How to use Super Intelligence to challenge your assumptions begins with a difficult truth: many decisions are determined before the visible reasoning even starts. The hidden driver is often an assumption you never wrote down.
You assume the deadline is fixed. You assume more study time is the answer. You assume the expensive product will last longer. You assume a colleague is unavailable. You assume a student is weak at a whole subject. Once the assumption enters the model, everything built on top of it can look perfectly logical.
Super Intelligence can help because it can externalise assumptions, generate counterexamples, test alternative interpretations and identify what evidence would change the conclusion. The value is not automatic disagreement. The value is making invisible premises visible enough to inspect.
In the eduKateSG life series, Super Intelligence, or SI, is our editorial name for practical AI assistance. This guide follows How to Think More Clearly with Super Intelligence and How to Use Super Intelligence to Break Down Complicated Problems.
The goal is not to doubt everything. The goal is to locate the assumptions that materially control the decision, then test them in proportion to the consequence.
What an Assumption Is
An assumption is something your reasoning is treating as true without establishing it fully inside the current decision.
Assumptions are not automatically mistakes. Every practical decision uses them. You assume tomorrow’s calendar will mostly remain intact. You assume a familiar route will take roughly the usual time. You assume an ordinary household item will behave similarly to previous examples.
The problem is unpriced assumptions: beliefs that control the plan even though nobody has noticed how much the plan depends on them.
A strong SI conversation therefore asks two questions: which assumptions are present, and which assumptions are load-bearing enough to deserve verification?
The Assumption Stack
Factual assumptions
Claims about current reality: the deadline is Friday, the price includes delivery, the student understands fractions, the meeting requires a presentation.
Causal assumptions
Claims about why something happened: marks fell because the topic became harder; the project is late because staff are overloaded.
Behavioural assumptions
Claims about how people will act: the family will follow the new calendar, customers will pay for the service, a colleague will reply quickly.
Resource assumptions
Claims about available time, money, attention, staffing, devices or expertise.
Temporal assumptions
Claims about what will remain stable over time: prices, requirements, availability, priorities, permissions and workload.
Value assumptions
Claims about what matters most: lower price is better, speed matters more than flexibility, more tuition is better than more independent practice.
Identity assumptions
Claims that convert current patterns into permanent traits: I am bad at planning; my child is weak in Mathematics; this team is unreliable.
Boundary assumptions
Claims about what belongs inside the problem: the issue is productivity, when the real problem may be workload; the issue is motivation, when the schedule is physically impossible.
Different assumption types require different tests. Facts need sources. Causal claims need distinguishing evidence. Behavioural assumptions need observation or agreement. Value assumptions need explicit human judgement.
The Load-Bearing Assumption
Not every assumption deserves equal attention. A load-bearing assumption is one whose failure would materially change the conclusion, plan or next action.
Suppose you are considering a second course because you believe it will improve your career prospects. The decision may contain many assumptions: the content is good, the certificate is recognised, the timetable is manageable, the skill is relevant and the opportunity cost is acceptable.
Which one is load-bearing? Perhaps the course is only attractive because you assume the certificate matters to target employers. If that assumption fails, the whole case changes. That deserves stronger verification than whether the lectures are live.
Ask SI: “Which assumption in this reasoning is doing the most work? If it were false, which conclusions would change?”
This directs verification toward the premise with the largest downstream effect.
Assumption Propagation
One unsupported assumption can propagate through a whole workflow.
Assume a school event begins at 8:00am. The family calendar is built around 8:00am. Transport is planned for 7:15am. Breakfast is moved earlier. Another appointment is declined. If the actual event begins at 9:00am, the original error did not stay local.
Ask SI: “If this assumption is wrong, what downstream parts of the plan inherit the error?” The answer identifies where early verification has the highest value.
The Assumption Register
Assumption: what is being treated as true.
Type: factual / causal / behavioural / resource / temporal / value / identity / boundary.
Why it matters: which decision depends on it.
Current evidence: what supports it.
Contrary evidence: what weakens it.
How to test: source, experiment or conversation.
Status: confirmed / strongly supported / plausible / unresolved / rejected.
Review trigger: when it should be checked again.
The register should stay small. Record assumptions whose failure would change something important. Do not turn every ordinary belief into bureaucracy.
Seven Ways to Challenge an Assumption
1. Ask for the opposite
Assume the opposite is true. What would the world look like? This is not evidence by itself. It expands the model.
2. Search for a counterexample
Give me a plausible case where this rule fails. A counterexample can show that a rule is too broad even when it remains useful in many cases.
3. Ask what evidence would change your mind
This converts a protected belief into a testable one.
4. Find the hidden variable
What factor could explain the same observation without my assumed cause?
5. Run a pre-mortem
Assume this plan failed. Which current assumption most plausibly caused the failure?
6. Remove the assumption
What plan would I choose if I were not allowed to rely on this assumption?
7. Test the smallest version
Design a reversible experiment that creates evidence before the large commitment.
The best challenge method depends on the assumption type. Factual assumptions often need a source more than a debate. Behavioural assumptions often need a conversation. Causal assumptions often need a distinguishing test.
Challenge Without Becoming Contrarian
A common failure is to turn assumption testing into reflexive disagreement. If the evidence strongly supports the current view, the assistant should say so. If an alternative explanation requires many unsupported steps, it should not be presented as equally credible merely because both sides exist.
Ask for the strongest reasonable challenge, not unlimited scepticism.
- Which assumption could realistically overturn this plan?
- What is the best-supported alternative interpretation?
- Which objection matters to the decision rather than merely being possible?
- Which claim is least supported relative to how much the conclusion depends on it?
The purpose is calibration, not argument for its own sake.
The Assumption Budget
Every plan contains uncertainty. You cannot verify everything. Allocate verification effort according to consequence, reversibility and dependency.
A low-consequence reversible choice can proceed with several unverified assumptions. A high-consequence irreversible action should rely on fewer unverified load-bearing assumptions.
- impact if false;
- ease of verification;
- likelihood of changing;
- number of downstream dependencies;
- consequence of acting before verification.
This converts assumption testing from endless doubt into targeted risk management.
