Series: How to Prepare For — Global Examination Performance Edge Articles
Article P023
The student knows how to do the question.
The method is correct.
The concept is secure.
The answer should be there.
Then one sign changes.
One unit disappears.
One word in the question is missed.
One number is copied incorrectly.
One answer is transferred to the wrong line.
One paragraph answers the topic but not the command.
One final check is skipped because time is running out.
The marks leak.
This is a different examination problem from not knowing the content.
It is also different from being unable to do the questions at all.
The learner can do the work.
But the performance system is not yet precise enough to protect what the learner already knows.
This article owns that edge condition.
It follows How to Prepare for an Exam When You Understand the Topic but Cannot Do the Questions | Repair the Conversion From Knowledge to Performance. That article owns the gap between understanding and independent execution. This page begins after the learner can already execute many questions correctly and asks a narrower, higher-resolution question:
Why do small avoidable mistakes still survive, and how do we stop reliable knowledge from leaking marks?
eduKateSG already has a broader owner for the phrase “careless mistakes” at I Keep Making Careless Mistakes in Exams — What Should I Do?. This article therefore does not duplicate that page. It owns the examination-preparation layer: how to build an accuracy system that remains reliable under realistic time, switching, fatigue and pressure.
Current high-ranking exam-advice pages repeatedly cluster around a familiar keyword field: common exam mistakes, careless mistakes in exams, misreading the question, calculation errors, unit conversion mistakes, checking answers, exam pressure, error logs, and how to stop losing marks. The useful opportunity is not to repeat another generic “be careful” list. It is to explain the mechanism underneath these phrases and show how to train precision as a performance skill.
The 50-second answer
If you can do the questions but keep making small mistakes, do not simply tell yourself to concentrate harder.
- Separate knowledge errors from execution errors.
- Classify the exact mistake family: reading, copying, sign, unit, operation, label, transfer, scope, checking or timing.
- Find the step where the error first enters the work.
- Build a specific prevention cue at that step.
- Use a short personal error log, not a giant notebook of every wrong answer.
- Practise the prevention cue under normal time pressure.
- Use selective checking instead of rereading everything equally.
- Protect high-risk transitions: question → plan, line → line, calculation → answer, page → answer sheet, draft → final response.
- Track high-confidence wrong answers urgently.
- Train enough fluency that basic execution does not consume all attention.
- Retest on fresh questions to prove the error family is actually shrinking.
- Judge success by fewer repeated leaks, not by one perfect paper.
The central rule is:
A small exam mistake is worth training only when you can name the behaviour that creates it and install a better behaviour in the same place.
Alicia, Tricia and Kai Kai all know the answers
Alicia completes a Mathematics paper.
She loses a mark because she copies 36 as 63.
Another because a negative sign disappears.
Another because she writes centimetres instead of square centimetres.
She says:
“I was careless.”
Tricia completes an English paper.
She understands the passage.
But she answers “why” with a description of “what happened.”
Later she misses one part of a two-part question.
She also says:
“Careless.”
Kai Kai does something more useful.
She asks what kind of carelessness occurred.
The copied number was a transfer error.
The lost negative sign was a symbolic continuity error.
The unit mistake was an answer-completion error.
The English response was a command-scope error.
The missed second part was a question-completeness error.
Now the mistakes are no longer one vague personality trait.
They are trainable mechanisms.
What this article owns
This article owns small-mark leakage after the learner is already broadly capable of doing the task.
- Error classification: replacing the word careless with a specific failure family.
- Entry-point detection: finding where the wrong information first enters the work.
- Prevention cues: installing a small check at the point of risk.
- Selective verification: checking what is most likely to fail rather than rereading everything.
- Transition control: protecting handoffs between reading, working, recording and submission.
- Load control: keeping accuracy stable when time, switching and fatigue rise.
- Error-log design: tracking recurring mechanisms rather than archiving every wrong answer.
- Fresh retesting: proving that the leak is shrinking on new material.
First principle: “careless” is a label, not a diagnosis
Students, parents and teachers often use “careless” because the learner knew the answer.
But knowing the answer does not explain why execution failed.
“Careless” can hide:
- rushing;
- weak attention at transitions;
- poor working layout;
- fragile symbolic control;
- overconfidence;
- weak checking habits;
- timing pressure;
- fatigue;
- method overload;
- unclear answer-completion rules.
The repair depends on which one is happening.
