Wait, what? One wrong decision at the top of a Science question can create four wrong answers underneath it.

A student misreads the graph scale. That produces the wrong comparison. The wrong comparison makes them choose the wrong concept. The wrong concept creates the wrong explanation. Then they “correct” the final sentence without fixing the graph reading—and the same error returns in the next paper.

This is error propagation.

This preserved Hougang Primary 6 Science URL now owns one precise job: how an early reasoning error propagates downstream, and how to repair the root rather than polishing the symptoms. The old duplicated 2019 tuition advertisement, stale locality claims, grade promises and unrelated image stack have been removed.

This page is deliberately different from the Hougang PSLE post-paper audit, which maps where marks leaked across a paper. Here the focus is inside one reasoning chain:

Which mistake happened first—and which later mistakes exist only because that first mistake was never repaired?

The first wrong move matters more than the last wrong sentence

Suppose a student’s final explanation is scientifically wrong.

There are many possible root causes:

• they read the wrong value;
• they compared final value instead of amount of change;
• they missed a time interval;
• they chose the wrong concept;
• they reversed cause and effect;
• they assumed an unstated condition;
• they ignored a control variable;
• they misunderstood the command word.

If the tutor corrects only the final sentence, the root mechanism remains.

Repair should therefore begin with:

Where did the reasoning first become invalid?

Root error versus downstream symptom

A root error generates later mistakes. A downstream symptom is a consequence of the earlier error.

Root error	Possible downstream symptoms
Misread graph scale	Wrong value → wrong comparison → wrong conclusion
Wrong concept selected	Wrong mechanism → wrong terminology → wrong prediction
Unstated assumption added	False cause → overconfident conclusion → wrong method evaluation
Command word misread	Correct Science in the wrong answer form
Cause/effect reversed	Whole causal chain runs backward

One root repair can remove several downstream symptoms at once.

Error propagation through a graph question

Imagine a graph where each small vertical interval represents 2 units.

The student reads each interval as 1 unit.

1. Root: axis scale misread.
2. Result: data values copied incorrectly.
3. Comparison: difference between setups calculated incorrectly.
4. Interpretation: wrong setup appears to have changed more.
5. Explanation: student chooses a mechanism to explain the false comparison.

Correcting Step 5 alone is useless. The graph must be reread from Step 1.

Error propagation through final value versus change

Suppose the question asks which plant grew more.

The student compares only final heights.

1. Wrong comparison target: final value instead of change.
2. Wrong winner selected.
3. Wrong evidence sentence written.
4. Wrong causal explanation built around the selected winner.
5. Checking reinforces the wrong answer because the final height still “looks bigger”.

The correction should not begin with “Plant A is wrong.” It should begin with:

What quantity does “grew more” require?

Repair the task definition and the later answer may correct itself.

Error propagation through concept selection

A student sees the word “water” and retrieves the water-cycle chapter even though the question is about transport through a plant.

That early cue-driven selection can propagate:

• wrong concept family;
• wrong scientific verbs;
• wrong diagram interpretation;
• wrong explanation;
• wrong checking criteria.

The answer may look coherent because every later sentence is consistent with the initial wrong model.

This is why internal consistency is not enough. The first concept must match the evidence and task.

Error propagation through an unstated assumption

Suppose the question does not say that the two plants received equal amounts of water, but the student assumes they did.

Then:

• light becomes the only imagined difference;
• light is treated as the cause;
• the causal mechanism is written confidently;
• alternative explanations are ignored;
• the conclusion becomes stronger than the evidence allows.

The root is not “wrong explanation”. It is “unsupported equality assumption”.

Assumption control prevents entire branches of false reasoning.

Error propagation through reversed causality

If a student reverses A→B into B→A, every downstream step may inherit the reversal.

Example structure:

Correct: lower food availability → fewer surviving consumers.

Reversed: fewer consumers → lower food availability.

The reversed chain may then produce wrong predictions about later populations.

The repair is to justify every arrow before propagating the chain.

Error propagation through command-word misreading

A student reads “compare” as “describe”.

They may know the data perfectly and still:

• write two separate descriptions;
• fail to state the relationship;
• lose the mark;
• later conclude they “do not know the topic”.

The downstream symptom can become a false self-diagnosis.

Correct the task-decoding layer before reteaching content the child already knows.

Error propagation through missing data

Students often fill missing information unconsciously.

If starting values are missing, they may assume they were equal. If time is missing, they may assume equal duration. If one control is unstated, they may assume it was held constant.

Each invented value creates a new branch of reasoning that appears valid only because the gap was silently filled.

A completeness check therefore acts as an upstream error firewall.

Local errors versus global errors

Not every mistake propagates equally.

• Local error: affects one small part of the answer.
• Global error: changes the model, comparison or assumption used by many later steps.

Example of a local error:

Correct reasoning, but unit omitted in the final numerical statement.

Example of a global error:

Wrong concept selected at the start, so every later explanation uses the wrong mechanism.

