
How to use Super Intelligence to find gaps in your knowledge begins by distinguishing “I do not know this” from several other states that feel similar. You may know the idea but fail to retrieve it. You may remember the rule but not recognise when to use it. You may understand a familiar example but fail when the surface changes. You may be confident and wrong. You may be correct but unable to explain why.
A useful SI learning system does not label all of those states as “knowledge gaps”. It identifies which layer is missing and chooses the smallest repair.
In the eduKateSG life series, Super Intelligence, or SI, is our editorial name for practical AI assistance. This article follows How to Turn a New Subject into a Learning Roadmap with Super Intelligence. A roadmap tells you where to go. Gap detection tells you what is blocking the route right now.
The central rule is: do not repair a gap until you know what kind of gap you are looking at.
The Nine Knowledge-Gap Types
1. Fact gap
A needed piece of information is missing. Example: the learner does not remember the formula, definition, vocabulary item or historical date.
2. Concept gap
The learner knows words but not the relationship. Example: they can recite “photosynthesis uses carbon dioxide” but do not understand how carbon becomes part of glucose and plant mass.
3. Procedure gap
The learner understands the idea but cannot execute the method reliably.
4. Recognition gap
The method is known but the learner does not recognise when to use it.
5. Retrieval gap
The knowledge is available with notes or prompts but not independently.
6. Transfer gap
Performance succeeds only on familiar examples or contexts.
7. Integration gap
Individual skills work separately but fail when several must be combined.
8. Boundary gap
The learner knows the rule but not where it stops applying, leading to overgeneralisation.
9. Metacognitive gap
The learner cannot accurately judge what they know, where the error began or how much help they used.
The nine types explain why repeating the same explanation often fails. The error may live somewhere else.
Knowledge Gap Versus Performance Gap
A poor result does not automatically mean missing knowledge.
The learner may know the content but perform poorly under time pressure, fatigue, distraction, long working sequences or unfamiliar formatting.
That is a performance gap.
SI should compare supported, untimed and realistic performance before concluding that knowledge is missing.
Example: a student solves equations accurately untimed but loses marks in a mixed paper. The gap may be method selection or speed rather than equation knowledge.
Example: a professional can explain a process but misses steps during a rushed handoff. The gap may be workflow support rather than conceptual knowledge.
The Gap-Detection Loop
Use six stages: Elicit → Expose → Classify → Prioritise → Repair → Retest.
Elicit
Make the learner produce something: answer, explanation, classification, code, calculation, plan or example.
Expose
Change the surface, remove the notes, mix the skills or ask for mechanism.
Classify
Identify the gap type and first unstable step.
Prioritise
Ask whether the gap blocks other skills, causes major performance loss or is needed soon.
Repair
Use the smallest intervention that fits the gap.
Retest
Use changed evidence to confirm that the gap actually closed.
The First Unstable Step
The final wrong answer is often too late in the chain.
Ask SI to locate where the reasoning first diverged.
Mathematics: the learner chose the correct equation method but copied a negative sign incorrectly.
Writing: the evidence is accurate but the topic sentence does not answer the question.
Coding: the function logic is correct but the return value is not captured.
Science: the key terms are remembered but the causal direction is reversed.
Repair the first unstable step and retest the whole sequence.
The Blank-Page Test
Ask the learner to write what they know without notes.
This exposes retrieval and structural gaps quickly.
Prompt: “Without looking anything up, write the five most important ideas in this topic and show how they connect.”
Then compare with the roadmap or source.
Missing concepts indicate possible fact or concept gaps. Correct concepts with weak connections suggest integration gaps.
The blank-page test is especially useful before revision because it tells the learner what not to spend time rereading.
The Explain-Back Test
Ask the learner to explain the concept to a beginner without copying the original wording.
SI then checks definition, mechanism, example and boundary.
If the learner gives a definition but no mechanism, the concept may be shallow.
If the explanation works only through an analogy, ask for the formal relationship.
If the explanation omits a boundary, test a counterexample.
The Contrast Test
Show two similar cases where the correct response differs.
Can the learner identify the decisive difference?
Mathematics: two equations suited to different methods.
Vocabulary: two near-synonyms with different register.
Science: two phenomena that look similar but use different mechanisms.
Coding: two bugs with the same symptom but different causes.
Failure reveals a discrimination or recognition gap rather than a missing rule.
The Counterexample Test
Ask for a case where the rule does not apply.
If the learner cannot produce or recognise one, they may have a boundary gap.
Example: “Favourable outcomes divided by total outcomes” works for equally likely outcomes. Ask when that shortcut fails.
Example: “More practice improves performance” may fail when practice repeats the wrong method.
Counterexamples test whether the learner understands conditions, not only slogans.
The Confidence Calibration Test
Before answering, ask the learner to predict confidence.
Then compare confidence with performance.
High confidence + low performance signals hidden gaps or misconceptions.
Low confidence + high performance signals under-confidence or weak metacognitive calibration.
SI can track these mismatches across topics and prioritise areas where the learner’s self-assessment is unreliable.
The goal is not to make confidence perfectly numerical. It is to reduce surprise about what is and is not stable.
The “What Would Change Your Mind?” Test
For conceptual or decision knowledge, ask what evidence would make the learner revise their belief.
If the learner cannot name any possible disconfirming evidence, the model may be rigid rather than understood.
This is useful in science, history, research, professional judgement and personal decision-making.
Unknown Unknowns
The hardest gaps are the ones the learner does not know exist.
SI can help expose them through several routes:
- ask for common beginner blind spots;
- compare the learner’s concept map with an authoritative outline;
- generate prerequisite checks;
- ask what experts usually consider that beginners omit;
- present unusual but valid boundary cases;
- ask for failure modes;
- compare two neighbouring concepts;
- use a project that integrates several skills.
Unknown-unknown discovery should be bounded by the learner’s current goal. Do not expand the subject endlessly.
The Expert-Outline Comparison
Ask SI to compare the learner’s current map with a trusted syllabus, textbook or expert framework.
Classify differences as:
- missing essential;
- optional extension;
- different terminology;
- possible misconception;
- out of scope for the current goal.
This prevents every missing advanced topic from being labelled a gap.
Gap Severity
Not all gaps deserve immediate repair.
Rate gaps by leverage, consequence, frequency and time-to-need.
High leverage
Blocks several downstream skills.
High consequence
Creates large errors, safety issues or decision risk.
High frequency
Appears repeatedly across tasks.
High time pressure
Needed soon for an exam, project or real task.
SI can rank gaps provisionally, but the learner or teacher should confirm priorities.
The Gap Portfolio
Keep a small active list rather than every weakness ever observed.
Active repair: high-leverage gap being worked on.
Watch: possible gap needing more evidence.
Maintain: repaired gap needing spaced checks.
Closed: stable across transfer and delay.
Out of scope: real gap but irrelevant to current goal.
The portfolio prevents a learner from feeling permanently deficient because the system remembers every past error.
Worked Example: Mathematics
Symptom: student loses marks on simultaneous equations.
Gap hunt: direct equation solving is accurate. Negative-number arithmetic is accurate. Method choice is inconsistent when equations are not arranged neatly.
Classification: recognition gap.
Repair: contrast pairs—one ideal for elimination, one for substitution.
Retest: mixed set with no method labels.
Result: method selection improves; one sign error remains as an execution watch item.
The roadmap does not reteach simultaneous equations from the beginning.
Worked Example: Vocabulary
Symptom: learner recognises sophisticated words but does not use them in essays.
Gap hunt: definition recognition strong; free recall moderate; sentence production weak; register sometimes incorrect.
Classification: retrieval + transfer + boundary gap.
Repair: active recall, contrast near-synonyms, sentence production and paragraph transfer.
Retest: new topic without word bank.
Worked Example: Coding
Symptom: learner follows tutorials but cannot begin a new program.
Gap hunt: syntax recall acceptable; can read existing code; struggles to decompose a requirement into functions and steps.
Classification: design/recognition integration gap.
Repair: requirement → inputs → outputs → subproblems before writing code.
Retest: different small program with no template.
The gap was not “more Python syntax”.
Worked Example: Science
Symptom: learner memorises definitions but loses marks in open-ended questions.
Gap hunt: keywords remembered, mechanism explanations incomplete, causal chain breaks.
Classification: concept + integration gap.