The careless-to-specific conversion
| What the student says | Better diagnosis |
|---|---|
| “I was careless.” | I copied one value incorrectly between lines. |
| “I rushed.” | I began writing before checking the command word and scope. |
| “I forgot the unit.” | I do not have a final-answer completion check. |
| “I misread it.” | I skim negative and exception words under time pressure. |
| “I knew it.” | I had the knowledge but failed at execution or verification. |
Second principle: separate knowledge errors from execution errors
This distinction is foundational.
A knowledge error means the learner did not know enough to produce the correct route.
An execution error means the learner had a viable route but lost accuracy while carrying it out.
Do not repair one with the treatment for the other.
More content revision does not necessarily fix a copied number.
More checking does not fix a missing concept.
The two-question diagnostic
After a mistake, ask:
1. Before seeing the answer, did I know the correct method or idea?
2. If I redo the question slowly without help, can I complete it correctly?
If both are yes, the issue is likely downstream of knowledge.
Third principle: find the first wrong event
The final wrong answer is often several steps downstream.
Trace backward.
Example:
Wrong final answer ← wrong intermediate value ← copied 48 as 84 ← visual transfer error.
The repair target is not the final arithmetic.
It is the copy transition.
The first-wrong-event method
For every small mistake:
- Find the first moment the work stopped being correct.
- Name the action occurring at that moment.
- Identify what cue could interrupt the error next time.
- Practise that cue on fresh questions.
Fourth principle: transitions are high-risk zones
Many small mistakes do not happen while thinking deeply.
They happen while information is moved.
Common transitions include:
- question → working;
- diagram → equation;
- line → next line;
- calculator → script;
- draft → answer booklet;
- working → final answer;
- question number → answer sheet;
- source → written evidence.
These are handoffs.
Handoffs deserve control.
The handoff rule
At a high-risk transfer, use a tiny verification action.
Examples:
Copying a number → glance back once before continuing.
Transferring MCQ answer → say question number and option together silently.
Writing a final numerical answer → value + unit + required precision.
Moving evidence into an essay → claim first, then evidence that actually supports that claim.
Fifth principle: not every step deserves equal checking
Students are often told:
“Check your work.”
But checking everything equally is expensive.
Strong checking is risk-weighted.
Check more where:
- the learner has a known history of errors;
- one value propagates into later parts;
- the step is irreversible;
- the instruction contains a negative or exception;
- the answer requires units, labels or rounding;
- an answer is transferred elsewhere.
Sixth principle: checking should be designed before the exam
Do not wait until the final five minutes and then decide what “checking” means.
Build a checking system during practice.
Possible personal checks:
- negative/except words;
- sign changes;
- units;
- answer completeness;
- question number transfer;
- scope;
- labels;
- final judgement.
Use only what the learner actually needs.
The personal five-check rule
Keep the final checklist short enough to remember.
For example:
Command. Sign. Unit. Scope. Transfer.
Another student may need:
Negative. Working. Label. Evidence. Conclusion.
The checklist should come from the learner’s own error history.
Seventh principle: the best error log is small and alive
Current exam-advice competitors increasingly recommend error logs.
The idea is useful.
But a giant error archive can become another unread notebook.
Track error families.
| Trigger | Error | Prevention cue | Last fresh retest |
|---|---|---|---|
| Negative stem | Answered opposite | Box the negative | Green |
| Long calculation | Dropped sign | Line-by-line sign scan | Amber |
| Final value | Unit omitted | Value + unit lock | Green |
| Two-part question | Second part missed | Count requested outputs | Amber |
Eighth principle: prevention cues must be attached to triggers
A cue floating in memory is weak.
Attach it to a situation.
If I see “except” → I box it.
If I move a number from one line to another → I compare once.
If I write a final numerical answer → I check unit and requested precision.
If the question asks for two reasons → I count two before moving.
Trigger-action links reduce the need to remember to “be careful” everywhere.
Ninth principle: visual working layout matters
Messy working increases transfer risk.
Good layout helps the learner see:
- where values came from;
- where signs changed;
- which line is current;
- what remains to be answered;
- where the final answer belongs.
Layout is not decoration.
It is external memory.
The one-step-one-line principle
In multi-step quantitative work, do not compress too many transformations into one line while accuracy is unstable.
One meaningful transformation per line makes errors easier to prevent and easier to detect.
Tenth principle: speed should not be bought with ambiguity
Students sometimes make working shorter to become faster.
If the compressed route increases mistakes, the speed is false economy.