Repair priority should consider propagation reach, not only visible mark loss.

The dependency tree

Map the answer as dependencies:

task reading → evidence extraction → comparison → concept selection → mechanism → conclusion → wording

Each later stage depends partly on earlier stages.

If the conclusion is wrong, move upward through the tree:

• Was the mechanism wrong?
• If yes, was the concept wrong?
• If yes, was the evidence selected incorrectly?
• If yes, was the task misunderstood?

Stop at the earliest invalid node. That is the repair target.

The “first divergence” method

Compare the learner’s reasoning with a valid reasoning path.

Find the first point where they diverge.

Example:

Stage	Valid route	Learner route
Task	Compare amount of change	Compare amount of change
Data	Use start and final values	Use start and final values
Calculation	A changes by 8	A changes by 8
Calculation	B changes by 5	B changes by 5
Decision	A changed more	B changed more

The first divergence occurs at the decision stage, not at data extraction.

Different divergence points require different repairs.

Do not overdiagnose from the final answer

Two students can write the same wrong final answer for different reasons.

• Student A misread the graph.
• Student B read the graph correctly but chose the wrong concept.
• Student C knew the concept but reversed cause and effect.
• Student D had the correct reasoning and changed the answer during checking.

The same visible wrong answer hides four different roots.

Correction must inspect the route, not only the endpoint.

Correct answer, wrong route

Error propagation also matters when the final answer is accidentally correct.

A student may:

• misread a quantity;
• apply the wrong concept;
• make a second compensating error;
• land on the correct option.

That correct mark is fragile.

During diagnosis, ask the learner to explain high-confidence correct answers too. A correct endpoint does not guarantee a sound path.

Error propagation across multi-part questions

Some structured questions build later parts on earlier answers.

If Part (a) identifies the wrong variable, Part (b) may use that variable, Part (c) may predict from it and Part (d) may explain the prediction.

One early error can therefore contaminate several parts.

Students should use checkpoints between parts:

• Does my answer to (a) fit the diagram?
• Does the unit make sense?
• Does (b) depend on an assumption I have not verified?
• If (a) were wrong, would the rest collapse?

High-leverage early parts deserve careful checking.

Error propagation during checking

Checking can fix errors or spread them.

If the learner revisits a correct answer using the same wrong assumption that affected another question, they can change a correct answer into a wrong one.

Good checking therefore uses independent checks:

• reread the command;
• recheck axes and units;
• verify the comparison quantity;
• ask whether the conclusion depends on an unstated assumption;
• look for contradictory evidence.

Do not simply re-run the same flawed route more confidently.

The root-cause correction log

Question	Final error	First wrong move	Repair	Retest
Graph Q	Wrong conclusion	Scale misread	Axis-first routine	Changed-scale graph
Plant Q	Wrong concept	Keyword selection	Task/evidence before topic label	Unfamiliar wording
Experiment Q	Overclaim	Assumed controls equal	Given-vs-assumed audit	Incomplete-method question

The correction log should record the first wrong move, not merely the answer key.

High-reach errors deserve priority

Some errors affect many topics.

• graph-scale reading;
• final-value versus change;
• command-word decoding;
• cause/effect direction;
• assumption control;
• concept selection under mixed cues;
• time-interval interpretation.

Repairing one of these can improve Heat, Plants, Forces, Systems and experimental questions at once.

That makes high-reach root errors better intervention targets than isolated chapter facts.

Low-reach errors still matter—but differently

A misspelled technical term or omitted unit may affect one mark locally.

It should be corrected, but it may not deserve the same teaching time as a concept-selection error that contaminates ten questions.

Prioritise by:

• frequency;
• mark cost;
• propagation reach;
• cross-topic reach;
• ease of repair;

This makes revision strategic rather than merely exhaustive.

The upstream checkpoint routine

Before building the explanation, check the upstream layers:

1. What is the command?
2. What evidence is relevant?
3. Are the values and units read correctly?
4. What exactly is being compared?
5. Is the data packet complete?
6. Which concept fits the evidence?
7. Which direction does causality run?

Only then write the mechanism and conclusion.

This costs seconds and can prevent several downstream errors.

The downstream validation routine

After the answer is built, check whether later stages remain consistent with the earlier evidence.

• Does the mechanism predict the observed direction?
• Does the conclusion answer the command?
• Did a new unstated assumption appear?
• Does the wording accidentally reverse the causal chain?
• If one early value changes, which later statements must change too?

This ensures the reasoning chain is internally connected.

The isolation test

To locate the root error, isolate stages.

• Give the correct graph values. Can the student now compare correctly?
• Give the correct comparison. Can they now choose the concept?
• Give the correct concept. Can they build the mechanism?
• Give the mechanism. Can they communicate it precisely?

The first stage where performance breaks identifies the likely bottleneck.

This prevents reteaching the whole chapter when only one reasoning layer is weak.