Repair: cause → mechanism → effect explanation frames, then unfamiliar scenarios.
Retest: novel context using the same scientific concept.
Worked Example: Professional Knowledge
Symptom: a new manager knows project terminology but misses risks during planning.
Gap hunt: definitions stable, but no habit of mapping dependency, trigger, owner and contingency.
Classification: application/integration gap.
Repair: analyse several real project cases and compare missed risk categories.
Retest: new project plan without checklist, then compare with structured review.
The Gap Register
For each active gap, record:
Observed failure: what happened.
Gap hypothesis: current classification.
Evidence: representative tasks.
Priority: why it matters.
Repair: smallest intervention.
Retest: changed task.
Status: active, watch, maintain or closed.
This register is compact enough to feed into a personal SI tutor or roadmap.
False Gaps
A false gap is a weakness the system thinks exists but does not.
Causes include ambiguous questions, wrong marking, one-off slips, unfamiliar formatting, bad source material or model error.
Do not update the learner model from one surprising failure.
Use a second task targeting the same skill.
If performance is stable, downgrade the gap hypothesis.
False Strengths
A false strength is a skill that appears stable because the environment provides cues.
Examples: chapter heading names the method, notes remain visible, SI hints early, practice items are nearly identical, answers are multiple choice.
Remove cues and retest.
False strengths are one reason transfer testing is essential.
Knowledge Drift
A gap can reopen after time.
Use brief delayed checks for important skills.
If the learner once performed independently but later fails, classify it as drift rather than “never learned”.
Repair can often be shorter because the underlying structure may still exist.
Stale Knowledge
Some professional and technical knowledge becomes outdated even if memory remains perfect.
Software APIs change. Policies change. Standards change. Prices and requirements change.
A personal SI system should distinguish forgotten knowledge from stale knowledge.
The repair for stale knowledge is source refresh, not memory practice.
A Copyable Knowledge-Gap Prompt
“Help me find gaps in my knowledge without assuming every mistake means I do not know the topic. First ask me to produce an answer, explanation or task. Identify the first unstable step. Classify the gap as fact, concept, procedure, recognition, retrieval, transfer, integration, boundary, metacognitive or performance. Give me one diagnostic that could confirm the classification. Rank the gap by leverage and time-to-need. Repair only the confirmed gap, then retest with a changed task and update the gap register.”
Gap-Finding Failure Modes
Gap explosion
SI lists everything the learner does not know. Repair: scope to current goal and highest leverage.
Mistake equals gap
One slip becomes a permanent label. Repair: require recurrence or diagnostic evidence.
Explanation bias
Every failure triggers more explanation. Repair: classify first.
No transfer test
Gap appears closed on familiar examples. Repair: changed evidence.
No source frame
The system identifies “missing” material that is outside syllabus or goal. Repair: compare with roadmap.
Stale model
Closed gaps remain active forever. Repair: update state.
The Gap Audit
- Is this a knowledge gap or a performance gap?
- What is the first unstable step?
- Which gap type fits the evidence?
- Can one more diagnostic distinguish alternatives?
- Does the gap block other skills?
- Is it required soon?
- What is the smallest repair?
- What changed task proves closure?
- Should this gap be active, watched, maintained or closed?
- Is the source itself current and relevant?
The Knowledge Map Test
A learner can remember isolated pieces while still lacking a usable map of the subject.
Ask the learner to draw or describe the main concepts and relationships from memory. Then compare the map with a trusted roadmap, syllabus or reference.
Look for four kinds of gap: missing nodes, missing links, wrong links and misplaced emphasis.
Missing node: an essential concept is absent.
Missing link: two known concepts are not connected.
Wrong link: learner connects ideas through an incorrect relationship.
Misplaced emphasis: minor detail dominates while a core concept is weak.
SI is particularly useful for comparing the learner map against a reference and generating one diagnostic for each suspicious gap.
The Knowledge-Graph Gap
Some knowledge gaps are really relationship gaps.
A learner knows supply, demand and price but cannot explain equilibrium. They know variables, loops and functions but cannot see how a program decomposes into reusable components. They know vocabulary words but cannot choose them according to register.
These learners may score well on definition quizzes and still fail integrated tasks.
Ask SI to create a relationship test rather than another fact test.
The Link-Strength Test
For each important connection, ask the learner to complete one sentence: “X affects Y because ___.”
Then change the direction: “If Y changes, what can you infer about X?”
If the learner can state the terms but not the relationship, the gap is conceptual or integrative.
The Prerequisite Graph Audit
Many visible gaps are downstream symptoms.
Ask SI to map the target skill backward into prerequisites and then test the earliest unstable rung.
Example: quadratic factorisation may depend on multiplication facts, sign control, algebraic expansion and recognition of product-sum relationships.
Example: persuasive writing may depend on comprehension, claim formation, evidence selection, paragraph logic and vocabulary retrieval.
Example: statistical inference may depend on probability, distributions, variability and sampling concepts.
The audit prevents high-level reteaching when one prerequisite is doing most of the damage.
Hard Prerequisite Versus Helpful Background
Not every earlier topic is a hard prerequisite.
Ask SI to label each prerequisite as hard, soft or optional for the target output.
A hard prerequisite should be repaired before moving on. A soft prerequisite may be learned alongside the new skill. Optional background can wait.
This protects the learner from endless “preparation” that delays real practice.
The Misconception Probe
A misconception is not simply missing knowledge. It is an active wrong model.
Ask the learner to predict before seeing the answer.
If the same wrong rule generates several predictions, you have stronger evidence of a misconception.
Then ask SI to design a contrast where the misconception and correct model produce different outcomes.
That contrast becomes the repair point.
The “Why Did You Think That?” Probe
After an error, ask the learner why the wrong answer seemed reasonable.
The explanation may reveal a hidden rule, analogy or shortcut.
This is often more valuable than simply telling the learner the correct answer because it reveals the internal model that needs replacement.
The Retrieval Boundary
Sometimes the learner can retrieve part of the knowledge but not enough to complete the task.
Ask SI to find the boundary between remembered and forgotten.
Vocabulary example: learner remembers meaning but not spelling or register.
Science example: learner recalls the word equation but not the role of chlorophyll.
Coding example: learner remembers function syntax but not how return values are used.
The repair should target the missing component rather than reviewing the entire topic.
Cued Recall Versus Free Recall
If a cue immediately unlocks the answer, the gap may be retrieval rather than concept.
Track how much cue is required.
A learner who needs one first-letter cue is in a different state from a learner who needs a full explanation.
SI can use a help ladder to measure this difference.
The Recognition Gap Test
Ask the learner to classify several problems before solving any of them.
Which method applies? Which concept is being tested? Which evidence is relevant? Which code pattern is appropriate?
If classification fails while execution succeeds after the method is named, the gap is recognition rather than procedure.
This distinction is common in mixed examination papers and real-world work where nobody labels the method in advance.
The Method-Selection Matrix
For concepts with several neighbouring methods, build a matrix of cues, conditions and disqualifiers.
Example: when should you use substitution versus elimination? When should you use a bar chart versus scatter plot? When should a function return a value versus modify shared state?
The learner should use the matrix initially, then reproduce the selection logic without it.
The Transfer Gap Test
Change the surface while keeping the underlying skill constant.
Change wording, numbers, context, representation or output mode.
If performance collapses, ask whether the learner relied on surface cues.
Transfer gaps are often invisible in repetitive practice because the learner can match patterns without understanding the invariant structure.
Near Transfer Versus Far Transfer
Near transfer changes the surface slightly. Far transfer places the same idea in a substantially different context.
Do not jump directly to far transfer if near transfer is unstable.
SI can generate a ladder from near to far transfer so the learner’s boundary becomes visible.
The Integration Gap Test
Ask the learner to combine several individually stable skills inside one task.
Mathematics: algebra + geometry + ratio.
Writing: thesis + evidence + paragraph logic + vocabulary.
Coding: parsing input + functions + data structures + error handling.
If performance fails only when skills combine, the gap may be sequencing, working memory, planning or interface between skills.
Repair the integration process rather than re-teaching every component.
The Integration Sequence
Ask the learner to state the order in which subskills should operate.
Then compare with the actual attempt.
This exposes whether the learner knows the components but cannot orchestrate them.