First build a clean route.
Then remove only genuinely redundant steps.
Eleventh principle: error prevention should happen upstream
It is cheaper to prevent a mistake than to discover it five minutes later.
Examples:
- read the command correctly before writing;
- check the copied value before using it in three later lines;
- label a sample before two samples become visually similar;
- write the unit with the answer before moving on.
Upstream checks are usually faster than downstream repair.
Twelfth principle: some mistakes are propagation risks
A small error can remain local.
Or it can contaminate everything after it.
High-propagation errors deserve stronger checks.
Examples:
- wrong sign in a linked calculation;
- wrong sample label;
- wrong question number on an answer sheet;
- wrong premise in an essay;
- wrong unit conversion used repeatedly.
The propagation question
Ask:
If this step is wrong, how much downstream work becomes wrong?
The higher the propagation cost, the more valuable an immediate check becomes.
Thirteenth principle: negative wording is a special risk
“Which is NOT…”
“Except…”
“Least likely…”
“Incorrect…”
These words invert the task.
Under speed, the eye can skim past them.
Use a deliberate mark or pause.
The inversion lock
When the question reverses the ordinary direction, make the reversal visually or mentally explicit before looking at options.
Do not trust general familiarity.
Fourteenth principle: two-part questions create completion risk
Students answer the first demand and feel finished.
Train output counting.
If the question asks:
“Explain X and state one limitation,”
write mentally:
Output 1: explanation.
Output 2: limitation.
Fifteenth principle: units are part of the answer architecture
Units should not be an afterthought.
When the question asks for a quantity, think:
value + unit + required precision.
This simple completion structure prevents many small numerical leaks.
Sixteenth principle: rounding should happen at the right time
Premature rounding can create downstream error.
Late rounding can create answer-format error.
Follow the subject and exam conventions.
Keep sufficient intermediate precision where required, then round the final answer according to instructions.
Seventeenth principle: calculator entry is a separate risk surface
Correct mathematics can be damaged by incorrect calculator entry.
Train:
- bracket discipline;
- sign entry;
- mode awareness where relevant;
- order-of-operations awareness;
- reasonableness checking.
The calculator should reduce arithmetic load, not remove mathematical supervision.
Eighteenth principle: estimate before trusting the display
If the expected answer should be around 20 and the calculator gives 2,000, stop.
Estimation catches some entry errors cheaply.
Use order of magnitude, sign and rough range.
Nineteenth principle: copying is an examinable operation
Students copy:
- numbers;
- equations;
- quotations;
- labels;
- answer choices;
- question numbers.
Each copy is a transfer event.
Practise accurate transfer instead of assuming it is trivial.
Twentieth principle: high-confidence wrong answers are dangerous
If the learner is wrong and uncertain, the system already has a warning signal.
If the learner is wrong and confident, checking may never activate.
High-confidence wrong answers deserve priority diagnosis.
They may indicate:
- a stable misconception;
- a habitual misread;
- a bad automatic rule;
- overgeneralisation.
Twenty-first principle: low-confidence correct answers can waste time
A learner answers correctly but does not trust the answer.
They check three times.
They erase.
They rewrite.
Time leaks.
Calibration is part of accuracy.
Twenty-second principle: do not change correct answers without evidence
Changing an answer because it suddenly “feels wrong” is weak.
Changing because you identify a specific contradiction, calculation error or misread is stronger.
Require a reason to override a completed answer.
Twenty-third principle: review should search for known failure families
Random rereading has low precision.
If your history shows:
- units;
- negatives;
- scope;
- transfer;
review those first.
Personal checking is more efficient than generic checking.
Twenty-fourth principle: checking should not become compulsive
More checking is not always better.
Repeatedly checking secure work can consume the time needed to complete unattempted questions.
Use a checking hierarchy.
- Unattempted high-value work.
- Flagged uncertainty.
- Known personal error families.
- Global completeness check.
Twenty-fifth principle: answer-sheet transfer needs its own routine
Where answers are transferred to a separate sheet, the transfer process is a distinct performance stage.
Train it during mocks.
Possible routines include:
- transfer in small batches;
- say question number + option silently;
- check sequence continuity after each batch;
- protect final time for completeness.
Follow the exam’s specific rules.
Twenty-sixth principle: small mistakes increase when the basic method is not fluent
If too much working memory is consumed by basic operations, less remains for monitoring.
This is why “careless mistakes” can increase when foundations are weak.