Worked Example 1: My Child Needs More Tuition
A parent sees falling Mathematics marks and assumes more tuition hours are the solution.
The assumption stack may include lack of instructional time, weak prerequisites, insufficient guided practice, enough schedule capacity for more lessons and suitability of the current tuition method.
Diagnostic tasks show the student understands most concepts but repeatedly fails to choose a method on mixed questions and slows sharply under time pressure.
The original more-tuition assumption is not necessarily wrong, but the more precise need may be mixed retrieval and timed execution rather than more explanation.
A better intervention can now be tested before increasing total lesson hours.
Worked Example 2: The Expensive Product Will Last Longer
A buyer compares two devices and assumes the more expensive one is the safer long-term purchase.
Challenge the assumption by decomposing “last longer” into physical durability, battery replacement, software support, performance headroom, repair availability, warranty and actual workload.
The premium model may be stronger in some dimensions and identical in others. Ask SI to identify which longevity dimension matters for the expected use, then verify current manufacturer and repair information.
The final decision may still favour the expensive model, but the reasoning now rests on explicit factors instead of price as a proxy for durability.
Worked Example 3: The Team Is Overloaded
A project is late. The manager assumes the team lacks capacity.
Alternative explanations include unclear priorities, waiting for approval, duplicated work, unstable requirements, inconsistent inputs or one specialised bottleneck.
SI maps task flow and finds several tasks waiting on one approval process. Staff utilisation is high, but additional staff would not remove the approval delay.
The causal assumption changes from not enough people to decision latency may be the constraint. The manager can test that hypothesis by changing the approval workflow on one project.
Worked Example 4: I Need a New Productivity App
The user feels disorganised and assumes software is the missing component.
The evidence shows tasks are duplicated across three tools, priorities change daily and many commitments have no clear owner. A new app would probably add another container to the same structural problem.
The challenge suggests a smaller intervention: one current task list, one weekly review and explicit ownership. Only after the process stabilises should the user decide whether a tool limitation remains.
Sometimes the best assumption challenge removes technology from the solution.
Worked Example 5: The Course Certificate Will Help My Career
A professional considers an expensive course. The assumption contains several layers: the credential is recognised, target employers care about it, the course builds demonstrable skill and the opportunity cost is acceptable.
SI helps design verification: review target job descriptions, speak with people in relevant roles, inspect alumni outcomes cautiously and compare the course with a project that would produce portfolio evidence.
The result may show that the skill matters more than the certificate, or that the certificate is genuinely useful in that field. The assumption becomes a research question rather than a marketing claim.
Worked Example 6: If I Automate This, I Will Save Time
The user sees a repetitive report and assumes automation must create time leverage.
The baseline is 25 minutes manually. The automated workflow requires 8 minutes of data preparation, 3 minutes of generation, 9 minutes of verification and 5 minutes of allocated maintenance. The total is also 25 minutes.
The automation may still improve consistency, but the time-saving assumption is unsupported. A clearer next question is whether input preparation can be reduced or whether the report should be simplified or removed.
Assumptions about AI itself deserve the same scrutiny as assumptions about the original task.
Worked Example 7: This Person Does Not Respect My Time
Repeated late changes to plans can reasonably create frustration, but motive remains an assumption.
Separate observation—three same-day changes—from interpretation—lack of respect—from consequence—your schedule was disrupted.
Ask SI for alternative explanations that fit the observation without claiming they are true. Then prepare a direct question and a concrete coordination request.
The assumption challenge helps improve the conversation without pretending a model knows another person’s internal state.
The Default-Assumption Test
Defaults hide assumptions because nobody actively chose them. The meeting is one hour because meetings are usually one hour. Reports are weekly because they have always been weekly. The purchase must be new because repair was never considered.
Ask SI: “Which parts of this process are defaults rather than requirements?” Then test whether changing the default improves the outcome.
Challenging defaults often creates leverage because defaults can survive long after the reason for them disappears.
The Inherited-Assumption Test
Some assumptions came from someone else: a teacher, manager, family member, industry norm, old policy or previous version of the project.
“We have to do it this way” becomes: who requires it, where is the requirement documented and is it current?
The answer may confirm the constraint. Or it may reveal a convention that became mistaken for a rule.
Identity Assumptions
Identity assumptions influence which options you consider. “I am not technical.” “My child is not a language person.” “I am bad with money.” “Our team is slow.”
Ask SI to translate the identity claim into observable behaviour. “I am not technical” might become “I have limited experience configuring software and feel uncertain when error messages appear.” That description creates possible interventions.
The objective is not forced positivity. It is replacing a fixed label with a model that can be tested and changed.
Assumptions About Other People
Be especially careful with motive assumptions. A colleague does not reply. You assume they disagree. A child avoids homework. You assume laziness. A customer does not buy. You assume price is too high.
SI can generate many plausible motives, which may feel insightful but remain speculation. Use the system to separate observable behaviour from interpretations and design a question or experiment that creates evidence.
Do not let generated psychology become a substitute for direct communication.
Assumptions About the Future
Forecasts contain assumptions about what stays stable. A budget assumes future prices. A career plan assumes demand. A learning plan assumes future available time. A product purchase assumes future workload.
Instead of asking SI for one confident forecast, ask for scenario-dependent reasoning: what happens if workload stays the same, grows moderately or changes completely?
This makes temporal assumptions visible and supports flexible plans.
Assumption Half-Life
Some assumptions decay quickly. Opening hours, prices, software features, availability, policy details and deadlines may change often. Other assumptions are more stable, such as a long-term personal value.
Give volatile assumptions a shorter review cycle. Ask SI to mark which premises need current verification before each use.
A good workflow does not treat all context as equally permanent.
The Source Assumption
A common hidden premise is that the source itself is correct, current and relevant.
An old PDF can be authentic and stale. A current article can discuss the wrong jurisdiction. A product page can describe a different model. A course brochure can use broad marketing language rather than employment evidence.
Ask SI to challenge not only the claim but the source fit: date, authority, population, geography, version and purpose.
