphasises that feedback should be grounded in high-quality initial teaching and formative assessment. It also moves beyond the simplistic question of written versus verbal feedback and focuses on principles. See Teacher Feedback to Improve Pupil Learning.

This supports SETC’s design decision: the medium is secondary to whether the learner can use the information to close a meaningful gap.

EEF: feedback has opportunity cost

Feedback takes time. A system that asks teachers to generate large quantities of low-leverage comments can reduce time available for planning, teaching, diagnosis and responsive instruction.

Therefore, “more feedback” is not automatically better.

IES: retrieval, spacing and explanation strengthen durability

The Institute of Education Sciences practice guide Organizing Instruction and Study to Improve Student Learning recommends spacing learning over time, using retrieval through quizzes, interleaving worked examples and problems, integrating abstract and concrete representations and asking deep explanatory questions.

These principles matter after feedback. The corrected mechanism must be retrieved again, varied, mixed and explained if it is to become durable.

Feedback is not independent of instruction

Poor initial teaching cannot always be repaired efficiently through increasingly detailed feedback. If most learners misunderstand the concept, reteaching may be the appropriate response.

Feedback should not become a substitute for curriculum quality.

Feedback effects depend on the target

Correcting a factual error is different from developing argument, fluency or creative writing. Some targets have clear right answers; others involve judgement.

The teacher should avoid pretending that subjective decisions are mechanical and avoid treating objective errors as mere preferences.

Feedback effects depend on prior knowledge

Novices may need more explicit guidance because they lack the internal model needed to interpret a hint. Advanced learners may benefit from questions that require diagnosis and self-regulation.

One feedback style should not be imposed across all levels.

Feedback effects depend on motivation and trust

Learners can ignore, resist or misunderstand accurate feedback. The teacher needs enough relationship and clarity for the learner to engage with the message.

This does not mean feedback must always feel pleasant. It means challenge should remain interpretable as a route to improvement.

Feedback effects depend on opportunity to act

A comment on a final assignment that the learner never revisits may have little transfer unless the same mechanism is deliberately carried forward.

Build response time into the system.

Common myths about feedback

Myth: every error should be corrected

Correction volume is not equivalent to learning. Prioritise by leverage.

Myth: written feedback is always more serious

A two-minute verbal diagnostic conversation can change more than a page of written comments.

Myth: verbal feedback leaves no evidence

The evidence is the learner’s next performance. Important verbal targets can also be recorded briefly.

Myth: immediate feedback is always best

Immediate feedback is useful for some learning conditions. Delayed feedback protects uninterrupted performance and can support independent monitoring. The right timing depends on the task.

Myth: praise is feedback

Praise can support motivation, but instructional feedback must tell the learner something useful about performance or process.

Myth: feedback should explain everything

Excess explanation can remove the learner’s cognitive work. Sometimes one diagnostic question is enough.

Myth: red ink demotivates, therefore do not correct

The colour is less important than the meaning, density, tone and usability of the feedback.

Myth: learners should discover every correction themselves

Discovery has a cost. For simple factual or form errors, direct correction can be efficient. Preserve problem solving where it has learning value.

Myth: feedback should focus only on weaknesses

Learners also need to know which decisions worked so they can reproduce them.

Myth: positive feedback means saying something nice first

The common “praise–criticism–praise” sandwich can become predictable and dilute the message. Genuine strengths should be identified because they matter, not because feedback requires a ritual.

Myth: detailed rubrics automatically improve feedback

A rubric can make standards visible, but a learner may still not know which action will improve the next attempt.

Myth: AI gives unlimited feedback, so feedback scarcity is solved

AI can increase volume while also increasing noise, false rules and dependence. Verification and learner action remain essential.

Feedback quality audit

Take any comment and ask:

1. Does it identify a real performance issue?
2. Is the issue important enough to act on now?
3. Does it name or imply the cause?
4. Can the learner understand it?
5. Can the learner do something with it?
6. Does it preserve enough cognitive work?
7. Is there a fresh opportunity to apply it?
8. Will the repair be checked later?
9. Can support eventually be reduced?

If most answers are no, the comment may be marking rather than feedback.

Feedback density audit

For one page of student work, count:

• number of teacher marks;
• number of distinct underlying mechanisms;
• number of actions the learner is expected to take.

If thirty marks reduce to two mechanisms, present the two mechanisms.

Feedback language audit

Look for vague words:

• unclear;
• awkward;
• weak;
• develop;
• elaborate;
• better vocabulary;
• be careful;
• check grammar.

Each can be useful shorthand only if the learner already knows the operational meaning.

Feedback equity audit

Ask whether different learners receive:

• equally clear criteria;
• comparable opportunities to respond;
• feedback matched to current capability;
• consistent standards;
• language support where needed;
• fair distinction between dialect variation and target errors.

Feedback can unintentionally reproduce hidden expectations if criteria are not explicit.

Feedback evidence hierarchy

SETC uses a practical hierarchy for claims about feedback:

• systematic research or major evidence reviews for broad educational claims;
• official assessment criteria for examination-specific feedback;
• linguistic references for language claims;
• repeated learner performance for individual diagnosis;
• one-off anecdote as a hypothesis, not proof.

This mirrors the site-wide evidence architecture.

When feedback should become instruction

If the learner cannot understand the comment because the mechanism is absent, stop commenting and teach.

Example:

A Primary learner repeatedly receives use better conjunctions but does not understand clause relationships. The next step is a lesson on cause, contrast, addition and condition—not more margin comments.

When feedback should become practice

If the mechanism is understood but unstable, reduce explanation and increase retrieval/variation.

When feedback should become transfer

If the learner succeeds on familiar tasks, change context and cues.

