How to perform in the new G2 SEC examinations becomes more reliable when the learner can distinguish information that changes the answer from information that merely sets the scene. Examination questions often contain context, examples, labels, names, diagrams and explanatory detail. Some of those details carry a condition, relationship or constraint. Others help make the question realistic but do not alter the method. The challenge is not to read less. It is to identify what has decision value.
This eighty-eighth Learner’s Guide develops signal-to-noise filtering across G2 English, Mathematics and Science. The central rule is: keep every detail that changes the task, evidence, relationship or constraint; compress the rest. Filtering is not skimming past information. It is controlled reduction after understanding what each detail is doing.
SEAB’s 2027 G2 school-candidate directory currently lists English Language K200, Mathematics K210 and the Science combinations K223, K224 and K225. Use the official G2 syllabus directory for current subject documents. The filtering framework below is eduKateSengkang teaching guidance rather than an official examination procedure.
Signal and noise are question-dependent
The same detail can be signal in one question and noise in another. A date may matter when sequencing events but not when identifying tone. A diagram label may matter when calculating a length but not when interpreting a separate graph. A story detail may be irrelevant to an equation but essential to a constraint.
Therefore, do not mark permanent categories such as “context is noise” or “numbers are signal”. Ask what the current task needs.
The four kinds of signal
- Task signal — the command word or requested outcome.
- Evidence signal — the source line, value, observation or graph feature supporting the answer.
- Relationship signal — the connection among quantities, events or ideas.
- Constraint signal — the condition limiting what counts as a valid answer.
Most high-value information falls into one of these four categories.
Task signal
The task signal tells you what operation to perform. English may ask infer, compare, explain or summarise. Mathematics may ask calculate, show, determine, compare or decide. Science may ask describe, explain, predict, suggest, calculate or evaluate.
If the task signal is missed, excellent subject knowledge can still produce the wrong answer form.
Evidence signal
Evidence signal is the information the answer must use. In English it may be a phrase, sentence or source position. In Mathematics it may be a quantity, graph feature or geometric property. In Science it may be a measurement, pattern, observation or experimental condition.
The learner should be able to point to the evidence before expanding the answer.
Relationship signal
Relationship signal tells you how the evidence fits together. Percentage base, rate, contrast, causal link, scale factor, sequence, comparison and dependency are examples.
A question can contain all the required numbers and still remain unsolved until the relationship is recognised.
Constraint signal
Constraint signal tells you what the answer is allowed to be. Words such as at least, only, except, within, whole number, if and unless often carry more decision value than long descriptive phrases around them.
These small words are easy to overlook because they do not look technical.
The five-second filter
- What is the job?
- Which information directly supports that job?
- What relationship links the useful information?
- What condition limits the answer?
This filter should become fast. It is not an invitation to annotate every line heavily.
Noise is not useless information
Calling a detail “noise” does not mean it is badly written or meaningless. It means that for this particular task, the detail does not change the required reasoning.
A name, background story or decorative statistic may still help the question feel realistic. The learner simply does not need to carry it through the solution.
The deletion test
During practice, remove one detail mentally. If the question can still be answered in exactly the same way, that detail may be non-essential for this subpart.
If removing it changes the answer, method, evidence or validity, it is signal.
The replacement test
Replace a contextual noun with a neutral label. If “solar-powered delivery cart” becomes “vehicle A” and the mathematics remains unchanged, the brand-like description may be narrative rather than structural.
If replacing “battery capacity” with a generic label removes the unit relationship needed for the calculation, you have removed signal.
The condition-loss warning
Filtering becomes dangerous when the learner removes a condition because it looks like context.
“Each bus holds at most 40 passengers” is not background. “The trip is organised by the school eco-club” may be background for a seating calculation. The learner must distinguish story from constraint.
English signal-to-noise filtering
English reading contains many meaningful details, but each question selects a small evidence window.
The learner should not reread the whole passage equally for every question. Use the command word and referent to decide which span deserves close attention.
English example: literal detail
Question: Why did the organiser move the event indoors? If the passage states “heavy rain made the field unsafe”, the signal is the causal clause.
A nearby sentence describing the organiser’s previous successful event may be interesting but irrelevant to this answer.
English example: inference
Question: What does the student’s response suggest about the booking system?
Now both explicit comment and qualification may matter. A single isolated adjective may not be enough. Filtering should identify the small set of details needed to support the inference, not reduce the evidence too aggressively.
English example: two texts
When two sources are compared, source ownership itself becomes signal.
A detail from Text A cannot be imported into Text B merely because both discuss the same programme. Keep speaker, certainty and condition attached to each source.