The Boundary Gap Test
Ask for a counterexample or exception.
If the learner cannot identify conditions where the rule changes, they may be overgeneralising.
Example: not every correlation implies causation. Not every repeated task should be automated. Not every larger denominator means larger value. Not every loop is best replaced by recursion.
Boundary knowledge becomes especially important as the learner moves from beginner rules to expert judgement.
The Exception Library
For important concepts, store a small set of boundary cases.
These cases teach when the common rule fails and often reveal deeper structure.
SI can generate candidates, but verify technical exceptions against trusted sources.
The Metacognitive Gap Test
Ask the learner to estimate what kind of help they need before receiving it.
Do they think they need a definition, hint, example or full explanation?
Compare the request with what actually resolves the problem.
A learner who always asks for full explanations may not yet know how to diagnose their own gaps.
SI can gradually hand diagnosis back by asking the learner to classify the gap first.
The Self-Diagnosis Ladder
Stage 1: learner says only “I do not know”.
Stage 2: learner identifies where they got stuck.
Stage 3: learner identifies gap type.
Stage 4: learner proposes a diagnostic or repair.
Stage 5: learner tests and updates independently.
The long-term goal is moving the learner upward on this ladder.
The Gap Severity Matrix
Use four dimensions when several gaps compete.
Leverage: how many downstream skills depend on it?
Loss: how much performance does it cost?
Recurrence: how often does it appear?
Urgency: how soon is the skill needed?
A high-leverage low-urgency prerequisite may deserve steady repair. A high-loss high-urgency gap may deserve immediate triage.
SI can organise the matrix, but a teacher or learner may know contextual priorities the model does not.
The Gap Budget
Do not actively repair every gap at once.
Choose a small active set. Put lower-priority gaps on watch.
This reduces fragmentation and makes progress visible.
A learner with five active repair goals may make less progress than a learner who stabilises one high-leverage corridor first.
The Source Gap
Sometimes the gap is not inside the learner. The source is incomplete, outdated or misaligned.
A tutorial may skip a prerequisite. A textbook explanation may assume vocabulary. A software example may use an old API. A school note may be concise because the teacher expects classroom explanation.
Ask whether the learner is missing knowledge or the source is missing a bridge.
SI can compare two sources and identify what one assumes that the other makes explicit.
The Source Mismatch Test
If the learner repeatedly struggles despite multiple attempts, try a different representation or source before concluding the learner lacks ability.
Then retest with the same target output.
If performance improves, the original source may have been poorly matched.
The Gap Closure Protocol
A gap closes through evidence, not through explanation completion.
1. Confirm the gap.
2. Apply the smallest repair.
3. Retrieve the repaired idea.
4. Apply on a changed task.
5. Retest after a delay if the skill matters long-term.
6. Move the gap from active to maintain or closed.
If the changed task fails, reopen the classification. The repair may have targeted the wrong gap.
The Gap Reopen Rule
Closed does not mean permanent.
If a previously stable skill fails repeatedly after delay, move it to drifted or active repair.
This is normal learning maintenance, not evidence that the earlier learning was fake.
The 30-Day Knowledge-Gap Audit
Week 1: baseline, blank-page test and prerequisite audit.
Week 2: active gap repair plus contrast and transfer tests.
Week 3: integrated tasks and confidence calibration.
Week 4: delayed retest, close stable gaps and identify drift or new unknowns.
At the end of the month, the active gap register should be smaller and more precise.
The Gap Transfer Test
Use the same gap-detection method across subjects.
Mathematics: method, execution, recognition and transfer.
Language: recognition, retrieval, production and register.
Coding: syntax, decomposition, debugging and integration.
Professional learning: concepts, decision judgement, source currency and real-world application.
If the method transfers, the learner is building a general self-diagnosis capability.
The Final Gap Rule
Do not ask “What do I not know?” as though knowledge were one giant list.
Ask which capability is failing, where the first unstable step appears, what gap type best explains it and what smallest test can confirm the diagnosis.
The most useful knowledge gap is the one you can see clearly enough to repair.
The Blind-Spot Test
A blind spot is a gap the learner does not notice because their current model makes the wrong answer feel complete.
Blind spots are different from ordinary forgetting. The learner does not feel uncertainty.
SI can expose blind spots through contradiction, boundary cases, unfamiliar examples and comparison with a trusted reference.
Prompt: “Based on my explanation, what important assumption, boundary or neighbouring concept would an expert notice that I have not mentioned?”
Then verify the suggested blind spot rather than accepting it automatically.
The Surprise Test
Ask the learner what outcome would surprise them.
If a valid example produces exactly that surprising outcome, the learner’s model may be missing an important condition.
This is useful in probability, science, programming, economics and any subject where intuition can diverge from formal structure.
The Expert Question Test
Ask SI: “What question would a competent practitioner ask here that a beginner often forgets?”
Examples: What is the unit? What is the source? What is the baseline? What happens if the assumption fails? What is the exception path? Which stakeholder owns the handoff?
If the learner cannot answer, the gap may be procedural judgement rather than missing theory.
Confidence Mapping
Create a simple map of knowledge confidence.
High confidence / high performance: likely stable.
High confidence / low performance: dangerous blind spot or misconception.
Low confidence / high performance: under-confidence or insufficient evidence of stability.
Low confidence / low performance: known gap.
SI can use the matrix to choose different interventions.
High-confidence errors deserve explanation and contrast. Low-confidence correct performance may need repeated independent success rather than more teaching.
The Confidence Update Rule
Confidence should change when evidence changes.
Do not remain confident because a topic once felt familiar. Do not remain doubtful after repeated independent transfer.
Ask the learner to predict, perform and update.
This makes confidence evidence-sensitive rather than personality-based.
The Source-Freshness Gap
A learner can possess correct old knowledge that has become wrong for the current environment.
Examples include software APIs, exam formats, legal requirements, technical standards, company policies and product features.
Ask SI to distinguish memory failure from source freshness.
Prompt: “Which parts of this topic are stable fundamentals and which parts are version-, policy- or date-sensitive?”
The repair for a freshness gap is source verification and update, not more spaced repetition of obsolete information.
The Version Gap
Sometimes two explanations are both correct for different versions or contexts.
A programming tutorial may use an older library. A school may use a different notation convention. A professional framework may have revised terminology.
Attach the version to the knowledge state.
“Correct under version X” is better than treating the old knowledge as universally wrong.
Project-Based Gap Discovery
Projects expose gaps that isolated drills miss.
A learner may know all the required components separately but fail to plan, sequence, choose or verify when building something real.
Use a small project before declaring a learning roadmap complete.
Coding: build a tool from a short specification.
Writing: produce an argument using unfamiliar evidence.
Science: explain a novel scenario.
Data: clean and analyse a new dataset.
During the project, record where the learner stops, seeks help, misclassifies the task or cannot verify their own work.
The Project Gap Map
Classify project gaps as planning, source, concept, execution, integration, checking, communication or recovery.
This broader taxonomy is useful because real work includes more than subject knowledge.
A learner may know the science but fail to communicate it. Know the code but fail to test it. Know the vocabulary but fail to organise the essay.
The Recovery Gap
A learner may perform well until something goes wrong.
Can they diagnose and recover?
Mathematics: can they notice an impossible answer and backtrack?
Coding: can they isolate the failing component?
Writing: can they identify why a paragraph is off-topic and repair it?
Professional work: can they respond when an assumption or dependency fails?
Recovery is a distinct capability. SI can simulate failures and ask the learner to repair them.
The Verification Gap
Some learners can produce an answer but cannot check whether it is reasonable.
Ask for a verification method.
Substitute the value back. Estimate magnitude. Compare with source. Run a unit test. Use a counterexample. Check another representation.
If the learner has no checking strategy, the gap may be verification rather than production.
The Communication Gap
Knowing something and communicating it are different capabilities.
A learner may solve the problem but cannot explain the reasoning. A professional may understand the analysis but cannot produce a concise decision brief.
Ask SI to test the same knowledge through another output mode.
Explain orally. Write a summary. Produce a diagram. Teach the concept. Defend a recommendation.
If performance changes sharply by output mode, the gap belongs partly to communication or representation.
The Representation Gap
A concept may be stable in words and weak in symbols, or stable in diagrams and weak in prose.
Ask the learner to translate between representations.