The changed-representation retest

After repairing a root error, retest it in a different form.

• Graph error → retest with a table.
• Table comparison error → retest with a diagram and labels.
• Concept-selection error → change the wording and context.
• Causal-direction error → reverse the diagram orientation.

If the repair survives representation change, it is less likely to be tied to one memorised example.

The delayed retest

A root repair is not complete because the student can redo the question immediately.

Return after several days.

• Can they still read the scale correctly?
• Can they still identify the comparison quantity?
• Can they still resist the misleading keyword?
• Can they still distinguish given facts from assumptions?

Delayed return checks whether the root repair entered durable memory.

The transfer retest

Finally, move the same reasoning skill into another topic.

• final-value versus change: from plant growth to temperature;
• cause/effect direction: from food webs to body systems;
• missing baseline: from height to water loss;
• assumption control: from plant investigations to circuits.

A root repair has high value when it transfers across topics.

Error propagation in MCQ

MCQ can hide propagation because only one final option is visible.

A learner may eliminate three options using one wrong premise and land confidently on the fourth.

During review, ask:

• What was the first rule you used?
• Which option did it eliminate?
• If that rule were wrong, which options would return?
• What direct evidence actually controls the choice?

High-confidence wrong MCQs are especially useful for finding root errors.

Error propagation in structured answers

Open-ended answers reveal the propagation path more clearly.

Mark the chain:

evidence → selected concept → mechanism → conclusion

If the conclusion is wrong, move left until the first invalid link appears.

Then repair that link and rebuild the answer forward.

The rebuild-forward rule

Once the root is corrected, do not simply patch the final sentence.

Rebuild every dependent step.

1. Correct the root.
2. Recompute or reinterpret the next step.
3. Reselect the concept if needed.
4. Rebuild the mechanism.
5. Rewrite the conclusion.

This ensures no downstream statement remains contaminated by the old error.

Five Primary 6 error-propagation failure modes

1. Final-answer corrector

Changes only the last sentence. Repair by tracing to the first wrong move.

2. Whole-chapter relearner

One root error is treated as total topic weakness. Repair by isolating the failing reasoning stage.

3. Correct-answer trust

A correct final option is assumed to prove the reasoning was sound. Repair by sampling the route behind fragile correct answers.

4. Same-route checker

Rechecks using the identical flawed assumptions. Repair with independent checks on task, units, scale, comparison and assumptions.

5. No-transfer repair

Can redo the corrected question but fails the same reasoning in another topic. Repair with changed-representation and cross-topic retests.

A Phase 4 Primary 6 error-propagation lesson

• Trace: reconstruct the learner’s actual route.
• Divergence: find the first invalid step.
• Classify: task, evidence, representation, comparison, assumption, concept, causality or expression?
• Reach: identify which later answers depend on the root.
• Repair: teach the earliest weak mechanism.
• Rebuild: recompute every downstream step.
• Isolate: test the reasoning stage independently.
• Delay: retest later.
• Represent: change the surface form.
• Transfer: test the same root skill in another topic.

Why small groups help error-propagation diagnosis

Three students can produce the same wrong conclusion from three different roots.

• Student A misreads the scale.
• Student B selects the wrong concept.
• Student C adds an assumption.

Comparing the routes teaches everyone that a wrong answer is not a diagnosis.

It also helps students learn to verbalise the first step they took rather than jumping straight to “careless”.

What parents can practise at home

• Ask “What was the first thing you decided?”
• Ask “Where did your answer first differ from the correct route?”
• Do not stop at the final corrected answer.
• Ask whether one mistake affected later parts.
• Retest the same reasoning skill in another topic.
• Check high-confidence correct answers occasionally for fragile routes.
• Replace “careless” with a specific first-wrong-move description.

How to tell whether error control is improving

• The learner can name the first wrong move.
• Root errors and downstream symptoms are distinguished.
• Corrections target reasoning mechanisms rather than only final answers.
• Multi-part contamination decreases.
• Checking uses independent evidence.
• Fragile correct answers become less common.
• Repairs survive delay and representation change.
• Cross-topic recurrence of the same root error declines.

How this page fits the Hougang Science network

This eduKateSingapore page owns error propagation and first-wrong-move repair. It complements post-paper mark-loss audit, unfamiliar-question decomposition, scientific checking and uncertainty, and sensitivity and robustness.

For the complete P3-to-PSLE reasoning map, use Hougang Primary Science Learning Library.

Official curriculum and examination boundary

The Ministry of Education’s Science Teaching & Learning Syllabus: Primary Three to Six develops scientific analysis, explanation, evaluation and communication. For current examination information, families should refer to SEAB’s PSLE formats examined in 2026.

---

Primary 6 Science correction becomes much more powerful when we stop asking only, “What answer should this have been?” Ask where the reasoning first went wrong, repair that upstream mechanism, rebuild every dependent step, then test whether the repair survives delay, a changed representation and a different topic.