When feedback should stop

If the learner can monitor and correct independently across fresh tasks, continued teacher prompting can create dependence.

Stop active feedback and monitor at longer intervals.

Feedback and expert performance

Advanced learners still need feedback, but it shifts toward finer distinctions:

• claim calibration;
• style and audience;
• evidence quality;
• efficiency;
• strategic choice;
• disciplinary convention;
• originality and synthesis.

At high levels, feedback often becomes conversation among informed participants rather than correction from authority.

Feedback and creativity

Creative work needs criteria without formula.

Feedback can address:

• reader effect;
• causal coherence;
• viewpoint;
• selection of detail;
• rhythm;
• voice;
• structural payoff.

Do not turn every creative choice into a universal rule.

Feedback and rhetoric

Persuasive writing feedback should distinguish reasoning from rhetoric.

A vivid example can be rhetorically strong and evidentially weak. Feedback should preserve the rhetorical success while repairing the evidence claim.

The Rhetoric & Persuasion owner provides the audience-effect layer.

Feedback and evidence literacy

If a learner uses a misleading chart, feedback should not stop at “wrong interpretation”. Ask which axis, denominator or comparison produced the mistake.

The Evidence, Data & Multimodal English owner provides the evidence-reading system.

Feedback and reference literacy

If teacher and learner disagree about usage, verify rather than escalate authority.

Use the Corpus, Dictionary & Reference Literacy owner to distinguish dictionary evidence, corpus patterns, style preference and institutional rules.

Feedback and digital literacy

Feedback on online writing should consider platform genre, permanence, audience collapse, multimodality and AI assistance.

The Digital & AI English owner handles the wider environment.

Feedback as institutional learning

A mature education system collects feedback about itself.

If a curriculum repeatedly creates the same misunderstanding, the curriculum needs repair.

If a rubric produces inconsistent interpretations, the rubric may need clarification.

If an AI feedback tool repeatedly flags legitimate Singapore English as incorrect, the tool needs auditing.

Feedback is not only something institutions give learners. It is also evidence institutions should use to improve their own systems.

The SETC Feedback Operating Manual

The operating manual below converts the whole page into a repeatable workflow for tutors, teachers and independent learners.

Step 1: establish the target

Before giving feedback, identify what successful performance requires in this task.

Questions:

• What is being assessed?
• Which criteria matter most?
• What does success look like at this learner’s current stage?
• Which conventions are fixed and which allow judgement?

Step 2: collect performance evidence

Use the learner’s actual work. Do not diagnose from reputation or previous marks alone.

Possible evidence:

• draft;
• oral explanation;
• timed response;
• annotation;
• planning notes;
• error pattern;
• confidence rating;
• time log.

Step 3: locate the visible loss

Where did performance first become observably weaker?

Step 4: trace backward

Ask what earlier decision could have produced that loss.

Continue until the cause is:

• plausible;
• supported by evidence;
• teachable;
• important enough to repair.

Step 5: identify the first repairable cause

Name it in one sentence.

The learner over-infers because plausible background knowledge is not being separated from passage evidence.

Step 6: define the replacement action

Before writing an inference, state the textual evidence and build only the smallest conclusion needed to explain it.

Step 7: choose feedback mode

Direct correction? Diagnostic question? Contrast example? Mini-lesson? Peer comparison? Automated hint?

Choose the least support that still produces useful learning.

Step 8: practise immediately

Use one or two fresh examples.

Step 9: record the rule

The learner states the replacement action in their own words.

Step 10: schedule delayed retrieval

Do not leave the repair inside one lesson.

Step 11: vary the surface

Change topic, wording, genre or context.

Step 12: interleave

Mix the target with neighbouring skills.

Step 13: restore authentic performance

Return to a whole comprehension, composition, oral or examination task.

Step 14: reduce support

Move down the prompt ladder.

Step 15: release

Archive the target when independent performance is sufficiently stable.

The learner version

A learner can use a shorter loop:

What was I trying to do? → What happened? → Where did it first go wrong? → What rule or decision should replace it? → Can I do it on a new example? → Can I still do it later without help?

The tutor version

Before leaving a lesson, the tutor should be able to answer:

• What did this learner improve today?
• What evidence supports that claim?
• What is still unstable?
• Which support should be removed next?
• When will the repair be retested?

The parent version

Parents do not need the full technical system. They can ask:

• What are you working on?
• What was the teacher’s main feedback?
• What will you do differently next time?
• Did you get a chance to try it again?

If the child cannot explain the feedback at all, that is useful information.

The three-student tutorial version

For Maren, Iona and Leonie:

1. all three attempt independently;
2. compare answers/processes;
3. each student identifies one difference;
4. tutor diagnoses individual first causes;
5. repair targets differ where needed;
6. fresh task tests each target;
7. group discusses why different feedback was appropriate;
8. next lesson begins with delayed retest.

The group learns not only English but how learning itself is diagnosed.

Feedback sentence bank: useful prompts, not canned comments

Reading

• What exactly does the question require?
• Which sentence is your evidence?
• What is the smallest inference the evidence supports?
• Which word limits the scope?
• What does this pronoun refer to?

Writing

• What is this paragraph’s job?
• What changes if this paragraph disappears?
• Which evidence supports the claim?
• Where does the reader need more explanation?
• Which sentence carries the main load?

Grammar

• What relationship is the grammar expressing?
• Which noun controls the agreement?
• What is the timeline?
• What does this pronoun point to?
• Where does one clause end and the next begin?

Vocabulary

• What exact distinction do you need?
• Which word naturally collocates here?
• What register is this?
• Which word-family form does the sentence require?
• Would an unfamiliar reader interpret this the same way?