The learner may know the concept but still be using too much attention on:
- basic arithmetic;
- algebraic manipulation;
- sentence construction;
- formula recall;
- evidence retrieval.
Fluency can create monitoring capacity.
Twenty-seventh principle: accuracy first, then pressure
Do not train speed on top of unstable technique.
Use:
slow-clean → normal-clean → timed-clean → mixed-timed-clean.
Each stage adds load only after the previous one is reasonably reliable.
Twenty-eighth principle: pressure reveals weak routines
Under time, learners revert to defaults.
If the default is:
rush → skip working → hope → check randomly,
pressure exposes it.
If the default is:
read → identify → execute → verify key transition → move,
pressure has less room to create drift.
Twenty-ninth principle: time checkpoints protect accuracy
When students fall behind without noticing, they suddenly accelerate.
That acceleration often increases error rate.
Use paper checkpoints so timing corrections happen gradually.
Thirtieth principle: rushing is often downstream of earlier overinvestment
Late-paper carelessness may be caused by early-paper perfectionism.
Trace time upstream.
If thirty extra minutes were spent polishing earlier answers, the final rush is not simply a concentration problem.
Thirty-first principle: perfectionism can create careless errors later
This seems paradoxical.
But overchecking one answer can consume the buffer that protects the rest of the paper.
Precision needs allocation.
Thirty-second principle: fatigue changes the error profile
Late in long papers, students may show more:
- copying errors;
- missed qualifiers;
- omitted units;
- incomplete answers;
- answer-transfer mistakes.
Compare the first and final quarters of practice papers.
If small errors cluster late, train stamina and protect earlier pacing.
Thirty-third principle: error location across the paper matters
Early errors suggest a different mechanism from late errors.
Early:
- rushed start;
- poor instruction reading;
- state transition into the exam;
- overconfidence.
Late:
- time pressure;
- fatigue;
- reduced checking;
- incomplete work.
Thirty-fourth principle: practise the first five minutes
The opening routine should be stable.
Read instructions → map the paper → begin at planned pace.
A controlled start reduces early careless errors.
Thirty-fifth principle: practise the final ten minutes
What exactly happens when ten minutes remain?
The learner should already know.
For example:
complete unattempted high-value work → resolve flagged items → personal error checks → submission/transfer check.
Adapt to the exam format.
Thirty-sixth principle: small errors should be measured by family, not embarrassment
Do not record:
“Stupid mistake.”
Record:
“Dropped negative when distributing bracket under time.”
The second entry contains a trigger, action and context.
That can be trained.
Thirty-seventh principle: frequency matters more than drama
A spectacular one-off mistake may feel memorable.
But three repeated one-mark leaks can be more important.
Count recurrence.
Repeated families deserve priority.
Thirty-eighth principle: mark value matters too
Some errors cost one mark.
Some propagate through a ten-mark linked problem.
Prioritise by:
frequency × expected mark cost × propagation risk.
Thirty-ninth principle: use fresh questions to test prevention
Redoing the same question after correction may prove memory of the correction.
Use a fresh question containing the same risk.
Example:
Sign error repaired → fresh algebra problem with several sign changes.
Unit omission repaired → fresh quantitative problem with derived units.
Scope error repaired → fresh two-part language question.
Fortieth principle: delayed retesting matters
A prevention cue that works immediately may disappear tomorrow.
Retest after delay.
The new behaviour should become default, not short-term compliance.
Forty-first principle: error correction should end with a new default
The goal is not:
“I understand why I made that mistake.”
The goal is:
“Next time this trigger appears, I automatically perform the safer action.”
That is behavioural repair.
Forty-second principle: subject-specific checks beat generic reminders
Mathematics:
- sign;
- unit;
- substitution;
- reasonableness;
- domain restrictions.
Science:
- unit;
- variable;
- label;
- mechanism;
- data direction.
English:
- command;
- scope;
- evidence;
- grammar;
- completeness.
Humanities:
- source;
- provenance;
- comparison direction;
- evidence;
- judgement.
Forty-third principle: the learner should know their top three leak families
Not twenty.
Three.
For example:
- negative wording;
- unit omission;
- answer transfer.
These become the personal attention targets for the next training cycle.
Forty-fourth principle: once a leak is green, retire it from prime attention
Do not keep mentally checking a solved problem forever.
If several fresh retests show the error family has disappeared, move it to maintenance.
Attention is finite.
Forty-fifth principle: new error families can appear as speed increases
When the learner becomes faster, different leaks may emerge.