This is a powerful gap test because many advanced tasks require moving between them fluently.
The Decision Gap
In professional or advanced learning, the gap may be judgement rather than information.
The learner knows the concepts but cannot choose between options under uncertainty.
Use cases with competing criteria.
Ask the learner to state which fact, assumption or value drives the choice.
SI can generate alternatives and counterarguments, but the learner should own the judgement.
The Decision-Quality Gap
A correct outcome can hide weak reasoning, and a poor outcome can follow a sound decision under uncertainty.
Ask the learner to review the process: evidence, assumptions, alternatives, criteria and review trigger.
This turns advanced learning into judgement practice rather than answer matching.
The Self-Diagnosis Session
Once a week, ask the learner to diagnose their own active gaps before SI comments.
“What failed?”
“What kind of gap do you think it is?”
“What test would distinguish your hypothesis from an alternative?”
“What repair do you propose?”
Then ask SI to critique the diagnosis.
This shifts the learner from passive recipient to active controller of the learning loop.
The Self-Diagnosis Transfer Test
Give the learner a completely new subject or task and ask them to apply the same gap framework.
If they can identify prerequisite, retrieval, recognition, transfer and performance gaps without being told, the diagnostic skill itself has transferred.
That is one of the most valuable outcomes of using SI well: the learner begins to know how to find what they do not know.
The Knowledge-Gap Dashboard
Keep the dashboard small.
Active gap.
Gap type.
Evidence.
Priority.
Repair.
Retest.
Status.
Do not turn the learner into a database of weaknesses.
The dashboard exists only to choose the next useful repair.
The Gap Simplification Review
Every few weeks, remove closed gaps and merge duplicate hypotheses.
If several errors trace back to one prerequisite, collapse them into the prerequisite corridor.
The active gap set should become smaller as the learner improves.
A good gap system creates focus, not a permanent catalogue of deficits.
Knowledge Gaps in Examination Preparation
Examination revision often treats every lost mark as a content gap. That is too crude.
Use SI to classify exam errors into content, recognition, execution, timing, checking, transfer and question-reading.
A wrong answer after choosing the correct method may need execution repair. A blank question may reflect recognition, time allocation or confidence. A correct answer reached too slowly may indicate fluency rather than understanding.
Build revision priorities from the error distribution rather than from chapter order.
The Mark-Loss Map
For each paper, record where marks were lost and why.
Topic.
Question type.
First unstable step.
Gap type.
Marks lost.
Repair status.
SI can aggregate several papers and show repeated corridors.
If algebra sign errors account for small losses across many topics, the total impact may exceed one dramatic mistake in a rarely tested extension topic.
The Exam Gap Priority Rule
Prioritise recurring high-mark losses in prerequisite skills, then recognition and timing, then lower-frequency extension gaps.
The exact order depends on the exam and learner, but the principle is to repair what changes the expected performance most.
Knowledge Gaps in Research
Research gaps are not only missing facts. They include missing definitions, missing populations, missing time periods, missing alternative explanations and missing source provenance.
Ask SI to inspect a research note and identify where a claim lacks evidence, where evidence does not match scope and where interpretation outruns the source.
Example: the learner knows that one study found an effect but does not know whether the sample, method or time period supports the broader claim they want to make.
This is a source-scope gap rather than a content gap.
The Citation Gap
A statement may be remembered correctly but detached from its source.
For academic or professional work, that can become a real gap because the learner cannot verify or defend the claim.
SI can help reconnect claims to sources and identify statements whose provenance has been lost.
The Interpretation Gap
The learner may understand each source separately but struggle to compare them.
Ask for synthesis: where do sources agree, differ in method, differ in population or support different interpretations?
This moves the learner from source collection to research reasoning.
Knowledge Gaps in Teams and Organisations
A team can have knowledge gaps even when individuals are competent.
Information may exist in one person’s head, one document, one system or one department but fail to reach the next decision.
These are coordination knowledge gaps.
Ask SI to map who knows what, who needs it, when they need it and which interface carries the information.
The Handoff Knowledge Gap
A handoff fails when the next person receives an output without the assumptions, context or status needed to use it.
Example: a report contains numbers but not data definitions. A project status says “done” but not whether approval is complete. A school summary says “weak topic” without representative errors.
SI can turn repeated handoff failures into a required-information checklist.
The Single-Expert Risk
If only one person understands a critical process, the organisation has a knowledge concentration gap.
Use SI to help document definitions, decision rules, failure modes and recovery procedures while the expert remains available to correct the record.
The goal is not to replace the expert. It is to make essential operational knowledge less fragile.
The Gap Between Knowing and Deciding
Some learners know the relevant facts but cannot turn them into a decision.
This is common in professional judgement, essay evaluation, project planning and personal decisions.
The learner may need criteria, thresholds, trade-off reasoning or risk structure rather than more information.
Ask SI to identify whether the missing capability is knowledge or decision architecture.
The Criteria Gap
A learner cannot compare options well if they do not know which criteria matter.
Example: choosing an experimental method requires validity, feasibility, cost and ethical constraints. Choosing a database design may require integrity, performance and maintainability. Choosing a study method may require retrieval, transfer and sustainability.
SI can propose criteria, but the learner should understand why each one matters.
Hallucinated Knowledge Gaps
SI can invent a missing concept, requirement or weakness that is not actually relevant.
This can happen when it generalises from a generic curriculum or assumes an advanced framework.
Never let the assistant’s claim that “you need to know X first” become the source of truth automatically.
Check the roadmap, syllabus, documentation or task requirements.
The Gap Verification Rule
A gap hypothesis becomes active only after evidence.
Evidence can be failed diagnostic, missing explanation, source mismatch, repeated project failure or inability to transfer.
The model’s suggestion alone is not enough.
The False-Correction Risk
If SI incorrectly marks a correct answer as wrong, it can manufacture a gap.
For high-value learning, use source-aligned marking and ask the tutor to show the rule or source behind the correction.
If the learner disagrees, switch to verification mode rather than forcing conformity.
The Knowledge-Gap Interview
For a new subject, use a short interview before testing.
What have you studied before?
What tasks can you already complete?
Where do you feel uncertain?
Which source or course are you using?
What independent output are you trying to produce?
Then use performance evidence to correct the self-report.
Self-Report Is Useful but Incomplete
Learners often know where they feel confused, but they may misidentify the cause.
Use self-report as one source, then compare with attempts, diagnostics and transfer.
This respects learner experience without treating it as the only evidence.
The Gap Review after a Course or Project
At the end of a course, ask what the learner can now do and where they still depend on prompts, templates or examples.
Ask which knowledge remains unused, which skills never reached transfer and which assumptions about the subject changed.
This review often identifies the next learning roadmap more accurately than another course catalogue.
The Gap Closure Evidence Standard
Use evidence proportional to the skill.
A simple vocabulary fact may close after repeated retrieval. A complex professional judgement may require several cases. An exam skill may require mixed timed tasks. A coding skill may require a working project and debugging.
Do not use one universal definition of “mastered”.
The Long-Horizon Gap Review
Every few months, review not only active weaknesses but also the structure of knowledge.
Which concepts became central?
Which once-important gaps no longer matter?
Which knowledge became stale?
Which new responsibilities created new gaps?
Which area now deserves a deeper roadmap?
A learner’s gap profile should evolve with their goals.
The Knowledge-Gap Continuity Card
Current outcome: what capability matters.
Active gaps: only confirmed high-priority gaps.
Watch list: hypotheses needing evidence.
Closed gaps: recently repaired, with latest retest.
Freshness risks: knowledge needing source review.
Next diagnostic: one test with highest information value.
This card is enough to continue gap work across sessions without preserving every old mistake.
The Final Gap-Ownership Test
Ask the learner to describe their own active gaps and why they matter.
If the learner can only repeat labels SI generated, self-diagnosis has not yet transferred.
If they can identify the evidence, classify the gap, propose a diagnostic and choose the next repair, the learner is becoming capable of managing their own knowledge state.
The deepest benefit of gap detection is not finding more weaknesses. It is learning how to see the boundary of your own understanding.
Gap Density: When Many Errors Share One Cause
A learner may appear to have many gaps when one unstable prerequisite is generating errors across several topics.
Ask SI to cluster errors by possible common cause.