English example: writing prompt
Situational Writing may include several source points. Not all need equal development, but every required point must appear accurately.
Filter for recipient need: what information helps the audience act, understand or respond? Decorative source language can be compressed; required facts cannot be dropped.
Mathematics signal-to-noise filtering
Mathematics word problems often contain narrative packaging around a compact quantitative structure.
The learner should reduce the story to target, quantities, relationship and constraints without discarding a unit, exception or boundary.
Mathematics example: fixed fee
A delivery company charges a $7 booking fee plus $3 per parcel. The fact that the parcels contain books may be irrelevant to cost.
The $7 fixed fee and $3 variable fee are signal because they define the model.
Mathematics example: travel story
A journey problem may describe landmarks, weather and purpose. The useful core might be two distances, two speeds and a waiting time.
However, if the question asks arrival time and the weather causes a stated 15-minute delay, that detail becomes signal.
Mathematics example: graph context
A graph may describe water level, cost or distance. The context can determine what gradient or intercept means even when the graph calculation is identical.
Do not discard labels so aggressively that the final interpretation becomes impossible.
Science signal-to-noise filtering
Science questions often introduce unfamiliar devices, organisms or processes. The name can create intimidation even when the underlying syllabus principle is familiar.
Replace the unfamiliar name with its functional role: sensor, sample, conductor, reactant, barrier, source or measured system.
Science example: unfamiliar device
A new cooling container is described with several design details. The question asks which one reduces heat transfer through the lid.
Focus on the lid material, thickness or relevant physical feature. A brand name or colour may be narrative unless the question links it to a property.
Science example: graph with context
A graph showing enzyme activity may be wrapped in a story about food production. If the question asks for the optimum temperature, the production story may be noise.
If a later question asks whether the process is suitable for a factory operating at a stated temperature, the context becomes signal.
Science example: experiment
An experimental description may include apparatus arrangement, variable levels, repeats and safety details.
The subpart asking for the independent variable requires one subset; a method-evaluation subpart may require another. Filtering resets at every subpart.
The reset rule
Do not carry one subpart’s filter into the next automatically.
The evidence that was noise for part (a) may become essential for part (c). Re-read the new command before deciding what matters.
The over-filtering error
Over-filtering removes a detail that actually changes the answer.
- dropping “only”;
- ignoring a unit;
- forgetting a time interval;
- removing a condition attached to permission;
- ignoring a comparison baseline;
- discarding an experimental control.
These are high-impact losses because the learner may solve a simpler question that was never asked.
The under-filtering error
Under-filtering keeps every detail active and overloads working memory.
The learner may repeatedly reread names, background stories and numbers that never enter the method. The solution becomes slower because irrelevant information competes with the signal.
The signal stack
For long questions, write a tiny stack:
- Target;
- Evidence;
- Relationship;
- Constraint.
Everything else can remain on the page without occupying the centre of attention.
The signal stack should be temporary
Do not create a permanent four-line annotation for every easy question.
Use it when a prompt is dense or unfamiliar, then let the skill become internal.
The compression test
After identifying signal, compress the question into one sentence without losing the target, relationship or constraint.
Example: “A school orders folders with a fixed delivery fee and per-folder cost; find total cost for eight folders.” The story may originally contain far more detail. If the compressed sentence still produces the same method, the filtering is probably sound.
The reconstruction test
After compression, reconstruct the original answer requirements. If you can no longer tell who the audience is, which unit is required or what condition applies, the compression removed too much.
Good filtering reduces load while preserving every decision-bearing feature.
The cue-word scan
Small words often carry large signal. During difficult reading, scan for:
- only;
- except;
- unless;
- if;
- before;
- after;
- at least;
- at most;
- per;
- each;
- total;
- remaining;
- approximately.
These words frequently define scope, sequence, denominator or constraint.
The number-role scan
Every number in a quantitative question should be assigned a role before use.
A number might be a total, rate, starting value, final value, capacity, count, scale, percentage or distractor. Numbers without labels create calculation risk.
The unit-role scan
Units can reveal whether two numbers are comparable and what operation is plausible.
Kilometres and hours suggest a rate relationship; square centimetres signal area rather than length. Unit information is often stronger signal than the story surrounding it.
The named-entity test
Names can be removed mentally when they do not change relationships.
“Company A”, “School B” and “device C” can replace unfamiliar labels during reasoning. Restore the real names only when the final answer requires them.
The emotional-language test
In English texts, emotionally charged wording may be signal when the question asks tone, attitude or persuasion.
The same wording may be noise for a factual question about time or location. Signal depends on task.
The visual-feature test
In a visual text, colour, size, layout or image can matter if the question asks how meaning or appeal is created.