The Model Assumption
When using SI, one more assumption enters the system: that the model has understood the task and is using the intended evidence.
Ask it to restate the assumption register, identify which claims come from your input and which come from its own inference, and expose anything it cannot verify.
Fluent reasoning is not proof that the model’s premises match yours.
The Assumption Challenge Prompt Library
Load-bearing assumption
“Identify the assumption whose failure would change the most downstream conclusions.”
Counterexample
“Give me one realistic counterexample that shows where this rule stops working.”
Alternative cause
“What other cause fits the same observations, and what evidence would distinguish it?”
Opposite-world test
“Assume the opposite of my current belief. What evidence would I expect to see if that were true?”
Source test
“Which claims in this plan require an external source rather than reasoning?”
Default test
“Which constraints are true requirements and which are inherited defaults?”
Value test
“Which part of this conclusion depends on what I value rather than on factual evidence?”
Reversibility test
“If this assumption is uncertain, what smaller reversible action could create evidence?”
Future-stability test
“Which assumptions are likely to become stale first?”
Failure-propagation test
“If this premise is wrong, what parts of the workflow become wrong with it?”
The Assumption Audit
Decision: What am I about to choose?
Preferred option: What do I currently want?
Key assumptions: What must be true for that preference to make sense?
Load-bearing assumption: Which one matters most?
Evidence: What supports it?
Counterevidence: What weakens it?
Source freshness: Is the evidence current?
Alternative interpretation: What else fits?
Test: What can I verify or experiment with?
Action under uncertainty: What is the smallest sensible next step?
Review trigger: What new evidence would make me reconsider?
This audit is most useful before major purchases, course choices, career moves, new automations and other decisions where the narrative already feels convincing.
The Assumption Review After Action
After the outcome arrives, inspect which assumptions were accurate. Do not only ask whether the decision worked. A good decision can get unlucky. A weak decision can get lucky.
- Which assumption was most accurate?
- Which assumption was most wrong?
- Which error was visible beforehand?
- Which unknown genuinely could not have been known?
- Which verification step was worth the effort?
- Which assumption should become a standard checkpoint next time?
This turns experience into a better assumption model rather than a story about personal success or failure.
Common Failure Modes
Challenging everything
The conversation becomes paralysed. Focus on assumptions that materially change the action.
Treating alternatives as equal
A possible alternative is not automatically a well-supported alternative. Compare evidence.
Using SI as the counterevidence
A generated objection is a hypothesis, not external evidence. Verify important claims.
Protecting the preferred conclusion
The user accepts challenges only when they do not affect the preferred option. State in advance what evidence would change your mind.
Replacing facts with narratives
A plausible causal story is accepted because it feels coherent. Return to observations and distinguishing tests.
Ignoring value assumptions
A recommendation appears objective even though it depends on an unstated preference. Make the preference explicit.
Failing to update
The assumption is disproved but the downstream plan remains unchanged. Trace propagation and repair affected components.
Keeping stale assumptions
A once-correct premise remains active after the environment changes. Add freshness checks.
A 21-Day Assumption Training Plan
Days 1–3: identify assumptions in ordinary low-stakes decisions.
Days 4–6: classify assumptions by type.
Days 7–9: find the load-bearing assumption in one plan each day.
Days 10–12: run counterexample and opposite-world tests.
Days 13–15: replace motive assumptions with observable behaviour and questions.
Days 16–18: test defaults and inherited rules.
Days 19–20: review a past decision and identify which assumption most affected the result.
Day 21: save one compact assumption-audit template in your personal SI operating system.
Assumption Sensitivity: Which Premise Actually Flips the Decision?
Once several assumptions are visible, test their sensitivity. Sensitivity asks how much the final conclusion changes when one premise changes. Some assumptions can move a little without affecting the decision. Others sit directly on a threshold.
Suppose you are comparing two courses. Course A remains preferable whether the employer subsidy is 20% or 30%. That assumption is low sensitivity. But if Course A only makes sense when your employer covers at least half the fee, subsidy becomes a threshold assumption.
Ask SI: “Change one assumption at a time. Which changes leave the preferred option the same, and which changes flip the conclusion?”
This is more useful than simply ranking assumptions by how uncertain they are. A highly uncertain premise may barely matter. A moderately uncertain threshold premise may deserve immediate verification.
Sensitivity is not probability
Do not confuse “this assumption could flip the decision” with “this assumption is likely to be false”. Sensitivity describes impact. Probability describes confidence. High-impact, low-confidence assumptions are the strongest candidates for testing.
The Assumption Risk Matrix
A simple assumption matrix uses two dimensions: impact if wrong and confidence in the assumption.
High impact + low confidence: verify before major action.
High impact + high confidence: keep visible and monitor for change.
Low impact + low confidence: usually tolerate the uncertainty.
Low impact + high confidence: do not waste review effort.
For time-sensitive or changing information, add a third dimension: freshness. A once-high-confidence assumption can become stale.
This matrix gives your personal SI system a practical verification order instead of treating every uncertainty as equally urgent.
Falsifiability: What Result Would Show the Assumption Is Wrong?
A useful assumption has a failure condition. Ask what you would observe if the belief were wrong.
Assumption: “The student understands the concept but makes careless errors.”
Falsifying evidence: on an untimed fresh question, the student cannot explain why the method applies. That suggests the problem is deeper than carelessness.
Assumption: “Customers are not buying because the price is too high.”
Falsifying evidence: potential customers say the value proposition is unclear even when offered a lower price. Price may not be the primary barrier.
Assumption: “This automation saves time.”
Falsifying evidence: total input, verification, correction and maintenance effort equals or exceeds the manual baseline.
SI is useful here because it can help write concrete failure conditions before the result is known.
Avoid impossible falsification
If every possible outcome can be explained as supporting the same assumption, the belief has become protected. Ask whether you are evaluating evidence or defending a narrative.
The Evidence Ladder for Assumptions
Different assumptions need different evidence. Use the strongest practical evidence available for the consequence.