Continue measurement.
Accuracy training is adaptive.
Forty-sixth principle: the goal is not zero human error
No realistic exam system eliminates every mistake.
The objective is to reduce predictable, repeated and high-cost errors enough that the final score reflects actual knowledge more closely.
Forty-seventh principle: small mistakes are often easiest to fix after the learner becomes broadly competent
When core knowledge is unstable, too many errors compete for attention.
Once knowledge and method are reliable, the remaining leaks become more visible.
This makes late-stage precision training unusually valuable for strong students.
Forty-eighth principle: top-grade preparation often becomes mark protection
At higher performance levels, improvement may come less from learning huge amounts of new content and more from:
- cleaner interpretation;
- faster retrieval;
- more precise execution;
- better checking;
- fewer small leaks.
The ceiling and the floor converge through precision.
Forty-ninth principle: accuracy is a system property
The learner is not simply “careful” or “careless.”
Accuracy emerges from:
- clear knowledge;
- clean working;
- stable timing;
- good transitions;
- risk-weighted checking;
- trained error cues;
- fresh retesting.
Fiftieth principle: the paper should become boringly reliable
The highest compliment for a precision system is not that it feels clever.
It is that the student keeps doing the small correct things without needing to think dramatically about them.
Question read.
Method chosen.
Working clean.
High-risk transition checked.
Answer completed.
Move.
Reliability should become ordinary.
The small-mistake error ledger
| Error family | Trigger | Prevention cue | Fresh retest | Status |
|---|---|---|---|---|
| Negative wording | NOT / EXCEPT / LEAST | Box inversion word | 10 mixed items | Amber |
| Sign drift | Bracket expansion | One transformation per line | 5 fresh problems | Green |
| Unit omission | Final numerical answer | Value + unit + precision | 8 quantitative items | Green |
| Scope miss | Multi-part language question | Count outputs | 6 fresh questions | Amber |
| Transfer | Separate answer sheet | Number + option lock | Full MCQ set | Amber |
The 14-day accuracy runway
Days 14–13: collect recent scripts and classify small errors by family.
Days 12–11: identify the top three repeated leak families.
Days 10–9: build prevention cues and practise them slowly.
Day 8: fresh mixed questions under normal pace.
Day 7: first timed section with personal checklist.
Day 6: inspect which errors survived time pressure.
Day 5: targeted micro-drills on surviving leaks.
Day 4: full representative paper.
Day 3: compare error-family counts with the baseline.
Day 2: light fresh retest and finalise the personal checking hierarchy.
Day 1: short retrieval, logistics and taper.
The 7-day accuracy runway
Day 7: classify error families.
Day 6: train two highest-cost prevention cues.
Day 5: fresh mixed questions.
Day 4: timed section.
Day 3: repair surviving leak.
Day 2: full or representative simulation.
Day 1: personal checklist and taper.
The 3-day accuracy rescue
- Inspect the latest marked paper.
- Choose the three most repeated avoidable errors.
- Create one prevention cue for each.
- Run fresh questions containing those triggers.
- Run one timed representative set.
- Carry only the short personal checklist into the final preparation phase.
Subject application: Mathematics
Mathematics small-mark leakage often clusters around:
- signs;
- copied numbers;
- brackets;
- units;
- rounding;
- calculator entry;
- wrong operation;
- incomplete final answers.
Do not use the phrase “careless in Math” as the final diagnosis.
Trace the first wrong line.
If the method is correct, isolate the micro-operation that failed.
Subject application: Additional Mathematics
A-Math magnifies symbolic propagation.
One sign or algebraic drift can damage several later lines.
Use:
- one meaningful transformation per line when risk is high;
- domain/condition checks;
- substitution checks where appropriate;
- reasonableness checks on final results.
Accuracy comes from structure, not simply slower handwriting.
Subject application: Science
Science small mistakes often include:
- wrong unit;
- missing keyword;
- confusing observation and explanation;
- wrong variable;
- incorrect graph label;
- answering a neighbouring question.
Use output-specific checks rather than generic rereading.
Subject application: English comprehension
English small leaks may look like:
- misreading the command;
- missing a qualifier;
- answering outside the requested lines;
- using evidence without explaining it;
- pronoun-reference confusion;
- incomplete two-part answers.
Train command, scope, evidence and completeness as separate checkpoints.