Speaking

• What is your position in one sentence?
• What reason comes next?
• Which word needs clearer stress?
• Did you answer the question or your prepared point?
• How will you repair if the listener looks confused?

Listening

• Where did the signal disappear?
• Which word did you expect?
• Did you know the word in writing?
• Did your notes overload attention?
• How quickly did you re-enter after missing it?

Exam execution

• Which item consumed too much time?
• What was your stop rule?
• Which marks were knowledge losses and which were execution losses?
• What should trigger a skip-and-return?
• Where did checking produce a useful correction?

Frequently Asked Questions

What is feedback in learning?

Feedback is information that helps a learner compare current performance with a target and decide how to improve a future attempt. It is broader than correction and more actionable than a mark alone.

What makes feedback effective?

Effective feedback is grounded in a clear goal, accurate diagnosis and an action the learner can perform. It should be followed by an opportunity to use the feedback and later prove that the repair survives without support.

Should every mistake receive feedback?

No. Prioritise errors by leverage, severity, recurrence and position in the failure chain. Too much feedback can reduce attention to the most important target.

Should feedback be written or verbal?

Either can work. Choose the medium that best serves the learning job. Written feedback preserves a record; verbal feedback supports interaction and immediate diagnosis. The quality of the feedback principles matters more than assuming one medium is inherently superior.

Should feedback be immediate?

Sometimes. Immediate feedback can prevent repeated error and support new learning. Delayed feedback can protect independent performance and allow the teacher to observe the complete process. The retest should often include delay regardless.

Is marking the same as feedback?

No. Marking records or annotates performance. It becomes feedback when the learner can use the information to change a later performance.

Are grades feedback?

Grades provide performance information but usually do not identify how to improve. They are one piece of evidence, not a complete feedback system.

Should teachers correct grammar in compositions?

Yes when correction serves learning, but not necessarily every error at once. Group recurring patterns, repair high-leverage grammar relationships and require learners to detect some errors themselves.

Should teachers rewrite weak sentences?

A rewritten model can demonstrate a possibility, but if the teacher rewrites most of the text the learner loses authorship and practice. Use modelling strategically, then return the decision to the learner.

Do model answers count as feedback?

They can support feedback if the learner compares decisions and understands why the model works. Simply reading or copying a model is not sufficient.

How many feedback targets should a learner have?

Usually a small active set. One primary target and one secondary target can be more usable than ten simultaneous goals. Advanced learners may manage more.

How long should a feedback target remain active?

Until evidence shows it survives fresh tasks, delay, reduced support and realistic conditions. Some targets stabilise quickly; others require repeated spiralling.

What if a student keeps making the same mistake?

Do not repeat the same comment indefinitely. Re-diagnose. The learner may not understand the feedback, may lack a prerequisite, may not retrieve the rule under load or may not recognise when to apply it.

What if feedback makes performance worse temporarily?

This can happen when learners are restructuring a process or when practice becomes more difficult and mixed. Check whether the underlying decision is improving and whether the temporary decline is caused by higher task demand.

Can peers give useful feedback?

Yes, especially when tasks ask peers to report observable reader experience or apply clear criteria. Peer feedback requires training and should not be treated as infallible technical authority.

Can students give themselves feedback?

Yes, and that is a major goal. Self-feedback improves as learners internalise criteria, detect patterns and calibrate confidence. Expert feedback remains useful for blind spots and advanced judgement.

Can AI give feedback?

AI can generate rapid feedback and practice, but its claims should be verified where accuracy matters. The educational test is whether the learner can later perform independently without the tool.

How do I know whether feedback transferred?

Change the surface, delay the task, mix the skill with others, reduce support and restore real constraints. If quality survives, transfer is stronger.

Should feedback focus on strengths too?

Yes. Learners need to know which decisions produced successful performance so they can repeat them. Strength feedback should be specific, not generic praise.

What is feedforward?

Feedforward emphasises what the learner should do in future performance rather than only what was wrong in the past. SETC includes feedforward inside feedback: every major comment should imply a next action.

What is the difference between feedback and coaching?

Feedback provides information about performance. Coaching can include goal setting, questioning, practice design, motivation and longer-term development. Feedback is one component of coaching.

What is the difference between feedback and assessment?

Assessment gathers and interprets evidence about performance. Feedback uses evidence to support improvement. Formative assessment and feedback are therefore closely connected but not identical.

External evidence routes

• Education Endowment Foundation — Teacher Feedback to Improve Pupil Learning
• EEF Teaching and Learning Toolkit — Feedback
• Institute of Education Sciences — Organizing Instruction and Study to Improve Student Learning
• National Academies — How People Learn II

These sources support the broad learning principles used here. SETC’s contribution is to translate those principles into a domain-specific English diagnostic and transfer system.

Continue through SETC

• English Atlas — locate feedback inside the complete English system.
• How English Works — understand the language mechanisms being repaired.
• How English Learning Transfers — follow the repair into independent performance.
• Three-Student English Tutorials — see the teaching environment built around visibility and diagnostic feedback.
• English Learning Library — move into the relevant skill or stage.
• Corpus, Dictionary & Reference Literacy — verify language feedback and usage claims.
• Digital & AI English — use automated feedback without surrendering verification or independence.

The SETC Feedback Standard

Feedback is successful when the learner does not merely understand the comment but changes a later decision.

A mature feedback system can answer:

• What was the intended performance?
• What evidence did the learner produce?
• Where was the first repairable cause?
• What replacement action was taught?
• How was it practised?
• When was it retested?
• Did it survive delay?
• Did it survive variation?
• Did it survive reduced support?
• Did it survive realistic performance pressure?