- current authoritative source for factual requirements;
- direct measurement or observation for performance;
- controlled or bounded experiment where practical;
- multiple relevant examples for behavioural patterns;
- qualified professional judgement for specialised high-stakes interpretation;
- high-quality secondary evidence for broader context;
- anecdote as a clue rather than proof;
- model-generated possibility as a hypothesis rather than evidence.
A generated counterargument can tell you what to investigate. It cannot, by itself, prove the assumption wrong.
Assumption Maps: Show How Beliefs Depend on Each Other
Some assumptions are nested.
You believe a course is worth taking because you assume the skill is valuable. You assume the skill is valuable because target roles appear to request it. You assume those role descriptions reflect actual hiring decisions. You assume the course will build the skill well enough to matter.
Draw the chain.
Target role demand → skill value → course relevance → completion feasibility → career benefit.
Now a weak link becomes visible. If the course does not actually build demonstrable skill, the downstream career-benefit assumption weakens even if demand is real.
Ask SI to map assumption-to-assumption dependencies, then identify which premise sits closest to the source evidence and which premises are mostly inference.
This prevents a long chain of “reasonable” assumptions from acquiring false certainty through repetition.
Assumption Clusters
Several assumptions may rise or fall together because they share one underlying premise.
Example: a parent believes a child needs more worksheets, more tuition and more study time. All three may rest on a shared assumption that the main problem is insufficient practice volume.
If a diagnostic shows the real issue is method selection, several surface assumptions weaken at once.
Cluster analysis helps you find the common root rather than challenging each surface belief separately.
The Pre-Mortem as an Assumption Detector
A pre-mortem asks you to imagine that the plan failed and then work backward to plausible causes already visible today.
Suppose you plan to automate monthly household renewals. Assume the system causes a missed renewal in six months.
Possible failure causes: source email format changed; the assistant treated a reminder as confirmation; a new family member’s responsibility was never agreed; the automation kept using an old date; the system created a reminder but nobody saw it.
Each failure reveals an assumption: source format will stay stable, reminders equal completion, ownership is clear, dates remain current, notifications are noticed.
The pre-mortem converts vague optimism into a set of testable premises before the workflow becomes difficult to unwind.
Do not confuse pre-mortem with prediction
The exercise generates plausible failure modes. It does not claim those failures are likely. Prioritise the ones with meaningful consequence and realistic pathways.
The Post-Mortem Without Hindsight Bias
After a bad outcome, people often rewrite history: “It was obvious.”
Use the original assumption register if you have one. Which uncertainty was visible at the time? Which warning was absent? Which assumption was reasonable based on the evidence then?
This protects the review from learning the wrong lesson. A decision can be reasonable and still produce a bad result under uncertainty.
The goal is better future assumptions, not punishment for imperfect prediction.
Worked Example 8: Family Scheduling
A family repeatedly misses preparation tasks before school activities. The first assumption is that everyone needs more reminders.
SI helps map the workflow. The notice reaches one parent. The event enters the family calendar. Materials are discussed in a chat. Transport is assumed rather than assigned.
The missed tasks occur mostly at the handoff between “event known” and “owner confirmed”.
The load-bearing assumption is not “people forget”. It is “someone will naturally take responsibility once the event is visible”.
The repair becomes explicit ownership and confirmation. More reminders may still help, but the deeper assumption has changed.
This example shows why behavioural and coordination assumptions should be challenged before adding notification volume.
Worked Example 9: Travel Planning
A traveller wants the cheapest itinerary and assumes a lower ticket price means a lower trip cost.
Challenge the assumption by including baggage, airport transfers, overnight accommodation, connection risk, arrival time and cancellation terms.
A cheaper flight may produce a higher total cost or a worse time trade-off.
SI can calculate total-trip scenarios, but current fares, baggage rules and entry requirements need current verification.
The assumption challenge changes the comparison from ticket price to total travel cost and practical fit.
Worked Example 10: Personal Finance
A household assumes discretionary spending is the reason savings have stalled.
The data show several irregular annual expenses were excluded from the monthly view. When insurance, school costs and repairs are normalised across the year, the apparent “overspending” picture changes.
The load-bearing assumption was that ordinary monthly spending represented the full cost structure.
SI can help classify and annualise expenses, but the source figures should come from actual records. The final financial decision remains human and may require qualified advice when consequential.
Worked Example 11: Health Administration
A person experiences fatigue and assumes the problem is sleep quality.
A general-purpose SI system should not convert that assumption into diagnosis. It can instead help organise a timeline: sleep, workload, symptoms, medication changes and questions for a healthcare professional.
The assumption is preserved as one hypothesis among several, not promoted to medical fact.
The useful action is better preparation for appropriate care, not autonomous treatment.
This is an example where challenging the assumption means reducing certainty rather than choosing a competing diagnosis.
Assumptions About Capability
People often assume they cannot do something because they have never done the final version.
“I cannot write a report.” “I cannot learn coding.” “My child cannot handle algebra.”
Break the capability into components. Which part fails? Vocabulary? Tool use? Method selection? Planning? Practice under time?
SI can help design a diagnostic that distinguishes missing knowledge from missing experience.
A capability assumption becomes more useful when expressed as a current performance statement rather than a permanent identity.
Assumptions About Capacity
Capacity is not only the number of free hours.
A schedule may assume that every open hour is usable. It may ignore transition, attention, recovery, travel and the unpredictability of family life.
Ask SI to challenge calendar availability with realistic operating constraints.
The plan should distinguish free time from usable capacity.
This matters because many “productivity failures” are actually capacity-model failures.
Assumptions About Motivation
When a task is repeatedly avoided, motivation is an easy explanation.
Alternative causes include unclear next action, excessive difficulty, missing prerequisite, low expected value, competing obligations, anxiety, poor environment or simple exhaustion.
Ask SI to generate observable tests instead of personality conclusions.
A motivation assumption should be the beginning of diagnosis, not the end.
Assumptions Inside AI Workflows
SI workflows create their own assumption layer.
- the model has access to the intended source;
- the source is current;
- the model interpreted the source correctly;
- the user supplied all important constraints;
- the output format preserves meaning;
- the connected action targets the intended destination;
- the system can tell whether the action completed;
- the user will notice a notification;
- a successful previous run predicts the current run.