Subject application: English writing
Writing leaks can include:
- tense drift;
- subject-verb agreement;
- missing words;
- spelling of high-frequency words;
- paragraphs without clear claims;
- conclusions that do not answer the exact task.
Do not proofread every sentence with equal intensity.
Use the learner’s known grammar and task-completion error families.
Subject application: Humanities
Common small leaks include:
- wrong source reference;
- missing provenance;
- comparison without both sides;
- evidence copied but not interpreted;
- judgement omitted.
Build a response-completeness check.
Subject application: multiple-choice exams
Small errors include:
- negative-stem misread;
- answer-sheet transfer;
- changing correct answers without evidence;
- calculator-entry errors;
- skipping a question accidentally.
Use structured transfer and completion checks.
Subject application: practical exams
Practical mistakes often arise at transitions:
- sample → label;
- instrument → reading;
- reading → record;
- observation → interpretation;
- procedure → safe close.
Embed checks into the procedure itself.
Subject application: online exams
Digital exams add:
- wrong field entry;
- navigation errors;
- unsaved work;
- file-upload mistakes;
- answer transfer between digital spaces.
Interface accuracy is part of exam accuracy.
The parent’s role
Parents should stop repeating:
“Be more careful.”
Ask instead:
- What exact mistake happened?
- Where did it first enter the work?
- Has this happened before?
- What cue will prevent it?
- When will you test that cue on a fresh question?
This shifts the conversation from character judgement to system repair.
The tutor’s role
The tutor should classify errors by mechanism.
A good tutor distinguishes:
- knowledge gap;
- retrieval gap;
- method-selection gap;
- execution leak;
- checking failure;
- timing failure.
Then the intervention can be small and precise.
The teacher’s role
Feedback becomes more actionable when it identifies the operation that failed.
Instead of:
“Careless.”
Use:
“You copied the denominator incorrectly when moving from line two to line three.”
Or:
“You answered the topic correctly but did not address the comparison command.”
Precision in feedback creates precision in repair.
The independence test
The learner is becoming accurate when they can:
- name their recurring error families;
- apply the right prevention cue automatically;
- maintain clean working under time;
- check selectively rather than compulsively;
- complete transfers accurately;
- finish with units, labels and required outputs;
- show fewer repeated errors on fresh papers;
- keep accuracy stable late in the paper.
Return to Alicia
Alicia stops writing “careless” beside every error.
She writes:
copy transfer.
sign continuity.
unit completion.
Three vague mistakes become three routines.
Return to Tricia
Tricia learns that rereading the whole English paper is not helping.
Her repeated leak is command-scope mismatch.
She starts marking command and output before writing.
The error count falls.
Return to Kai Kai
Kai Kai enters a full mock.
She does not try to be hyper-careful everywhere.
She protects the high-risk points she has trained.
The paper feels faster, not slower.
Because the checking system is small.
And because fewer mistakes need repairing later.
The deeper lesson
Marks can be lost after knowledge has already done its job.
The concept was understood.
The method was selected.
The learner was capable.
Then execution drifted.
This is why top-level examination preparation eventually becomes a reliability problem.
Knowledge must survive the final centimetres of the journey:
read accurately → work cleanly → transfer faithfully → answer completely → verify intelligently.
What proper preparation finally means
Proper preparation when you can do the questions but keep making small mistakes means stopping the vague instruction to “be careful.”
Find the leak.
Name the family.
Find the trigger.
Install the cue.
Practise slowly.
Add time.
Add switching.
Add a full paper.
Retest fresh.
Keep only the checks that earn their place.
Then make the final examination reflect what you actually know.
Where to go next
If the broader issue is still “careless mistakes” across many contexts, use I Keep Making Careless Mistakes in Exams — What Should I Do?.
If you understand the topic but cannot start or solve exam questions reliably, use How to Prepare for an Exam When You Understand the Topic but Cannot Do the Questions.
If one weak mock exposed many errors at once, use How to Prepare for an Exam After a Bad Mock Exam.
If checking itself is unreliable, use I Check My Paper but Still Miss Mistakes — How Should I Check Better?.
Next in this series: How to Prepare for an Exam When You Keep Running Out of Time | Protect the Marks That Never Reach the Page.
Final principle
If you already know how to do the question, the remaining work is precision engineering.
Do not ask yourself to become a more careful person.
Build a more reliable process.
Protect the transitions.
Check the high-risk points.
Use clean working.
Train under time.
Retest fresh.
And stop donating marks to mistakes your knowledge had already earned the right to keep.