If the answer reaches the final question, feedback has become capability.

Diagnose precisely. Repair the mechanism. Retest on something new. Delay the proof. Reduce support. Release the learner.

Advanced feedback: ambiguity, judgement, creativity and expert performance

Not every English task has one correct answer. As learners advance, feedback increasingly deals with judgement: which interpretation is better supported, which structure is more effective, which rhetorical choice is more appropriate, which evidence deserves more weight and which style best serves the audience.

Advanced feedback therefore cannot always say “right” or “wrong”. It needs criteria, evidence and comparison.

Feedback under ambiguity

Suppose two comprehension interpretations are both plausible.

The teacher should not resolve the ambiguity by authority alone. Ask:

• Which interpretation explains more of the text?
• Which requires fewer unsupported assumptions?
• Which fits tone and context?
• What evidence would weaken each interpretation?
• How certain should the answer sound?

The learner develops proportionate judgement.

Feedback on literary interpretation

Literature feedback should distinguish:

• textual evidence;
• interpretation;
• alternative interpretation;
• quality of explanation;
• historical/contextual knowledge;
• unsupported projection.

Weak feedback:

This interpretation is wrong.

Stronger feedback:

Your reading of the image as threatening is plausible, but the paragraph needs to connect that effect to the repeated darkness imagery and the character’s earlier avoidance. Without those links, the interpretation looks asserted rather than demonstrated.

Feedback on style

Style feedback should identify effect and fit.

This short sentence creates a useful pause after the long descriptive paragraph.

The metaphor is vivid, but it changes the tone from restrained to melodramatic. Decide whether that is the effect you want.

There may be more than one defensible choice.

Feedback on voice

Do not turn every strong student into the same polished institutional voice.

Ask:

• Is the voice coherent?
• Is it appropriate to genre?
• Does it obscure meaning?
• Does it rely on imitation?
• Can the learner vary it deliberately?

The Voice owner provides the deeper language system.

Feedback on rhetoric

A persuasive sentence may be effective and ethically weak.

Example:

Only a reckless parent would ignore this programme.

The line may pressure action through identity threat.

Feedback can say:

The sentence creates urgency, but it attacks the parent rather than demonstrating programme value. Build urgency from the evidence or consequence instead.

This teaches ethical persuasion.

Feedback on evidence

Advanced learners need feedback on evidence quality, not merely presence.

You have a source, but it studies university students and your claim is about Primary learners. Either narrow the claim or find population-matched evidence.

Feedback on uncertainty

Strong writing calibrates certainty.

Your paragraph moves from “associated with” in the source to “causes” in your conclusion. Restore the evidential boundary.

Feedback on synthesis

A literature paragraph can cite five sources and still lack synthesis.

Feedback:

Right now the paragraph is organised by authors. Reorganise it around the relationship: which sources agree, which differ, and why?

Feedback on originality

Originality does not require inventing facts. It can appear in:

• question framing;
• source connection;
• analogy;
• application;
• counterexample;
• new synthesis;
• style.

Feedback should protect originality while demanding evidence.

Feedback on creative writing

Creative writing feedback needs a model of reader effect and craft rather than a checklist of mandatory devices.

Character

The character’s final choice surprises me because the earlier scenes do not show enough conflict. Add one earlier moment that makes both choices possible.

Plot

The ending resolves the external problem, but the internal decision has not changed. Decide what the character has learned or refused to learn.

Description

You use five sensory details, but they all describe the room generally. Select details through the frightened child’s viewpoint.

Dialogue

The dialogue tells the reader information both characters already know. Give each speaker a real conversational objective.

Pacing

The argument occupies three paragraphs and the consequence one sentence. The story’s emotional weight is reversed.

Voice

The informal voice is coherent until the final paragraph suddenly becomes abstract and formal. Either motivate the shift or preserve the established register.

Creative feedback should preserve ownership

The tutor should not rewrite the scene into the tutor’s story.

Ask the writer what effect they want, then evaluate whether the current choices produce it.

Feedback on advanced oral communication

Advanced speaking feedback includes:

• argument structure;
• interaction;
• response to challenge;
• evidence calibration;
• audience adaptation;
• delivery;
• timing;
• recovery.

Seminar feedback

You contributed relevant evidence, but you entered as a new monologue rather than building on the previous speaker. Begin by naming the relationship to the prior point.

Presentation feedback

Your slide is accurate, but the audience must read six lines while you explain a different relationship. Reduce the slide to the evidence they need visible.

Question-and-answer feedback

You answered quickly but changed the question. Pause, paraphrase the question and then respond only to the part your evidence supports.

Feedback across cultures and varieties

Feedback can become unfair when one cultural or dialect norm is treated as universal without explanation.

Accent

Target intelligibility, not prestige.

The final consonant in these technical terms is disappearing for several unfamiliar listeners. Let’s practise that contrast.

This is more useful than sound more native.

Register

A local colloquial form can be legitimate and still be wrong for a formal task.

This expression works naturally in Singapore informal speech. The exam requires formal Standard English, so use this equivalent here.

The learner gains repertoire without shame.

Pragmatics

Directness can be interpreted differently across cultures. Feedback should describe audience effect rather than moralising one convention.

In this formal request, the imperative may sound stronger than you intend. Add a request form that preserves authority but reduces unnecessary directness.

Multilingual transfer

When another language influences English, contrast the systems explicitly where helpful.

But avoid assuming every error is transfer. Collect evidence first.

Feedback on research and academic writing

The Academic English owner provides the full system. Feedback in this domain should often prioritise:

• research question;
• source representation;
• synthesis;
• argument architecture;
• method/evidence boundaries;
• citation;
• disciplinary register;
• reader navigation.