Do not assume every visible feature needs comment. Select the feature that does explanatory work for the task.
The graph-feature test
A graph contains many visible details. The question may need one trend, one point, one gradient, one intersection or one anomaly.
Filtering protects the learner from describing the entire graph when a precise feature is required.
The experiment-feature test
An experimental description can contain method, apparatus, variables, repeats and observations.
If the question asks for an improvement, identify the weakness first. If it asks for the independent variable, apparatus detail may be secondary. Do not answer every experiment question with the whole method.
Signal-to-noise in MCQ
Multiple-choice questions often contain plausible but non-decisive information.
Read the stem for the exact decision criterion, then use options as competing hypotheses. Eliminate options that fail the key condition rather than comparing every phrase equally.
Signal-to-noise in long English passages
Map paragraph jobs before searching details. One paragraph may introduce the problem, another explain consequences, another present a response.
When a question asks for a consequence, begin in the paragraph whose job matches that operation.
Signal-to-noise in long Mathematics scenarios
Write the target first, then mark only values that connect to it through a valid dependency chain.
This works especially well with multi-stage costs, geometry and rate problems.
Signal-to-noise in long Science structured questions
Use discipline → command → evidence → operation. The surrounding context can be compressed once those four are known.
This links to Vol 0032.
The signal-to-noise drill
- Take one dense question.
- Underline the task signal.
- Circle evidence.
- Box constraints.
- Write the relationship in a few words.
- Cross out or grey out details that do not change the method.
- Solve.
- Restore all conditions during the final check.
This drill makes filtering visible before it becomes automatic.
The reverse-filter drill
Take a short clean question and add irrelevant story detail without changing the method.
Ask another learner to identify which additions are noise. This teaches that noise is defined by decision value, not by length or difficulty.
The condition-reinsertion drill
Start from a simplified problem, then add one condition that changes the answer.
Examples include “at least”, “without replacement”, “after a 20% discount”, “only on Tuesdays” or “for students under 16”. The learner should detect exactly when the method or answer set changes.
The distractor-number drill
Add one unused number to a Mathematics or Science prompt.
The learner must justify why it is unused rather than merely ignore it. This builds confidence that not every number is required.
The distractor-sentence drill
Add one plausible but irrelevant English sentence to a short passage.
The learner answers a specific question and identifies why the extra sentence does not affect the answer.
The small-word drill
Create pairs differing only by one word: at least/greater than, may/will, before/after, some/all.
Ask whether the answer changes. This reveals how tiny textual features can carry more signal than long descriptive phrases.
The time-pressure filter
Under time, use a compressed version of the full system: job → evidence → condition → move.
This is enough to prevent most over-reading without turning filtering into another lengthy procedure.
The error ledger
- irrelevant detail kept active too long;
- important condition discarded as background;
- number used without a role;
- unit ignored;
- source ownership lost;
- graph described too broadly;
- experiment detail answered instead of command;
- context replaced the scientific or mathematical relationship.
These errors show whether the learner under-filtered or over-filtered.
Under-filtering repair
If the learner keeps too much information active, practise compression and role labels.
The goal is a smaller working set: target, evidence, relationship and constraint.
Over-filtering repair
If the learner repeatedly drops conditions, practise the cue-word scan and final reconstruction test.
The solution should remain compact without becoming incomplete.
Filtering and mark security
Use Vol 0025. Filtering protects time for accessible marks by reducing unnecessary processing.
Filtering and evidence-window control
Use Vol 0063 for English rereading. Signal-to-noise filtering is the cross-subject version of the same attention principle.
Filtering and uncertainty quarantine
Use Vol 0076. An unresolved detail does not need to dominate the whole question if the known signal can still be processed.
The four-week filtering build
Week 1 — task and conditions
Mark commands, quantifiers, timing words and boundaries.
Week 2 — roles for numbers and evidence
Label every quantitative value and every English evidence span by function.
Week 3 — dense prompts
Compress long word problems, two-text passages and unfamiliar Science contexts.
Week 4 — timed transfer
Use mixed questions and require filtering to remain fast and selective.
The PSLE bridge
The PSLE habit Read Before You Solve remains the foundation.
At G2, reading becomes more selective: understand the whole question, then keep only the information that changes the next decision.
Use Examination Craft
For pacing and return decisions, continue through the Examination Craft hub. Filtering should save attention, not consume it.
Final rule
The strongest reader does not treat every detail as equally important.
Find the job, evidence, relationship and constraint. Compress narrative that does not change them. Restore every condition before submitting. Good filtering removes cognitive noise without removing mathematical, linguistic or scientific meaning.