Example: negative-sign errors appear in expansion, equations, coordinates and substitution. Instead of four topic gaps, the learner may have one sign-control corridor creating broad loss.
Example: weak vocabulary retrieval appears in comprehension explanation, oral response and essay writing. One active-language gap may be affecting several outputs.
Repairing a dense root gap creates more leverage than treating each symptom separately.
The Error-Cluster Test
Group recent errors by structural similarity rather than chapter name.
Ask: What repeated mental move is failing across these tasks?
If one move explains several errors, test it directly.
If the errors remain unrelated, keep separate gap hypotheses.
Repair Economics: Fix the Gap with the Highest Return
Not every gap deserves the same amount of time.
Estimate the expected return of repair.
How many downstream skills improve?
How frequently does the error appear?
How expensive is the repair?
How soon is the skill needed?
SI can help prioritise, but the learner or teacher should decide whether a high-cost repair is justified by the real goal.
The Cheap Diagnostic Rule
Before investing heavily in a repair, use the cheapest diagnostic capable of changing the plan.
Three targeted algebra questions may be enough to test a prerequisite. One short code task may reveal whether the issue is syntax or decomposition. A two-minute speaking prompt may reveal active vocabulary retrieval.
Good gap work reduces uncertainty quickly before consuming large amounts of study time.
The Diagnostic Question Bank
- Can you do this without notes?
- Can you explain why the method works?
- Can you recognise when the method applies?
- Can you distinguish it from a neighbouring method?
- Can you use it when the wording changes?
- Can you use it after a delay?
- Can you combine it with another skill?
- Can you find and fix your own error?
- Can you state when the rule does not apply?
- Can you verify your own answer?
These questions locate the boundary of understanding much more efficiently than asking “Do you understand?”
The Gap Escalation Rule
Escalate when a gap remains unclear after several diagnostics, when the source or marking convention is disputed, when the learner’s performance changes sharply without explanation, or when the subject requires expert judgement beyond the SI system.
Use SI to prepare the evidence for a teacher, tutor, supervisor or appropriate expert.
The handoff should include representative attempts, the suspected gap, diagnostics already tried and what remains unresolved.
A Full Case Study: Secondary Mathematics Gap Hunt
A student scores poorly across algebra, coordinate geometry and word problems.
Chapter-by-chapter analysis suggests three weak areas.
Structural analysis shows that many errors begin when negative values enter the working.
Diagnostic: short negative-number and sign-distribution tasks.
Result: arithmetic with negatives is stable, but distributing a negative coefficient across brackets is not.
Classification: procedure plus transfer gap.
Repair: explicit sign-distribution contrast, retrieval, then mixed algebra.
Retest: coordinate and equation questions where the same sign structure appears.
Outcome: one repaired corridor reduces errors in three apparent topics.
A Full Case Study: Knowledge Work Gap Hunt
A professional says they need to learn “data storytelling”.
Baseline shows they can create charts and calculate measures but struggle to connect analysis to a decision.
Gap classification: communication and decision relevance, not technical analysis.
Repair: each chart must answer one stakeholder question, state the evidence and identify the implication or uncertainty.
Retest: new dataset and a one-page decision brief.
The learner avoids an unnecessary advanced visualisation course because the real gap is narrative judgement.
A Full Case Study: Language Learning Gap Hunt
A learner has completed many vocabulary lessons but still speaks slowly.
Recognition test: very high.
Free recall: moderate.
Sentence production: low under time pressure.
Classification: active retrieval and fluency gap.
Repair: timed retrieval, short speaking prompts, phrase chunks and delayed repetition.
Retest: spontaneous response on an unfamiliar but everyday topic.
The learner stops adding large amounts of passive vocabulary until active use catches up.
The Gap Discovery Workflow
1. Define the target capability.
2. Produce representative evidence.
3. Find the first unstable step.
4. Generate two or three gap hypotheses.
5. Design the cheapest distinguishing diagnostic.
6. Confirm the gap type.
7. Rank by leverage and time-to-need.
8. Apply the smallest repair.
9. Retest with changed evidence.
10. Close, maintain or reopen.
This workflow converts “I need to learn more” into a precise control loop.
The Final Gap Check
Before adding another course, book, tutor, prompt or practice set, ask whether the active gap has been demonstrated.
Can you name the failed capability?
Can you show representative evidence?
Can you distinguish it from neighbouring explanations?
Can you state the smallest repair?
Can you define the retest?
If yes, the gap is actionable.
Find fewer gaps, but find the right ones.
The Gap Confidence Statement
Every active gap should have a confidence statement. Example: “High confidence that bracket expansion is unstable because the same sign error appeared across six changed questions; lower confidence that time pressure is also contributing because timed evidence is limited.”
This prevents the gap register from turning hypotheses into facts. It also tells the next learning session whether to repair immediately or gather one more diagnostic.
The Gap Closure Confidence Rule
Close a gap only when the learner has produced enough independent evidence for the real task. A single correct familiar example is weak closure. Several changed examples, a mixed transfer task and a delayed retest provide stronger closure where the skill matters.
The evidence standard should match consequence. A low-stakes vocabulary fact may need less proof than a core Mathematics prerequisite or professional safety procedure.
The Gap Continuity Rule
Carry forward only current confirmed gaps, not every historical weakness. When a gap closes, move it to maintenance or archive. When a new gap appears, connect it to the current target before activating it.
This keeps the learner focused on what matters now and prevents the system from creating a permanent identity out of old mistakes.
A knowledge-gap system should make learning lighter as understanding improves, not heavier because the history keeps growing.
The Gap Ownership Sentence
Ask the learner to complete one sentence: “The gap I am working on is ___; I know it is real because ___; the smallest repair is ___; I will know it is closed when ___.”
If the learner cannot explain the evidence or closure condition, the diagnosis may still live only inside SI rather than inside the learner’s own understanding.
The Gap Review Trigger
Reopen the gap analysis when performance changes unexpectedly, the source or syllabus changes, a previously stable skill drifts, a project exposes a new integration failure or the learner repeatedly needs more help than the current state predicts.
Do not reopen every closed gap after one isolated mistake. Use repeated evidence or a representative diagnostic.
The Final Gap Principle
Learning becomes faster when the learner stops treating every difficulty as a demand for more content. Sometimes the next move is not another chapter, explanation or course. It is one precise test that reveals what is missing.
Super Intelligence is most useful when it narrows the uncertainty around the gap, then gets out of the way while the learner repairs it.
The Gap Handoff Rule
When a gap needs a human teacher, tutor or expert, hand off evidence rather than a label. Share two representative attempts, the first unstable step, the suspected gap type, what diagnostics were tried and what remains unresolved.
“Weak at algebra” is a poor handoff. “Accurate on linear equations until a negative coefficient must be distributed across brackets; same error recurs on four changed questions” gives the next person something actionable.
This rule also protects the learner from broad deficit labels. A precise gap can be repaired. A vague identity can follow the learner unnecessarily.
The Gap-Minimisation Rule
Do not reward the system for finding the maximum number of weaknesses. Reward it for finding the smallest set of gaps that explains the observed performance and changes the learning plan.
When several hypotheses explain the same error, use the cheapest diagnostic to eliminate them. When one prerequisite explains several symptoms, prefer the simpler model until evidence requires more complexity.
The strongest gap analysis reduces the problem to the fewest confirmed repairs that restore independent performance.
Final gap checkpoint: before starting another learning resource, ask whether the current obstacle has been diagnosed with evidence. If not, collect the smallest diagnostic first. If yes, repair the confirmed gap and retest before widening the scope. This simple order prevents learning systems from becoming resource-collection systems and keeps attention on the capability that actually needs repair.
Final gap system note: every new learning difficulty should enter as a hypothesis, not a permanent label. Confirm it with the smallest useful evidence, repair only what is demonstrated, and remove it from the active system when independent transfer proves the gap is no longer controlling performance.
The Gap Finality Rule
Do not keep analysing a confirmed gap after the repair and retest already show stable independent performance. Move the skill into maintenance and direct attention to the next active corridor.
This prevents gap finding from becoming a habit of permanent deficiency scanning. The purpose is to restore capability and then release the learner from the gap.