Research question feedback

Your question asks whether AI “works”, but the sources measure different outcomes. Define the outcome before expanding the literature search.

Literature review feedback

You have enough sources. The next improvement comes from organising relationships, not adding more papers.

Method feedback

The method is described accurately but the reader cannot see why this design answers the research question. Add the methodological rationale.

Discussion feedback

The results are from one population, but the discussion generalises to all learners. Restore the population boundary.

Feedback on professional communication

Professional/career English is federated outward to eduKateSG, but the language mechanisms remain relevant.

Feedback may target:

• decision clarity;
• actionability;
• risk communication;
• audience knowledge;
• plain English;
• record accuracy.

SETC supplies the mechanism; the adult/career estate owns the application.

Longitudinal case: Maren’s writing feedback across four months

Month 1: symptom

Maren’s essays are polished but vague.

Diagnosis: paragraphs do not have clear claims.

Repair: write one claim sentence before evidence selection.

Month 2: improvement

Claims are clearer. Evidence remains loosely connected.

New target: warrant—explain why evidence supports the claim.

Month 3: transfer

Maren applies claim–evidence–warrant in argumentative writing but not in source synthesis.

Target shifts from isolated mechanism to cross-genre transfer.

Month 4: release

Maren independently reverse-outlines an unfamiliar essay, detects one unsupported paragraph and revises before feedback.

The tutor removes claim architecture from the active target list.

Feedback evolved with performance rather than repeating “develop your ideas”.

Longitudinal case: Iona’s comprehension feedback

Month 1

Iona reads accurately but over-infers.

Repair: evidence → smallest supported inference.

Month 2

Inference accuracy improves, but answer scope expands under difficult questions.

Repair: identify command word and scope limiter.

Month 3

Scope is stable untimed but collapses when the passage is long.

Repair: time/attention recovery rather than inference reteaching.

Month 4

Iona monitors uncertainty, flags one answer and returns later without losing the rest of the paper.

The feedback target has moved from comprehension knowledge to examination regulation.

Longitudinal case: Leonie’s timing feedback

Month 1

Leonie runs out of time because she overinvests in early questions.

Repair: section checkpoints and skip-return rule.

Month 2

Timing improves, but skipped items are forgotten.

Repair: visible return queue.

Month 3

Return queue works, but difficult questions trigger confidence collapse.

Repair: containment routine and confidence reset.

Month 4

Leonie finishes consistently and uses final minutes for high-value checking.

Feedback has changed the control system, not just speed.

Feedback under plateau

If progress stalls, do not automatically increase effort.

Audit:

• Is the target correctly diagnosed?
• Is prerequisite knowledge missing?
• Is practice too similar?
• Is retrieval delayed?
• Is support being removed?
• Is task difficulty too high?
• Is the learner fatigued?
• Is the metric sensitive enough to show change?

A plateau can be a design problem.

Feedback under regression

Skills can appear to regress when complexity increases.

A student who wrote accurate simple sentences may make more grammar errors while attempting sophisticated syntax. The teacher should distinguish productive stretch from genuine loss.

Feedback:

Your sentence complexity increased, and reference accuracy dropped. Keep the complexity; we’ll repair reference inside it rather than returning to only simple sentences.

Feedback under fatigue

Late-task errors can reveal endurance problems rather than knowledge gaps.

Compare early and late sections. If accuracy deteriorates systematically, train stamina and recovery.

Feedback under success

Strong performance still provides information.

Ask what made the success possible:

• better plan;
• stronger reading model;
• timing choice;
• specific retrieval;
• more appropriate difficulty;
• effective checking.

Success should also be diagnosed so it can be repeated.

Feedback and identity statements

Avoid turning temporary patterns into identities:

You are weak at comprehension.

Prefer:

Your literal understanding is strong; inference scope is currently unstable when questions are ambiguous.

The second statement is more accurate and more teachable.

Feedback and responsibility

As learners mature, they should carry increasing responsibility for acting on feedback.

Responsibility does not mean blame. It means agency:

• record the target;
• complete the retest;
• ask for clarification;
• verify disputed language;
• notice recurrence;
• bring evidence of improvement.

Feedback and teacher humility

Teachers can be wrong. A student may challenge a correction with a dictionary, corpus or official criterion.

A strong teacher checks the evidence.

This models intellectual honesty and strengthens trust.

Feedback disagreement protocol

When learner and teacher disagree:

1. state the disputed claim precisely;
2. identify the target context;
3. consult the relevant authority;
4. compare evidence;
5. revise the judgement if necessary;
6. record the principle, not the victory.

The expert feedback standard

At advanced levels, feedback becomes less about fixing obvious errors and more about sharpening judgement.

The learner should increasingly be able to ask:

What would make this decision more defensible, more precise, more efficient or more appropriate to the audience?

That is feedback operating at the level of expertise.

Feedback consistency: ten rules for coherent teaching

Feedback becomes confusing when different lessons use contradictory principles without explaining why. A learner may be told to correct every error immediately in one task and to delay correction in another. Coherent teaching makes the performance target and reason for the feedback choice explicit.

Rule 1: every major feedback claim must name the performance target

Do not write “feedback improves learning” without specifying what type of learning or performance is being discussed.

Is the target:

• factual recall;
• grammar accuracy;
• reading inference;
• essay structure;
• spoken fluency;
• timed examination execution;
• independent transfer?

Different targets require different evidence and feedback designs.

Rule 2: separate correction from learning evidence

A corrected product is not proof of durable learning. SETC pages should say what later evidence would demonstrate transfer.

Rule 3: feedback should imply a next action

Major instructional comments should make the replacement behaviour visible.