Practice clinic one: English passage filtering
Read this invented passage: “The school will trial a quiet study room on Wednesdays from 4 p.m. to 6 p.m. for six weeks. Students must reserve a seat by noon. The room was suggested after a survey showed that some students found the canteen too noisy for revision. The room is on Level 3 beside the library. Students may bring water but not food. The trial will continue next term only if attendance is high enough to justify staffing.”
Question: What condition determines whether the trial continues next term? The signal is the attendance requirement. Wednesday timing, room location, food rule and survey background are noise for this specific question.
Question: Why was the room proposed? Now the survey result becomes signal. The continuation condition becomes noise. Same passage, new filter.
Question: When can students use the room? Wednesday, 4 p.m. to 6 p.m. becomes signal. Filtering resets with the task.
Practice clinic two: English comparison
Text A says a programme may continue if registrations remain high. Text B says the writer wants the programme extended because it has been useful.
Question: What do both texts say about continuation? Signal from A is conditional possibility; signal from B is support. Do not flatten them into “both confirm continuation”. The condition and source status remain signal.
Practice clinic three: Mathematics narrative
A school trip includes breakfast, two bus journeys, a museum ticket and a lunch voucher. Each bus seats 45 students. There are 128 students and six teachers. The bus company charges $320 per bus. The museum ticket costs $12 per student, but the question asks only for the minimum number of buses and total bus hire cost.
Student and teacher counts, bus capacity and bus price are signal. Museum ticket and lunch voucher are noise for this subpart. Total travellers = 134, so three buses are needed. Bus hire = 3 × 320 = $960.
If a later subpart asks total trip cost, museum ticket information becomes signal. Filtering changes with the target.
Practice clinic four: Mathematics hidden condition
A store offers a $20 delivery fee plus $5 per box. A customer has a budget of at most $95. The boxes contain stationery for a charity event.
For maximum boxes, the charity context is noise but “at most $95” is crucial signal. Solve 20 + 5n ≤ 95, giving n ≤ 15. The boundary condition determines the final integer.
Practice clinic five: Science unfamiliar device
A fictional “ThermaShell” container has a reflective lid, foam walls, a digital thermometer, a blue outer coating and a carrying strap. A student measures cooling over ten minutes. The question asks which feature is intended to reduce conduction through the walls.
Foam walls are signal. Reflective lid may matter to another heat-transfer mechanism; blue coating and strap may be irrelevant unless the question connects them to a property.
Filtering is not deleting all unfamiliar detail. It is matching each feature to the mechanism the subpart asks about.
Practice clinic six: Science experimental description
A plant investigation records height every three days, keeps water volume constant, uses identical pot size, changes light intensity, and labels each pot with a colour code. The question asks for the independent variable.
Light intensity is signal. Water volume and pot size are controlled-variable information but not the answer. Colour code is organisational detail. Height is the dependent measurement.
A later question asking how to improve control might turn water volume or pot size into signal.
Practice clinic seven: MCQ filtering
A multiple-choice question gives three values, a unit conversion and four answer options. One option matches a common error from ignoring the conversion.
Signal is the quantity relationship plus unit conversion. The story around why the measurement was taken may be noise. Eliminate options by checking the required unit before recalculating everything.
Practice clinic eight: English visual text
Imagine a poster with a large heading “LAST WEEK TO REGISTER”, a deadline date, photographs of previous participants and a paragraph describing programme benefits.
Question: Which feature most directly creates urgency? The deadline wording and date are signal. Programme benefits may support attractiveness, but not the specific urgency claim.
Question: What benefit is emphasised? Now the descriptive paragraph and perhaps images become signal. Again, task changes the filter.
Practice clinic nine: Science data table
A table contains temperature, time, colour score and sample mass. The question asks for average cooling rate.
Temperature and time are signal. Colour score and mass may be irrelevant unless they affect a later comparison. The learner should not calculate with every column simply because the data is present.
Practice clinic ten: Mathematics diagram
A diagram contains a rectangle inside a larger shape, several labelled lengths and one decorative label naming the building represented.
If the question asks shaded area, only lengths defining the relevant regions matter. The building name is noise. If one labelled length belongs to an unused extension, it may also be noise for that subpart.
The signal-confidence distinction
A detail can be signal even when the learner is uncertain how to use it. Do not discard unfamiliar information simply because it is difficult.
The test is whether changing the detail would change the question, not whether the learner feels confident about it.
The noise-confidence distinction
A familiar number can still be noise. Students often use every number because calculation feels productive.
Familiarity is not relevance. Assign a role before use.
The red-herring discipline
Some questions may include plausible extra details, but examination writers are not necessarily trying to trick the learner with deliberate red herrings.