Career-Skill Diagnostic Lab: Find the First Unstable Step
You do not need to turn a disappointing work sample into a judgement about yourself. You need a requirement you can inspect, an attempt you can show, and a test that separates the likely explanations. This lab applies the gap-detection loop to one job-related task: producing a checked service-performance brief. It ends with a specific repair and a new performance test, rather than a list of courses.
Everything in this lab is fictional and supplied for practice: the role, records, people, policies, attempts and model responses. None is an actual applicant, employee assessment or reported experiment. The worked answers illustrate what would satisfy this particular brief; they do not demonstrate anyone’s real progress. Super Intelligence means practical AI assistance here. It is not an authority on a person’s ability, suitability or future employment.
Follow the task packet, inspect the diagnosis, try the independent retest, then attempt the changed-rule transfer. Keep each answer panel closed until you have written your own response. An SI-generated answer is evidence about the answer, not evidence that you can produce it independently.
1. The supplied role and judging standard
Role card R1, training version 1: Nila is practising one task associated with a fictional Service Operations Coordinator role. Given a weekly export and its current measurement rule, prepare a short internal brief that calculates the within-24-hour first-reply rate, checks an 80% target, identifies the queue needing investigation and describes one proportionate next check. The task is a training exercise, not a hiring decision. No qualification, certificate or software brand is required by this role card.
Rule M1: each real, unique customer ticket in the supplied reporting cohort belongs in the denominator. Exclude test records and duplicate records. A ticket succeeds only if its first reply arrived within 24 hours, including exactly 24 hours. Late first replies and tickets with no first reply remain in the denominator. All eligible tickets in these packets were received early enough that more than 24 hours had elapsed at the snapshot. There is no not-yet-due exception to infer.
Data dictionary: “input rows” counts all export rows before exclusions. Test and duplicate counts are disjoint exclusions; no row belongs to both. Within-24-hour, late and no-reply counts are disjoint outcomes for the remaining real, unique tickets. Those three outcomes must sum to eligible tickets. A combined rate uses combined successes divided by combined eligible tickets, not a simple average of queue percentages. Round only the final percentages to one decimal place.
Permission card P1: the learner may calculate and prepare an internal training draft. Only the fictional supervisor can approve a report for release or authorise an operational change. The supplied counts do not explain why a reply was late. A lower queue rate identifies a place to investigate, not a person to blame. Do not upload actual customer records or confidential employer material to an AI service without the relevant permission.
Use the following six checks for this task only. A check passes only when every condition in that row is met. There is no partial credit concealed inside a pass. Six passes would show that this particular response meets the supplied brief; it is not a validated skill scale, a hiring score or evidence of performance across a whole occupation.
| Check | What the submitted brief must show |
|---|---|
| G1: population | Correct eligible counts for both queues and the total; the applicable exclusions and no-reply treatment are respected. |
| G2: calculation | Correct queue rates and combined rate, each to one decimal place, with the relevant numerator and denominator visible. |
| G3: decision relevance | Correct target conclusion, the lower-rate queue, and a specific next investigation that does not assume a cause. |
| G4: traceability | The packet and rule version are named; the meaning of the rate is clear enough for another reader to reproduce it. |
| G5: inference boundary | No claim that these counts establish a cause, a worker’s general competence or readiness for employment. Any proposed explanation is explicitly unconfirmed. |
| G6: authority | The output is labelled a draft for supervisor review; no release or operational change is represented as authorised. |
2. Two baseline attempts you can inspect
For the baseline, Nila uses the supplied packets with a calculator, without SI help, hints or a pre-filled formula. The attempts below are authored examples, not observations of a real learner. Read the rule before reading the response. Ask where the calculation first stops representing the required measure.
| Packet A: queue | Export rows and exclusions | Eligible-ticket outcomes |
|---|---|---|
| Cedar | 50 input rows; 4 tests; 6 duplicates | 30 within 24 hours; 6 late; 4 no reply |
| Elm | 40 input rows; 5 tests; 5 duplicates | 24 within 24 hours; 3 late; 3 no reply |
Baseline attempt A: “Training draft for supervisor review, Packet A, rule M1. I excluded the test and duplicate rows, then used the tickets that received a reply. Cedar is 30/36 = 83.3%; Elm is 24/27 = 88.9%. Combined, 54/63 = 85.7% received a first reply within 24 hours. We meet the 80% target, so no further check is needed. The table does not tell us why any reply was late.”
| Packet B: queue | Export rows and exclusions | Eligible-ticket outcomes |
|---|---|---|
| Cedar | 45 input rows; 2 tests; 3 duplicates | 28 within 24 hours; 8 late; 4 no reply |
| Elm | 35 input rows; 3 tests; 2 duplicates | 27 within 24 hours; 1 late; 2 no reply |
Baseline attempt B: “Training draft for supervisor review, Packet B, rule M1. Cedar is 28/36 = 77.8%; Elm is 27/28 = 96.4%. Overall, 55/64 = 85.9% received a first reply within 24 hours. We meet the target overall, so I would stop here. As before, I excluded no-reply tickets because they are not completed. These figures alone do not establish the cause of the late replies.”
Both attempts contain correct division for the denominators Nila selected. Both select the wrong denominator. In A, eligible totals are 50 − 4 − 6 = 40 and 40 − 5 − 5 = 30. In B, they are 45 − 2 − 3 = 40 and 35 − 3 − 2 = 30. Removing unanswered tickets changes the question from “What fraction of all eligible tickets met the reply standard?” to “What fraction of replied-to tickets met it?” The brief required the first question.
| Visible check | Correct Packet A result | Correct Packet B result |
|---|---|---|
| Queue rates | Cedar 30/40 = 75.0%; Elm 24/30 = 80.0% | Cedar 28/40 = 70.0%; Elm 27/30 = 90.0% |
| Combined result | 54/70 = 77.1%, below 80% | 55/70 = 78.6%, below 80% |
| Next check | Investigate Cedar’s 6 late and 4 no-reply cases | Investigate Cedar’s 8 late and 4 no-reply cases |
| Baseline gates | G1, G2 and G3 fail; G4, G5 and G6 pass | G1, G2 and G3 fail; G4, G5 and G6 pass |
The two missed target conclusions are downstream of the same population error. Do not record three separate personal weaknesses simply because three checks failed. Traceability passes here because the attempts identify the packet, label the measure and expose their actual calculations; their error is visible rather than hidden. The decision-relevance check fails because the conclusion and proposed stopping point are wrong under M1.
3. Make SI show its diagnostic evidence
Bounded prompt: “Use only R1, M1, P1 and attempts A and B. Identify the first unstable step and quote the part of each attempt that supports it. Separate demonstrated performance, a confirmed task-specific gap, untested requirements and external blockers. Give competing explanations and one small diagnostic that could distinguish them. Do not infer personality, protected traits, global ability, employment prospects or a need for credentials. Do not write the independent retest answer for me.”
Illustrative flawed SI response: “Nila is weak at analytics and needs an advanced dashboard certificate before taking on this work.” This response introduces a broad label and a credential requirement absent from R1. Neither follows from two incorrect denominator selections. A human should reject the claim rather than convert it into a training purchase or a career judgement.
Checked replacement: “Across A and B, Nila excludes no-reply tickets despite M1 including them. Candidate explanation: difficulty selecting the reporting population when an outcome is incomplete. Division may be intact because the displayed divisions match the selected denominators. A second explanation is misunderstanding the wording of M1; a third is an execution slip. Use an explicit-denominator calculation and a record-classification probe to distinguish them. Dashboard operation has not been tested.”
| Hypothesis | Evidence already available | Diagnostic and interpretation |
|---|---|---|
| The division procedure is unstable | Displayed calculations agree with the selected denominators in A and B | Give denominators explicitly. Repeated correct calculations weaken this explanation within these examples. |
| The eligible population is misclassified | No-reply tickets are excluded in both attempts; B supplies the reason | Ask which record types belong in M1. Excluding an overdue no-reply ticket supports a rule-selection or boundary error. |
| A one-off copying slip caused the result | The same exclusion appears in two different packets | Ask for the rule in the learner’s own words. A repeated deliberate exclusion is more informative than an isolated typo. |
| Time pressure or an interface caused the error | No comparable timed task or interface test is supplied | Do not diagnose either from this packet. Gather matched-condition evidence only if it becomes relevant. |
Diagnostic D1, arithmetic without population selection: “The eligible denominators have already been checked. Calculate 18/24, 9/10 and the pooled rate for 18 + 9 successes out of 24 + 10 eligible cases.” The supplied learner attempt is “75.0%; 90.0%; 27/34 = 79.4%.” All three are correct. This supports the division procedure and pooled calculation under explicit inputs, not every possible numerical skill.