Rule 4: diagnosis comes before volume

Do not prescribe more practice until the weak mechanism is known.

Rule 5: do not let feedback become permanent scaffolding

Every scaffold should have a fading plan.

Rule 6: feedback should preserve ownership

Teacher and AI assistance should not produce finished work the learner cannot reproduce independently.

Rule 7: local language variation must be distinguished from target error

Correct for the context accurately. Do not call every dialect feature broken English.

Rule 8: feedback claims need evidence proportional to strength

Use major educational evidence for broad claims, official criteria for exam rules and linguistic references for language usage.

Rule 9: feedback systems should be sustainable

A method requiring impossible teacher marking loads will not survive implementation. Prioritisation and whole-class feedback can preserve quality.

Rule 10: the final goal is learner self-regulation

A feedback system that makes the learner permanently dependent has failed its deepest purpose.

How feedback connects to the rest of English learning

The general feedback mechanism is:

diagnose → repair → respond → retest → delay → transfer → fade → release.

Other English systems apply that mechanism:

• How English Learning Transfers owns the movement from supported learning to independent performance.
• Three-Student English Tutorials owns the teaching environment that makes learner decisions visible.
• Reasoning, Argument & Knowledge owns reasoning quality.
• Evidence, Data & Multimodal English owns evidence interpretation and verification.
• Reference Literacy owns language-evidence verification.
• Digital & AI English owns tool-assisted feedback and authorship/verification boundaries.
• Skill owners—Reading, Writing, Grammar, Vocabulary, Listening and Speaking—apply feedback inside their domains.

This keeps the feedback model consistent while allowing reading, writing, speaking, examinations and other domains to apply it differently.

Stage mastery map: Primary 1–2

Feedback at the beginning of formal English learning should help children notice concrete relationships.

High-value targets include:

• sound–print mapping;
• word recognition;
• sentence completeness;
• basic punctuation;
• literal comprehension;
• oral explanation;
• simple revision.

Feedback language should be short and observable.

Point to the word that tells you who did the action.

Read the sentence again and stop where the full stop tells you.

Your story has the event. Add what happened because of it.

Stage mastery map: Primary 3–4

Learners can begin managing larger relationships:

• main idea and supporting detail;
• reference;
• basic inference;
• paragraph structure;
• vocabulary in context;
• grammar editing;
• oral development.

Feedback should increasingly ask the learner to locate evidence before receiving the answer.

Stage mastery map: Primary 5–6

Feedback should prepare learners for greater independence and examination integration:

• question interpretation;
• answer scope;
• evidence-based inference;
• composition planning;
• paragraph development;
• editing routines;
• timing;
• oral interaction;
• listening recovery.

By Primary 6, the learner should begin carrying a small active feedback log and using self-check prompts.

Stage mastery map: Secondary 1–2

The transition requires feedback on denser text, less literal questions and wider writing decisions.

Targets include:

• paragraph function;
• summary;
• source viewpoint;
• situational audience/register;
• longer argument;
• editing under increased load;
• independent study routines.

Stage mastery map: Secondary 3–4

Feedback should become strongly performance-oriented:

• unseen text control;
• evidence and inference;
• argument and evaluation;
• source transformation;
• genre transfer;
• timing and recovery;
• score stability;
• independent revision.

The teacher should increasingly remove skill labels from practice so learners must select the method.

Stage mastery map: JC / Advanced English / GP

Feedback moves toward:

• definition control;
• knowledge selection;
• source synthesis;
• claim calibration;
• evidence quality;
• counterargument;
• rhetoric and audience;
• academic register;
• time allocation under high cognitive load.

Students should be able to debate feedback on evidential grounds.

Stage mastery map: university / Academic English

Feedback increasingly concerns:

• research question;
• source representation;
• literature relationships;
• disciplinary genre;
• methodological limits;
• citation and authorship;
• academic voice;
• research independence.

The teacher or supervisor should not become a permanent co-author of the learner’s work.

One month of feedback: complete worked cycle

Consider a Secondary 3 learner whose comprehension marks fluctuate.

Week 1, Day 1: baseline task

The learner completes an unseen passage under ordinary class time.

Lost marks:

• 2 marks literal detail;
• 4 marks inference;
• 2 marks summary scope;
• 1 mark expression.

A naive response would be “practise inference and summary”.

The tutor investigates.

Week 1, diagnosis

Literal errors occur because the learner searches by repeated words and misses paraphrases. Inference errors occur because the learner uses background knowledge rather than passage evidence. Summary scope errors occur because the learner cannot distinguish examples from central points.

There are three mechanisms, but the tutor prioritises inference because it costs most marks and influences summary selection.

Primary target:

Evidence-before-inference.

Week 1, repair

The learner practises five short items:

1. underline evidence;
2. state literal meaning;
3. state smallest supported inference;
4. identify one plausible but unsupported extra inference.

Prompt level: 4. Teacher names the mechanism and guides each item.

Week 1, immediate retest

New short passage. Four of five inferences are controlled.

No claim of mastery yet.

Week 1, learner rule

The learner records:

Before I infer, I must point to the evidence and avoid adding a cause the text does not need.

Week 2: retrieval

No notes shown initially.

Question:

What is your inference rule?

Learner retrieves it with one hint.

Prompt level falls to 3.

Week 2: mixed practice

Ten questions include literal, reference, vocabulary and inference items. Learner must identify which need inference.

Accuracy: 80%.

Two errors are not over-inference; they are pronoun-reference failures.

The tutor does not misdiagnose them as inference relapse.

Week 2: secondary target appears

Reference becomes secondary target. Tutor uses two short diagnostic items but keeps inference primary.

Week 3: full passage

Longer unseen text under time.