These assumptions should be part of workflow testing.
Tool-access assumption
A filename mentioned in conversation does not prove the file was opened. Ask the system to identify what source it actually used.
Completion assumption
A message drafted is not a message sent. A calendar proposal is not an event created. Verify external action in the destination system.
Memory assumption
Do not assume the system remembers the latest project state correctly. Important current work should point to a record you can inspect.
The AI Assumption Preflight
Before a consequential AI-assisted action, ask:
- Which facts come directly from my input?
- Which facts came from external sources?
- Which statements are model inference?
- Which source may be stale?
- Which constraint could be missing?
- What permission is required?
- What completion evidence will exist?
- What would make the action inappropriate?
This is assumption control applied to the intelligence layer itself.
The Assumption Sandbox
When a decision depends on uncertain premises, create a sandbox rather than changing the live system.
A family can test a new weekly schedule for one week.
A professional can run a new reporting workflow in parallel with the old process.
A learner can test a new revision method on one topic before using it across the syllabus.
A small business can test one offer before building full infrastructure.
The sandbox reduces the cost of being wrong and turns the assumption into an experiment.
Ask SI to define the scope, evidence and stop condition before the sandbox begins.
The Assumption Freeze
Sometimes an assumption has been verified enough for the current stage and should stop consuming attention.
Mark it temporarily frozen: “Use this premise unless new evidence appears.”
This prevents long conversations from repeatedly reopening settled low-risk questions.
A freeze is not permanent truth. It is an attention-management choice with a review trigger.
The Assumption Reset
When the environment changes materially, reset assumptions instead of patching the old model.
A new school term, new job, relocation, health event, major tool change or changed family responsibility can invalidate several assumptions simultaneously.
Ask: “Which assumptions from the previous context should no longer be carried forward automatically?”
This prevents legacy context from governing a new operating state.
A 30-Day Assumption Control Practice
Week 1 — Visibility: capture one assumption each day from a real decision. Classify its type and write why it matters.
Week 2 — Challenge: find one counterexample, alternative cause or opposite-world scenario for each assumption. Do not treat generated alternatives as evidence.
Week 3 — Testing: verify high-impact assumptions through sources, direct questions or small experiments. Record what changed.
Week 4 — Integration: add assumption checks to one recurring workflow, one learning process and one personal decision. Remove checks that create more overhead than value.
At the end of the month, keep only the assumption practices that changed real decisions or prevented meaningful errors.
The objective is not maximum scepticism. It is better calibration.
The Assumption Control Tower
For a complex project, create one compact view:
Current decision: what is being chosen.
Load-bearing assumptions: top three.
Confidence: support level.
Freshness: last verified.
Downstream dependencies: what breaks if wrong.
Test: next evidence step.
Owner: who confirms the premise.
Status: active / frozen / rejected / replaced.
The control tower is useful when several people or workflows rely on the same assumptions. It prevents a rejected premise from continuing to circulate in downstream work.
When Not to Challenge an Assumption Further
Stop challenging when the premise is sufficiently verified for the consequence, the action is easily reversible, further evidence is expensive and unlikely to change the route, or the next experiment will teach more than more analysis.
Also stop when the “challenge” has become speculative noise. A model can always invent another possibility. Possibility alone is not a reason to delay action.
The standard is not certainty. It is enough justified confidence for the next consequence.
The Assumption Audit: From Hidden Belief to Testable Claim
An assumption becomes manageable when you can write it as a sentence that could, in principle, be wrong. “This course is worth it” is too compressed. “This course will give me a capability that the roles I want actually value, and I will have enough time to practise that capability” is more useful because its parts can be checked.
Use six fields: assumption, current evidence, counterevidence, decision impact, cheapest reliable test and review trigger.
Assumption: the extra course will improve career options.
Current evidence: several target roles mention the skill.
Counterevidence: current workload already limits practice time.
Decision impact: if practice is insufficient, the course may create a certificate without usable capability.
Test: compare target roles, speak with practitioners and estimate realistic weekly practice capacity.
Review trigger: revisit after the current course ends or when a target opportunity appears.
This format is powerful because it prevents an assumption from remaining vague enough to survive every outcome.
The Evidence Ladder for Assumptions
Not all support is equal. When SI challenges an assumption, ask what kind of evidence currently supports it.
- current authoritative source;
- direct measurement or observation;
- repeated personal experience under similar conditions;
- relevant expert judgement;
- multiple independent reports;
- single anecdote;
- analogy to a different situation;
- model-generated inference;
- intuition with no identified source.
The purpose is not to dismiss intuition. Intuition can be an efficient summary of experience. The purpose is to know when intuition is carrying a consequential decision without enough external support.
A strong assumption challenge therefore asks not only “What evidence supports this?” but “What kind of evidence is it, and how directly does it apply to this situation?”
The Opposite-World Test
Ask SI to imagine that the opposite of your main assumption is true and describe what you would expect to observe.
Assumption: a student is weak because they do not understand the concept.
Opposite world: the student understands the concept but performs poorly under time pressure.
Different predictions follow. In the first world, untimed diagnostic work should also be weak. In the second, untimed work may be accurate while timed mixed work deteriorates.
Now the disagreement can be tested.
This method is useful because it converts two stories into competing predictions. It moves the conversation away from “which story sounds more plausible?” and toward “what evidence would look different if each story were true?”
The Removal Test
Remove the assumption mentally and see whether the conclusion still stands.
Plan: buy a more powerful laptop.
Assumption: future work will require heavy local processing.
Remove that assumption. Does the current workload justify the upgrade on its own? If not, the purchase is partly an insurance decision against a future possibility.
The removal test is especially useful when an assumption has become so familiar that it is embedded in the plan. By deleting it temporarily, SI can show how much of the conclusion depends on it.
The Substitution Test
Replace one assumption with a plausible alternative and see how the plan changes.
Assumption: the project delay comes from insufficient staff.
Alternative: the delay comes from unclear approval and rework.
If the second assumption is true, hiring more people may increase coordination load without removing the blocker.
Ask SI to produce a different action plan under each assumption. The difference reveals which evidence matters most.