Diagnostic D2, classification without division: “Apply M1 to five records: one test; one duplicate; one real unique ticket replied to in 12 hours; one real unique ticket replied to in 30 hours; one real unique ticket still unanswered after 30 hours. Name the denominator records and the success records.” The supplied attempt includes the 12-hour and 30-hour replies in the denominator, counts only the 12-hour reply as a success, and excludes the unanswered ticket “because it is unfinished”.
The checked D2 answer puts all three real, unique tickets in the denominator and only the 12-hour reply in the numerator: 1/3 = 33.3%. Tests and duplicates are excluded. This contrast localises the current problem: under M1, Nila is treating an incomplete outcome as a reason to exclude an eligible case. The repeated reasoning supports a task-specific recognition/boundary gap more strongly than a division gap.
Support probe D3: after being shown the sentence “No-reply tickets remain eligible once the 24-hour window has elapsed”, the constructed learner corrects D2 to 1/3. Record this as a supported correction. It does not count as an independent retest, because the decisive rule was just supplied as a cue.
4. The evidence register before repair
The register describes the scope of the evidence. “Untested” means no suitable performance sample exists; “blocked” describes a missing condition for a test; neither means the person lacks the skill. Self-reported confidence may help choose a test, but does not override the attempts.
| Capability or condition | Current classification | Evidence and next move |
|---|---|---|
| Select the M1 denominator | Active task-specific gap | A, B and D2 exclude overdue no-reply tickets. Repair that boundary, then test a changed packet without hints. |
| Calculate percentages from valid inputs | Demonstrated in these samples | D1 is correct; A and B divide their chosen inputs correctly. Do not prescribe a broad arithmetic course on this evidence. |
| Translate valid rates into a useful brief | Not independently established yet | A and B use invalid inputs. Retest the complete brief after the denominator repair instead of diagnosing a separate communication gap immediately. |
| Name sources and respect release limits | Demonstrated in the baseline text | Both attempts name their packet and rule and remain drafts for review. Continue checking this when the task changes. |
| Operate the employer’s dashboard | Untested; access condition absent | No dashboard task or authorised account is supplied. A written calculation does not prove tool operation, and absent access does not prove inability. |
| Choose a career or meet a whole occupation’s requirements | Outside this diagnostic | R1 contains one task. Do not infer occupational readiness or a hiring probability from it. |
Prioritise denominator selection because it changes the target conclusion in both supplied samples. Keep the active repair this narrow. There is no evidence here that Nila needs a new career direction, a particular course or a 12-week development schedule.
5. Targeted repair: freeze the population before calculating
Repair the rule at the point of failure. Before calculating, state the unit, reporting cohort, deadline, exclusions and success condition. Write the denominator as a sentence: “All real, unique customer tickets in this cohort after excluding tests and duplicates, including late and no-reply tickets.” Then check that eligible tickets equal within-24-hour + late + no-reply outcomes. Only then calculate the rate.
Repair contrast: a small M1 cohort has 10 eligible tickets: 8 replies within 24 hours, 1 late reply and 1 still unanswered after 24 hours. The rate is 8/10 = 80.0%. If the unanswered ticket later receives a late reply, the rate remains 8/10 = 80.0%. Its outcome category changes, but neither its eligibility nor its missed deadline changes. Excluding it while unanswered would have produced 8/9 = 88.9%, a different measure.
Second contrast: add 2 test rows and 3 duplicate rows to the same export. There are now 15 input rows, but still 10 eligible tickets and 8 successes: 8/10 = 80.0%. The raw file grew; the reporting population did not. The reason for an exclusion must come from the applicable rule, not from whether a row is convenient to count.
Constructed repair explanation: “I used completion status to choose the population. Under M1, an overdue unanswered ticket is eligible and unsuccessful. I will reconcile eligible totals before dividing, and I will re-read exclusions if the rule changes.” This is a useful explanation to practise, but repeating it alone does not show that the full task is repaired.
During repair, SI may ask a contrast question or point to M1. During the independent retest, it should withhold hints, formula completion and error flags until the learner has submitted the brief. If help is used, record what was provided and call the attempt assisted. Use another fresh packet before claiming independent performance.
6. Independent retest: new counts and a different priority queue
Packet C: use R1, M1 and P1 unchanged. River has 36 input rows, 3 tests and 3 duplicates; eligible outcomes are 25 within 24 hours, 3 late and 2 no reply. Pine has 44 input rows, 2 tests and 2 duplicates; eligible outcomes are 30 within 24 hours, 6 late and 4 no reply. All eligible tickets have passed their reply deadline at the snapshot.
Your task: without looking at the worked answer, produce the complete brief: both eligible counts, both queue rates, the combined rate, the target conclusion, one next check, the source/rule and the release boundary. Use a calculator and the supplied source packet as you would in this task, but no SI hints, earlier worked calculations or copied template answer. Source access is part of this job; memorising a policy is not the capability being tested.
Open the Packet C checking answer after submitting your brief
Reconciliation: River: 36 − 3 − 3 = 30 eligible, also 25 + 3 + 2 = 30. Pine: 44 − 2 − 2 = 40 eligible, also 30 + 6 + 4 = 40. Total: 70 eligible and 55 within-24-hour replies.
Rates: River 25/30 = 83.3%; Pine 30/40 = 75.0%; combined 55/70 = 78.6%. The overall rate is below the 80% target by 1.4 percentage points after rounding. Averaging 83.3% and 75.0% would give 79.15%, which does not preserve the different queue sizes.
Illustrative complete brief: “Training draft for supervisor review. Packet C under M1 counts all real, unique cohort tickets, including late and no-reply cases; tests and duplicates are excluded. River has 30 eligible tickets and 25 on-time first replies (83.3%); Pine has 40 and 30 (75.0%). Combined, 55/70 = 78.6%, below the 80% target. Pine is the lower-rate queue. Review its 6 late and 4 no-reply cases, checking receipt/reply timestamps and handoff records before proposing a cause or change. The aggregate counts do not explain the cause. Supervisor approval is required before release.”
Checking result: that supplied answer would pass G1–G6. An answer that excludes no-reply tickets fails G1 and G2 even if its division is accurate. An answer that recommends retraining a named person without causal evidence fails G3 and G5. Reusing Cedar as the priority without inspecting C fails G3. Saying “published” fails G6.
Interpretation: if your own first unaided answer passes, move the M1 population gap to maintenance pending a delayed check. This page does not claim that you have passed. If G1 fails, return to the population contrast. If G1 passes but G2 fails, inspect the arithmetic separately. If the numbers pass but the brief fails, target the specific interpretation, evidence or authority error rather than repeating the denominator lesson.
7. Countercase: missing access is not a demonstrated skill deficit
Supplied access record X: fictional learner Jules receives “Report view unavailable: account has not been granted access” before any dashboard task appears. The workflow note does not identify who grants access. An SI summary says, “Jules cannot use dashboards; enrol in dashboard training.” There is no dashboard performance attempt to support that diagnosis.
The supervisor then provides an approved offline training packet: 14 rows, 1 test and 1 duplicate; outcomes are 9 within 24 hours, 2 late and 1 no reply, with every ticket’s reply deadline elapsed at the snapshot under M1. Jules writes: “Draft, Packet X under M1: 12 eligible tickets, including the no-reply case. 9/12 = 75.0%, below 80%. Check the 2 late and 1 no-reply records before proposing a cause. Supervisor approval is needed for release.” This constructed response supports calculation and interpretation on the offline task. It does not prove dashboard navigation or live-system competence.
Record two separate facts: “The observed reporting task is blocked by missing access and an unclear access-owner handoff” and “dashboard operation remains untested”. Route the access question through the authorised supervisor, then use a permitted sandbox task to test the tool operation if it is genuinely required. Do not bypass permissions, request credentials in a prompt, or diagnose an individual skill deficit from the missing opportunity to perform.