Inference accuracy remains high early in the passage but declines late.

Interview reveals fatigue and rushing, not lost conceptual understanding.

Feedback shifts to performance regulation:

At the halfway point, pause for ten seconds and reset the evidence rule before the final section.

Week 3: prompt reduction

No teacher reminders during the task. Learner uses a self-mark on the page margin at the halfway point.

Week 4: delayed transfer

New genre: informational rather than narrative.

No skill label. No self-mark provided.

Learner applies the evidence-before-inference rule independently on four of five relevant questions.

Week 4: whole-paper simulation

Under realistic timing, inference marks are stable. Summary improves slightly because the learner is better at distinguishing textual evidence from elaborative background knowledge.

Week 4: release decision

Inference leaves the primary-target list and moves to monitoring.

Reference becomes primary target.

The learner’s comprehension programme evolves rather than repeating the same worksheet category.

What the month-long case demonstrates

Good feedback changed several things:

• the diagnosis;
• the practice design;
• the learner’s rule;
• the prompt level;
• the delayed evidence;
• the interpretation of relapse;
• the next target.

The final score mattered, but the feedback system explained why it changed.

One feedback mechanism across different learning contexts

The same feedback sequence can operate in very different settings without becoming a different theory each time.

• Recovery work uses it to find and repair the first unstable cause.
• School-stage learning uses it to move from guided practice to independent control.
• Examination preparation uses it to stabilise decisions under time and pressure.
• Reading, writing, vocabulary, grammar, listening and speaking use it inside their own mechanisms.
• Adult and professional learning use it to improve communication while preserving learner ownership.

One mechanism, many applications.

When a specialised feedback guide is useful

A specialised guide adds value when it explains a distinct feedback problem inside a skill, stage, examination or learner state. It should add diagnosis, practice or retesting that the general mechanism cannot provide on its own.

A guide adds little when it simply repeats the same general feedback theory under a new title. The learner should be able to see what is specific about the new task.

When the feedback system itself needs review

The core feedback loop is stable, but implementation should be reconsidered when the evidence, assessment conditions or learning tools change. A new examination may change what must be retested; a new AI tool may change which work the learner is actually performing; new research may refine how feedback timing or delivery should be interpreted.

The stable question remains: does this feedback help the learner make a better decision later with less external support?

Final mastery test

Give the learner a new task with no active feedback target displayed.

After completion, ask them to:

1. identify one strength;
2. identify one failure risk;
3. locate evidence;
4. name the first repairable cause;
5. make one revision;
6. state the rule;
7. design a fresh retest.

If the learner can do this accurately, the feedback system has begun to become self-regulation.

The final principle

Feedback is not the teacher’s commentary on the learner.

It is a temporary external control system whose purpose is to help the learner build an internal one.

The strongest feedback eventually makes itself less necessary.

Feedback Diagnostic Atlas: from visible symptom to first useful test

The table below is deliberately diagnostic rather than prescriptive. A symptom does not prove a cause. It suggests where to look first.

Quick route: when you have only five minutes

If time is limited, use this compressed feedback protocol:

1. Choose one high-leverage error.
2. Ask the learner why it happened.
3. Confirm or refine the cause.
4. State one replacement action.
5. Give one fresh example.
6. Schedule one delayed check.

Six steps can outperform twenty unexplained corrections.

Quick route: when the learner is upset

1. Stabilise the task: identify what actually happened.
2. Separate score from identity.
3. Name one thing that worked.
4. Name one first repairable cause.
5. Give one achievable next action.
6. Retest soon enough that improvement becomes visible.

Do not hide the problem, but make it solvable.

Quick route: when the exam is close

Prioritise by mark return and reliability:

• high-frequency costly errors;
• timing collapses;
• question interpretation;
• checking routine;
• recovery;
• stable language errors with quick repair.

Avoid rebuilding the entire English system days before the exam.

Quick route: when the learner is far ahead

Feedback should move from correctness toward:

• precision;
• efficiency;
• style;
• audience;
• evidence quality;
• originality;
• transfer into unfamiliar domains.

Quick route: when you are unsure what the error means

Do not guess. Collect one more piece of evidence:

• oral explanation;
• similar item;
• untimed version;
• prompt-free version;
• reference lookup;
• think-aloud process.

Precision in diagnosis is worth more than speed in commenting.

The four questions that should survive every feedback system

Whether the learner is seven or seventeen, whether the task is reading or GP, whether feedback is human or AI-assisted, the system should preserve four questions:

1. What were you trying to do?
2. What actually happened?
3. What will you do differently next time?
4. How will we know the change transferred?

Everything else is implementation.

Why this owner matters to the ten-year SETC build

A large English website can publish thousands of lessons and still fail educationally if those lessons do not contain a theory of change. Feedback is one of the mechanisms that connects information to improvement.

This owner makes that mechanism explicit. It prevents future SETC pages from treating marking, correction, praise, grading, AI suggestions and tutor commentary as interchangeable. It gives every stage and skill a common control loop and lets the wider eduKate federation reuse one coherent learning model.

Most importantly, it preserves SETC’s central educational principle: when a learner cannot yet do something reliably, the useful response is investigation, not blame. Find the state. Find the first repairable cause. Change the mechanism. Gather evidence.

The release sentence

The strongest evidence that feedback has worked is not:

The student corrected the page.

It is:

The student met a fresh problem, recognised the relevant risk, selected a better decision and no longer needed the person who first pointed it out.

The feedback evidence dossier: one reusable record for difficult cases

When a learner’s difficulty persists, a short structured dossier prevents feedback from becoming a sequence of guesses. The dossier is not bureaucracy. It is a way to preserve the evidence that led to a teaching decision.