Causal Assumptions Need Mechanisms
“A caused B” is stronger than “A and B occurred together”.
When your plan depends on a causal assumption, ask SI to state the proposed mechanism.
Example: “More tuition will improve Mathematics results.”
Possible mechanism: extra teaching identifies a missing prerequisite, targeted practice repairs it and feedback prevents recurrence.
Alternative mechanism: more tuition simply increases question volume without addressing the specific error, adding fatigue.
The intervention should test the mechanism.
If the suspected bottleneck is bracket expansion, test bracket expansion. If performance improves after the targeted repair and transfers to changed questions, the causal model gains support.
Do not use “the result improved after the change” as automatic proof that the change was the only cause.
Behavioural Assumptions Need Consent and Observation
Plans often assume how another person will behave.
“My colleague will review this tonight.”
“My child will study for an hour after training.”
“My partner can handle transport.”
These assumptions are not resolved by AI inference. They are resolved by asking the person, observing actual capacity or redesigning the plan so it does not depend on unconfirmed cooperation.
A useful SI role is to highlight behavioural dependencies and convert them into questions: “Which parts of this plan require another person’s agreement?”
This prevents a generated schedule from quietly allocating other people’s time.
Resource Assumptions Need Capacity Checks
A plan may be logically sound but impossible because it assumes time, money, energy, attention or access that is not actually available.
Ask SI to list every resource the plan consumes and identify which one is closest to its limit.
For a study plan, the bottleneck may be usable evening attention rather than clock time. For a household project, the bottleneck may be one person’s availability. For a work project, it may be approval capacity.
Challenge the resource assumption before optimising the schedule.
Temporal Assumptions: “This Will Still Be True Later”
Some assumptions are about persistence.
A price will remain similar.
A deadline will not move.
A software capability will remain available.
A student’s schedule will stay stable.
A role will still require the same skill next year.
Ask SI to mark assumptions whose validity decays with time and attach a freshness check.
Time-sensitive assumptions should be reverified near the action, not only when the plan was first designed.
Boundary Assumptions: What Is Inside the Problem?
Every analysis assumes a boundary.
A budget may exclude annual costs. A study plan may ignore sleep. A project plan may ignore legal approval. A career comparison may ignore commute or family constraints.
Ask SI: “What important factor might be outside the boundary I have drawn?”
Then decide whether it belongs inside. The goal is not to include everything. It is to prevent a material external factor from invalidating the internal optimisation.
Worked Example: Challenging a Parenting Assumption
A parent says, “My child is lazy because homework takes too long and they keep delaying.”
SI separates the assumption from observations.
Observations: homework often begins late; Mathematics takes much longer than English; the child asks for help repeatedly on algebra; work after 9pm is especially slow.
Assumption: laziness is the main cause.
Alternative explanations: specific skill gap, fatigue, unclear task sequence, avoidance caused by repeated failure, or too much total homework.
Diagnostic action: compare start time and completion on one easier subject and one difficult algebra task earlier in the day. Check whether the delay changes when the specific Mathematics gap is supported.
If the pattern changes, the identity label “lazy” loses explanatory power.
The parent may still need routines and boundaries, but the intervention becomes better matched to the evidence.
Worked Example: Challenging a Work Assumption
A manager believes a weekly report is necessary because “leadership needs visibility”.
Assumption chain: leadership reads the report → report affects decisions → weekly frequency is necessary → current format is useful.
SI proposes a test: identify decisions made from the report over the last six weeks and ask the recipients which sections they actually use.
The result shows that only one metric and one risk section drive discussion. The rest is archival.
The better solution may be a shorter dashboard with a monthly full report.
The assumption challenge does not merely criticise the report. It finds which part of the stated purpose is real.
Worked Example: Challenging a Travel Assumption
A family assumes staying in the city centre is always more convenient.
SI decomposes convenience into travel time to planned activities, transport cost, room size, noise, food access and airport transfer.
If most activities are outside the centre, the central hotel may increase daily travel despite the location label.
The assumption becomes measurable against the actual itinerary.
This is a useful pattern: replace prestige or category labels with the underlying mechanism that is supposed to create value.
Worked Example: Challenging an AI Assumption
A user assumes that because SI produced several good answers, a workflow can now be automated end to end.
Hidden assumptions: error rate will remain low under varied inputs; important errors will be detectable; connected permissions will behave as expected; retries will not duplicate actions; the user will still notice exceptions.
A proper challenge therefore stress-tests missing data, conflicting data, unusual amounts, new recipients and uncertain completion before autonomy expands.
Capability demonstrated in conversation is not the same as reliability demonstrated under workflow variation.
Assumption Drift
An assumption can be reasonable when created and wrong later.
This is assumption drift.
A course schedule changes. A family commitment changes. A market price moves. A software integration gains new limitations. A child improves and no longer needs the same scaffold.
Keep review triggers for assumptions that age quickly.
For example: “Recheck course timing before enrolment”, “recheck product price before purchase”, “retest student independence after three weeks”, “review permission scope after the new app connection”.
The goal is not to maintain a giant assumption database. Track the assumptions whose failure could materially change the plan.
Assumption Propagation Through a System
One wrong assumption can spread.
An unverified deadline enters the project brief.
The weekly plan uses it.
The calendar is changed.
A family commitment is moved.
A reminder is created.
Several later outputs now look consistent because they all inherited the same wrong source.
This is why provenance matters. Ask where the assumption entered the system and correct the source record first. Then identify downstream records that require update.
Do not merely patch the final output.
The Assumption Register in a Personal SI Operating System
For important projects, keep only a few active assumptions beside the project card.
Assumption: deadline remains 15 October.
Source: course portal checked 30 September.
Impact if false: study plan changes.
Review trigger: new course announcement.
Assumption: two evening study windows remain available.
Source: current calendar and family agreement.
Impact if false: second course should be delayed.
Review trigger: work schedule change.
The register makes the fragile parts of the plan visible without over-documenting everything.
Connect this to your personal SI operating system so current assumptions, decisions and source records remain separate.