Avoid a suspicious mindset. Simply determine whether the detail enters the reasoning required by the current task.
The first-pass and second-pass filter
On the first read, understand the whole prompt. On the second read, reduce it to signal.
Skipping directly to selective reading before understanding can cause over-filtering because the learner does not yet know which conditions matter.
Filtering and working memory
Working memory has limited capacity. Dense prompts become easier when the learner externalises signal through labels, underlining or a small sketch.
The goal is not to memorise the entire story. It is to preserve enough structure to reason accurately.
Filtering and unfamiliar language
If one word is unfamiliar, ask whether the sentence defines its role. A technical-sounding name may be replaceable with “device A” without losing the relationship.
If the unknown word itself carries the property being tested, it must be resolved rather than filtered away.
Filtering and multiple constraints
A complex problem may contain several constraints. Do not compress them into a vague phrase such as “within limits”.
List each boundary separately: budget, capacity, time, whole-number requirement. The solution must satisfy all of them.
Filtering and evidence synthesis
Some English and Science questions need multiple evidence points. Filtering should remove irrelevant detail without reducing the answer to one clue when the question requires two independent supports.
Use the smallest complete evidence set, not the smallest possible evidence set.
Filtering and checking
During checking, reconstruct the dropped conditions. Ask whether any ignored detail changes the final answer.
This catches over-filtering after an efficient solution has been found.
The two-risk model
- Under-filtering risk: cognitive overload and wasted time.
- Over-filtering risk: missing a condition and solving the wrong problem.
Good filtering sits between the two: compact but complete.
The signal-to-noise scorecard
- Task identified quickly;
- evidence window appropriate;
- numbers assigned roles;
- constraints preserved;
- irrelevant context ignored after understanding;
- final answer checked against omitted details.
This scorecard can be used after mixed practice without turning into a formal grading system.
Final perspective
Filtering is a form of disciplined attention.
The learner reads everything needed to understand the task, then gives mental priority only to information that changes the answer. This creates room for reasoning without sacrificing accuracy. The goal is not less reading. It is better allocation of attention.
Final signal-to-noise practice block
Use one dense prompt and create two summaries. The first summary may contain only the target, the evidence needed, the relationship and the constraints. The second summary may contain the surrounding context. Solve from the first summary, then compare with the original prompt to confirm that no condition was lost.
This drill teaches a crucial distinction: information can be useful for understanding the story without being necessary for the calculation or explanation. The learner should be able to acknowledge context without carrying all of it through working memory.
Filtering after a wrong answer
When an answer is wrong, ask whether the learner filtered too much or too little. Over-filtering usually shows up as a missing condition, ignored qualifier, lost unit or incomplete evidence set. Under-filtering usually shows up as repeated rereading, use of irrelevant numbers, oversized working or a response that answers several nearby questions instead of the one asked.
The repair should match the direction of the error. Over-filtering needs a stronger reconstruction check. Under-filtering needs more aggressive role labelling and compression.
Filtering during the return pass
When returning to a skipped question, do not reread the entire prompt at full depth automatically. First recover the signal stack you already identified. Ask what blocked the original attempt: missing relationship, uncertain evidence, forgotten condition or calculation difficulty. Then reread only the part needed to resolve that block.
This makes the second pass genuinely different from the first rather than a slower repetition of the same unsuccessful reading.
The advanced filtering standard
An advanced learner can hold a dense prompt lightly. They understand the full setting, but they know which four or five pieces actually drive the current decision. They can also restore a previously ignored detail instantly when a new subpart makes it relevant.
That flexibility is the goal. Signal-to-noise filtering is not deleting information from the page; it is deciding what deserves attention now.
The final filtering calibration
After a timed set, review every detail you actively used. Mark whether it was necessary, helpful but optional, or irrelevant to the final method. Then inspect every detail you ignored and ask whether any omitted condition should have altered the answer. This two-sided audit calibrates both under-filtering and over-filtering.
A learner who repeatedly carries irrelevant detail needs stronger compression. A learner who repeatedly drops boundary words needs a slower first read and stronger reconstruction check. The goal is not a universal level of simplification; it is a personal level that preserves every condition while freeing enough attention for reasoning.
Signal-to-noise control is mature when the learner can compress aggressively and still restore the exact wording that limits the answer whenever checking requires it.
The final filtering check is to ask whether every piece of retained information changes either the method, the evidence, the interpretation or the validity of the answer. If it changes none of them, it can remain outside the active working set. If it changes even one, restore it before submission.
Filtering is complete when the working set is smaller, but the original task is still fully reconstructable.