8. Independent transfer: the rule changes, not just the numbers
Role card R2 and rule M2, training version 2: prepare a fictional order-dispatch brief. The target is at least 90% dispatched on time. Eligible orders are real, unique, paid orders received before the stated cutoff. Exclude tests, duplicates and documented customer-requested holds placed before that cutoff. Include eligible orders delayed by stock shortages or still awaiting dispatch. Every eligible order’s dispatch deadline has passed at the snapshot. Categories are disjoint. The learner may draft; the supervisor must approve release and any operational change.
| Packet D: depot | Export rows and exclusions | Eligible-order outcomes |
|---|---|---|
| East | 62 input rows; 2 tests; 2 duplicates; 8 documented customer-requested holds placed before the cutoff | 45 on time; 3 late; 2 awaiting dispatch |
| West | 48 input rows; 2 tests; 1 duplicate; 5 documented customer-requested holds placed before the cutoff | 34 on time; 4 late; 2 awaiting dispatch |
Supplied flawed attempt D: “Packet D: holds must always remain in the denominator, just like unanswered tickets. East is 45/58 = 77.6%; West is 34/45 = 75.6%; combined 79/103 = 76.7%. We miss 90%. Buy more stock immediately and publish this report.” No stock-cause evidence or release approval is included in the packet.
Your independent task: diagnose the first unstable step without asking SI for an answer. Separate it from the cause and permission errors. Reconcile the counts under M2, calculate all three rates, write a corrected brief and name the smallest repair for the flawed reasoning. Finish by stating which skill is demonstrated, which gap is supported and what remains untested. Use the same six judging checks, substituting the M2 population, 90% target and dispatch outcomes.
Open the changed-rule transfer answer after completing your diagnosis
First unstable step: the attempt transfers M1’s inclusion rule as an absolute rule rather than reading M2. Customer-requested holds are explicitly excluded under M2. Overdue orders awaiting dispatch are still included. “Incomplete items always count” is not the principle. “Use the current, task-specific eligibility rule” is the principle.
Reconciliation: East: 62 − 2 − 2 − 8 = 50 eligible, matching 45 + 3 + 2 = 50. West: 48 − 2 − 1 − 5 = 40 eligible, matching 34 + 4 + 2 = 40. Combined: 90 eligible and 79 on-time dispatches. The flawed denominator has counted 13 excluded holds: 90 + 13 = 103.
Rates: East 45/50 = 90.0%; West 34/40 = 85.0%; combined 79/90 = 87.8%. East meets the target exactly. West and the combined population are below it; the combined shortfall is 2.2 percentage points after rounding. The flawed attempt happens to get the overall below-target conclusion right, but that does not validate its population or rates.
Corrected brief: “Training draft for supervisor review, Packet D under M2. Tests, duplicates and documented pre-cutoff customer holds are excluded; overdue orders awaiting dispatch remain eligible. East has 50 eligible orders and 45 on-time dispatches (90.0%); West has 40 and 34 (85.0%). Combined, 79/90 = 87.8%, below the 90% target. Check West’s 4 late and 2 awaiting-dispatch cases, including recorded delay reasons and timestamps, before proposing a remedy. The counts do not show that stock shortages caused the result. Supervisor approval is required before release or any stock purchase.”
Smallest repair: contrast one documented pre-cutoff customer hold with one stock-delayed eligible order. Under M2, exclude the former and include the latter. Ask the learner to cite the relevant sentence, then classify a fresh pair without a hint. Re-reading the current rule is part of the work, not a failure to memorise M1.
Evidence classification: the supplied flawed divisions are arithmetically consistent with their chosen denominators, so they do not by themselves support a broad arithmetic gap. The quoted “always” rule and the packet mismatch support a task-specific source/transfer boundary error. The stock purchase and publication sentences separately lack causal evidence and authority. Tool operation remains untested. The corrected supplied brief would pass all six checks; your independently produced answer must be judged on its own evidence.
Repair checks: a denominator of 103 is still wrong even if the conclusion is below target. Excluding the 4 awaiting-dispatch orders is also wrong: they are eligible under M2. A correct numerical answer with an unsupported stock purchase or claimed publication does not pass the complete task. Do not convert this one constructed transfer failure into a permanent label about a person.
9. Update the record without inventing progress
After a genuine attempt, preserve the submitted work, its source/rule version, date, permitted aids and six check results. If Packet C is independent and correct but D fails, record the distinction: “M1 task currently demonstrated; changed-rule transfer still needs repair.” If both pass, record the denominator repair as demonstrated on these task forms and retain a delayed-check date appropriate to the real work. Do not mark long-term closure before that check has occurred.
A useful delayed check should provide a new packet with the relevant source available, no model hints and a mix of excluded records and eligible unsuccessful outcomes. Compare the same evidence criteria. If a later rule has changed, test the current rule rather than rewarding faithful use of an obsolete one. If the source is contradictory or totals do not reconcile, pause the metric and ask its human owner to resolve the record; uncertainty in the input is not automatically a gap in the learner.
Completed handoff example after the baseline only: “Target: the M1 reply-rate brief. Evidence: A and B exclude overdue no-reply tickets; D2 repeats that choice. Explicit-denominator arithmetic in D1 is correct. Active repair: recognise that overdue no-reply tickets stay eligible under M1. Supported correction D3 is complete; independent C, changed-rule D and delayed performance are not yet established. Dashboard operation is untested. Next action: submit C without hints, then judge it against G1–G6.” This filled record makes the next step clear without claiming an outcome that has not happened.
Career-skill lab questions
Can I use a job advert instead of the fictional role card?
Yes, as a source for a specific current requirement, with its date and exact wording retained. Separate a required task from a desirable tool or credential. If the advert is vague, clarify the expected work product before testing yourself. A missing certificate is a missing stated credential, if genuinely required; it is not proof that an observed task skill is absent. This lab does not determine whether an employer will accept alternative evidence.
What if my answer is correct only after SI points out the denominator?
Keep it as an assisted repair attempt. Write down the cue, then use a different packet without that cue. Do not discard the improvement, but do not count it as independent performance. If you need the same hint repeatedly, the recognition step remains active even when the final arithmetic is correct.
What if the requirement or marking key is wrong?
Challenge it with the source and a reproducible calculation. An unreconciled export, an unclear cutoff or incompatible definitions calls for source clarification. Do not manufacture a precise rate or a personal deficit to make the assessment look complete. A supervisor or relevant subject expert should resolve disputed operational definitions.
Use the resulting evidence register to choose the next repair. For comparing possible career directions, continue to the career-guidance section of How Career Resilience Works. Once a direction and its requirements are supported, the same owner’s 90-day development section handles the broader planning task. Here, the useful result is smaller: a defensible description of what the evidence shows, one targeted repair and a fair chance to demonstrate it again.
Frequently Asked Questions
Can SI really identify what I do not know?
It can help expose likely gaps through diagnostics, explanation, contrast and source comparison. Gap classifications should be confirmed with performance evidence.
What is the difference between a knowledge gap and a retrieval gap?
A knowledge gap means the idea or relationship is missing. A retrieval gap means the learner can recognise or access it with support but cannot recall it independently.
What if I make one careless mistake?
Do not treat one slip as a permanent gap. Use another targeted task.
How do I find unknown unknowns?
Compare with trusted maps, ask for common blind spots, use projects and boundary cases, and inspect expert concepts adjacent to your current understanding.
Should I fix every gap?
No. Prioritise gaps relevant to your current goal, leverage, consequence and time-to-need.
How do I know a gap is closed?
The learner succeeds independently on changed evidence and retains the capability after an appropriate delay.
Can knowledge become stale even if I remember it?
Yes. Current rules, software, prices, standards and professional practices can change. Refresh the source, not only memory.
What comes next?
The next guide shows how to create practice questions with SI that diagnose, build and transfer the exact skill you are targeting.
Helpful Reading
- How to Turn a New Subject into a Learning Roadmap with Super Intelligence
- How to Build a Personal Super Intelligence Tutor
- How to Learn Anything Faster with Super Intelligence
- How to Use Super Intelligence to Challenge Your Assumptions
Find the Gap Before You Fill It
Knowledge is not one solid block.
A learner can know the rule and fail the transfer. Remember the fact and miss the mechanism. Understand the idea and fail under time. Be confident and wrong. Be correct and unsure.
SI becomes genuinely useful when it helps distinguish those states.
The shortest repair begins with the most accurate gap diagnosis.