Group Assumption Audits
Before a family, team or partnership makes a significant decision, ask each person to write what they believe must be true for the plan to work.
Compare the lists.
One person may assume budget is flexible. Another assumes it is fixed. One assumes the deadline cannot move. Another assumes the deadline is only an internal target.
SI can organise the statements into confirmed facts, assumptions, disagreements and questions to verify.
The people still resolve the disagreements. The assistant makes the hidden differences visible.
The Red-Team Protocol
Use a bounded red-team pass after a plan is mostly formed.
Prompt: “Identify only the three assumptions that create the largest downside if wrong. For each, show the failure path, evidence currently supporting it, evidence missing and one mitigation or test.”
Limit the number.
Unlimited red-teaming can produce an endless catalogue of hypothetical failure and make every plan look impossible.
The purpose is material robustness, not maximal scepticism.
The Steelman Protocol
Challenge is stronger when the alternative is represented fairly.
If you prefer Option A, ask SI to build the strongest reasonable case for Option B using the same evidence standard.
Then compare the two cases side by side.
A weak alternative proves nothing. A strong alternative reveals which trade-off is genuinely difficult.
Steelman the option, not the person. Avoid turning the exercise into speculation about motives.
The Assumption Threshold
Decide how much verification an assumption deserves by combining decision impact and test cost.
High-impact + cheap test: verify immediately.
High-impact + expensive test: seek a proxy, expert input or reversible experiment.
Low-impact + cheap test: verify if convenient.
Low-impact + expensive test: leave uncertain unless circumstances change.
This prevents assumption auditing from becoming more expensive than the decision itself.
A Copyable Assumption Audit Card
Conclusion or plan: [what I currently think]
Load-bearing assumption: [what must be true]
Type: factual / causal / behavioural / resource / temporal / boundary
Current support: [evidence]
Counterevidence: [evidence against]
Provenance: [where the belief came from]
Impact if false: [what changes]
Cheapest reliable test: [source / person / experiment / expert]
Status: confirmed / supported / plausible / speculative / unknown
Review trigger: [when to revisit]
The Assumption Exit Test
Before acting on an important plan, ask:
- Have I identified the one or two assumptions that most affect the decision?
- Are they factual assumptions or values?
- Do I know where the assumptions came from?
- Is there contradictory evidence I am ignoring?
- Can I test a high-impact assumption cheaply before committing?
- Does the plan depend on another person’s unconfirmed behaviour?
- Will a time-sensitive assumption be rechecked before action?
- If the assumption fails, is the plan recoverable?
- Have I challenged enough to improve the decision without creating paralysis?
If the answers are clear, move to action and review rather than continuing to generate more objections.
Assumption Challenge as a Transferable Skill
The method transfers because the structure remains the same even when the evidence changes.
In education, assumptions are tested with student work.
In shopping, with specifications, prices and real requirements.
In work, with project data, stakeholder confirmation and outcomes.
In relationships, with observable behaviour and direct conversation.
In health, with appropriate professional assessment rather than model speculation.
In AI workflows, with stress tests, permissions, error logs and completion evidence.
The evidence source changes. The reasoning discipline remains: make the premise visible, test what matters and update when reality disagrees.
Assumption Chains: Find the Root Premise
A decision may contain a chain of assumptions rather than one independent belief. If you challenge only the last link, the deeper structure remains untouched.
Example: “I need a second course because I need a stronger CV because employers want more credentials because my current experience is not enough.” Each step is an assumption. The root premise may be that employers value the additional credential more than stronger evidence from actual work.
Ask SI to rewrite the reasoning as an if-then chain. Then test the earliest premise whose failure collapses several later steps.
This is assumption root-cause analysis. It prevents spending time verifying downstream claims that only matter if an earlier premise survives.
Root premise versus surface premise
A surface premise sits close to the final action. A root premise explains why the action seemed necessary in the first place. Both matter, but the root premise often creates larger leverage when corrected.
Do not dig forever
Stop once you reach a premise that can be tested and sits within the decision’s practical boundary. Infinite philosophical depth does not improve ordinary life decisions.
Conflicting Assumptions Between People
Many coordination problems are not disagreements about goals. They are disagreements about assumptions.
A parent assumes the student has enough time after school. The student assumes homework will take two hours. A tutor assumes the school topic has already been taught. The school schedule assumes the student can complete revision over the weekend.
All four plans can look reasonable in isolation and collide in the same week.
Use SI to create a shared assumption table: who believes what, what evidence supports it, what source could resolve it and which assumption actually needs agreement rather than proof.
This is especially valuable because people often argue over solutions before discovering they are working from different models of the same situation.
Assumptions Hidden Inside Numbers
Numbers can make an assumption look objective.
A forecast says the project will take six weeks. That number may assume stable staffing, no approval delay and no scope change.
A budget says the household can save S$800 each month. That may assume no irregular annual costs.
A study target says ten hours a week. That may assume every open hour is equally usable.
Ask SI to expose the assumptions behind each important number: base, time period, unit, source, range and conditions.
Average assumptions
An average can hide variation. A commute that averages 30 minutes may routinely take 50 minutes on the day that matters. A student’s average score can hide one prerequisite topic causing repeated failure.
Percentage assumptions
Percentages require a base. “Costs increased 20%” means little without knowing the starting figure and period.
Forecast assumptions
Forecasts combine evidence with a model of what continues. Label the continuation assumptions instead of treating the projection as a fact.
Assumptions About Evidence Quality
A claim can be supported by evidence and still be weakly supported for the current decision.
Evidence may be old, from another country, from another age group, based on a different product version or generated under conditions that do not match yours.
Ask SI to challenge evidence transfer: “What has to be true for this evidence to apply to my situation?”
This turns “I found a source” into the stronger question “Does this source actually support the inference I am making?”
Assumptions in Plans
Every plan assumes a world in which the plan remains feasible.
Common planning assumptions include task duration, availability, travel time, response time, energy, cooperation, resource access and the absence of interruptions.
Use SI to produce an assumption layer underneath the schedule.
Plan item: finish report Wednesday evening.
Assumptions: source data arrives by Tuesday; Wednesday meeting ends on time; two hours of focused attention are available; no additional review is required.