1. Performance target

What was the learner expected to do?

2. Actual evidence

Quote or describe the answer, timing pattern, oral response, draft feature or score pattern.

3. Visible symptom

What failed at the surface?

4. Candidate causes

List two or three plausible causes before selecting one.

5. Diagnostic test

What extra evidence distinguishes the candidates?

6. First repairable cause

State the current best diagnosis in one sentence.

7. Replacement action

What will the learner do differently?

8. Immediate retest

What fresh item will test basic understanding?

9. Delayed retest

When will the same mechanism reappear after delay?

10. Transfer condition

What surface feature, genre, topic or constraint will change?

11. Prompt level

How much support was required?

12. Outcome

Stable, improving, unstable, regressed or released?

Worked dossier: persistent inference loss

Target: make text-supported inferences.

Evidence: learner identifies relevant sentence but adds motives not stated or required.

Candidate causes: weak comprehension; background knowledge interference; poor certainty calibration.

Diagnostic test: learner explains literal sentence accurately but cannot separate “possible” from “supported”.

First repairable cause: certainty/evidence boundary.

Replacement action: state evidence, then write the smallest conclusion that explains it.

Immediate retest: two short unseen passages.

Delayed retest: mixed comprehension one week later.

Transfer: move from narrative to informational text.

Prompt level: initially teacher asks for evidence; later learner self-prompts.

Release evidence: four of five correct in mixed unseen task after delay, with accurate self-explanation.

Worked dossier: sophisticated vocabulary that weakens writing

Target: precise formal vocabulary.

Evidence: learner inserts rare words whose collocations or connotations do not fit.

Candidate causes: meaning weakness; thesaurus dependence; belief that difficulty equals quality.

Diagnostic test: learner can define words approximately but cannot produce natural collocations.

First repairable cause: vocabulary-selection criterion.

Replacement action: choose by exact meaning + collocation + register, verified where needed.

Immediate retest: select among three candidate words and justify.

Transfer: use precise vocabulary in a new essay without thesaurus prompts.

Seven jobs that feedback can do

Across different tasks, useful feedback tends to perform one or more of seven jobs:

• Diagnosis — identifies the failure point;
• Repair — teaches the replacement mechanism;
• Retest — checks whether the repair works again;
• Transfer — moves the repair into a new context;
• Fade — reduces external support;
• Recovery — restores control after relapse, difficulty or pressure;
• Release — establishes independent performance.

A single comment does not need to do all seven jobs. Across the learning cycle, however, feedback should move beyond correction and eventually reduce the learner’s need for external control.

When the feedback plan should change

Keep a feedback method only while it is producing useful future performance. Reconsider the plan when:

• the same error persists despite repeated correction;
• the learner can perform the target in isolation but not inside a harder task;
• support has become so heavy that the learner waits for feedback before acting;
• a new examination format changes the performance condition;
• a digital or AI tool changes what the learner is doing independently;
• the learner has outgrown the current target and needs a higher-level challenge.

Stable principles do not need cosmetic rewriting. The method changes when the learner, task or evidence changes.

The complete feedback sequence

Across English learning, feedback should never be reduced to “tell the learner what was wrong”.

The system standard is:

observe → diagnose → prioritise → repair → retrieve → retest → vary → transfer → fade → release.

That sequence keeps feedback focused on future performance rather than on the corrected page alone.

Relapse and reactivation: when an old error returns

A released feedback target can reappear months later. That does not automatically mean the original learning failed. Performance changes with complexity, fatigue, unfamiliar genres, time pressure and competing demands. An old error can return because the learner is operating at a higher level of difficulty.

SETC treats relapse as a new data point, not a reason to restart the entire programme.

First ask whether the old mechanism is truly gone

Give one low-load diagnostic item. If the learner performs correctly there but fails inside a complex task, the issue is integration or load, not basic knowledge.

Reactivate narrowly

Bring the old target back into the active register temporarily:

• retrieve the rule;
• apply it to one fresh example;
• reconnect it to the new high-load context;
• reduce support quickly if control returns.

Do not automatically rebuild from the beginning.

Distinguish relapse from development

A learner attempting more complex syntax may temporarily make more agreement or reference errors. A writer attempting stronger evaluation may become less fluent while learning to qualify evidence precisely. A speaker entering spontaneous discussion may hesitate more than in prepared speech.

Temporary instability can be the cost of expanding the system.

Feedback should protect the new ambition while repairing the specific breakdown.

Reactivation threshold

One isolated recurrence may only need a reminder. Repeated recurrence across contexts may justify active retraining. Use frequency, severity and pattern before deciding.

Final durability standard

The strongest learner is not someone who never makes an old error again. It is someone who can notice recurrence, recover the relevant rule, repair the performance and continue without waiting for external rescue.

That ability to reactivate prior learning is itself part of independence.

Minimum evidence before a feedback target is released

A feedback target should not be marked complete because the learner produced one correct answer immediately after explanation. Before release, SETC expects enough evidence to answer five questions.

1. Can the learner perform the mechanism on a fresh example?
2. Can the learner recognise when the mechanism is needed without being told?
3. Can the learner retrieve it after a delay?
4. Can the learner use it while other demands compete for attention?
5. Can the learner recover independently if the old error reappears?

The amount of evidence required should match the importance of the skill. A minor spelling convention may need only a few successful checks. A high-cost examination habit, reading inference system or writing-structure repair needs broader evidence across tasks and time.

Release is therefore a judgement based on repeated performance, not a ceremonial end to a worksheet sequence.

This final rule protects the whole feedback architecture from premature success. The purpose of feedback is not to make the current page look corrected. It is to make the future learner more capable.