Now the plan has early-warning signals. If source data does not arrive Tuesday, the problem becomes visible before Wednesday evening fails.
A strong plan therefore contains not only tasks but the assumptions that make those tasks possible.
Assumptions in Forecasts
A forecast should be written conditionally.
Instead of “this project will finish by 15 October”, write “15 October remains feasible if approval arrives by 8 October, no critical defects appear and the current scope remains fixed.”
SI can help turn point predictions into conditional forecasts.
This is useful because a changing assumption can then update the forecast before the outcome arrives.
The user gains an early-warning model rather than a date that slowly becomes false.
Domain Template: Learning
Observed difficulty: what the learner cannot do independently.
Current assumption: why we think the difficulty exists.
Evidence: marked work, explanation, diagnostic attempt.
Alternative explanation: another error source that fits.
Test: one question or task that distinguishes them.
Repair: targeted intervention only after diagnosis.
Retest: changed problem without assistance.
This template prevents labels such as “weak at algebra” from becoming unquestioned premises.
Domain Template: Career
Desired change: what must improve in the working life.
Assumption: why the proposed role or course will produce that change.
Evidence: role descriptions, conversations, actual project experience.
Unknown: what cannot yet be known.
Experiment: a reversible project, shadowing opportunity or responsibility that tests fit.
Review: update the career model from real evidence rather than imagined future identity.
Domain Template: Major Purchase
Need: the problem the purchase must solve.
Assumption: which feature, brand or price characteristic is believed to solve it.
Evidence: current specification, warranty, repair, independent testing.
Alternative: repair, delay, cheaper option or different solution.
Total cost: purchase + recurring + maintenance.
Threshold: the condition that justifies buying now.
Domain Template: Family Coordination
Confirmed facts: dates, requirements and existing commitments.
Assumptions: who is available, who will own the task, how long preparation takes.
Agreement required: items that cannot be inferred.
Source: original notice or calendar.
Escalation: what happens when the requirement is unclear.
Ten Common Assumption Traps
The familiar-is-true trap
A belief feels reliable because it has been repeated, not because it has been checked.
The price-equals-quality trap
Higher cost is used as a proxy for durability, expertise or fit without examining the mechanism.
The time-equals-effort trap
Free calendar time is assumed to equal usable attention.
The activity-equals-progress trap
More lessons, reports, meetings or prompts are assumed to improve the outcome.
The last-result trap
One recent success or failure is treated as representative.
The motive trap
Observed behaviour is converted into a confident explanation of another person’s intention.
The stability trap
A price, rule, deadline, permission or preference is assumed to remain current.
The tool-capability trap
Because an AI can perform an action, it is assumed that delegation is appropriate.
The automation-saves-time trap
Generation speed is mistaken for total workflow speed.
The identity trap
A current difficulty is turned into a permanent trait.
Recognising a trap does not automatically tell you the correct answer. It tells you where evidence or a test is needed.
The Assumption Exit Test
Before closing an assumption review, ask:
- Have I identified the premise most capable of changing the decision?
- Is its evidence current and relevant?
- Have I confused possibility with probability?
- Have I separated values from facts?
- Have I checked whether another person’s agreement is being assumed?
- Do I know what would falsify the belief?
- Can a smaller experiment create evidence?
- If the assumption is wrong, do I know what downstream work needs repair?
- Is further checking likely to change the next action?
If the last answer is no and the next step is reversible, the assumption review is probably complete enough for action.
The purpose of assumption testing is not to eliminate uncertainty. It is to prevent invisible uncertainty from controlling the decision.
The final discipline is to record what changed after the assumption test. If the evidence strengthens the original view, keep the assumption but upgrade its status. If the evidence weakens it, change the plan rather than defending the earlier story. If the result is inconclusive, preserve uncertainty and choose a reversible next step. The point of assumption challenge is not to win an argument with yourself. It is to make the plan more faithful to reality.
Frequently Asked Questions
Why should I challenge assumptions with SI?
Because SI can quickly externalise hidden premises, generate reasonable alternatives and help identify what evidence would change the conclusion. The user still verifies facts and makes the final judgement.
Are assumptions bad?
No. Every practical decision uses assumptions. The problem is relying heavily on an assumption without recognising or pricing the risk.
Which assumptions should I test first?
Start with load-bearing assumptions whose failure would materially change the plan, especially when the action is consequential or hard to reverse.
How do I know whether an assumption is load-bearing?
Remove it mentally or assume the opposite. If the preferred decision changes, it is probably important.
Should I ask SI to argue against me?
Ask for the strongest reasonable challenge, not automatic opposition. Useful challenge is evidence-sensitive.
What is the difference between an assumption and an unknown?
An unknown is information you recognise you do not have. An assumption is a belief currently filling part of that gap or supporting the model.
How do I challenge causal assumptions?
Generate alternative causes and identify observations or interventions that would distinguish them.
How do I challenge assumptions about another person?
Separate observable behaviour from motive, generate alternative explanations cautiously and ask the person directly where appropriate.
What if the assumption cannot be verified now?
Use scenarios, reversibility and monitoring. Do not convert uncertainty into false certainty simply to complete the plan.
How many assumptions should I track?
Only the assumptions whose failure would materially change important decisions or workflows.
What comes next?
The next guide compares options systematically once the assumptions behind them are visible.
Helpful Reading
- How to Leverage Your Life with Super Intelligence
- How to Think More Clearly with Super Intelligence
- How to Use Super Intelligence to Break Down Complicated Problems
- How to Turn Super Intelligence from a Chatbot into a Thinking Partner
- NIST — AI Risk Management Framework
Make the Hidden Premise Visible
An assumption is most powerful when it is invisible. Once it is visible, it can be sourced, tested, bounded, updated or deliberately accepted.
Super Intelligence is useful here because it can expose structure quickly. But the generated challenge is not the evidence. The evidence still comes from sources, observation, experiments and people.
Do not challenge everything. Challenge the assumptions carrying the most weight.
Continue with How to Compare Options with Super Intelligence to turn challenged assumptions into a structured comparison.
