A student has revised algebra every night for two weeks.
They score well.
They keep revising algebra because it now feels safe.
Meanwhile, chemistry remains unstable.
A revision stopping rule is the evidence required before a topic can leave active repair and move into lower-frequency maintenance, freeing scarce time without pretending the topic can never weaken again.
This is the fourth pillar beneath How Revision Works. The master owns revision as a continuing loop. This page owns the exit condition: when is enough active revision enough?
Quick Read
A topic should not leave active revision simply because it feels familiar, was completed once, or crossed one arbitrary percentage threshold. Stronger evidence is repeated performance under progressively less supportive conditions: accurate retrieval after delay, changed-cue questions, successful discrimination from neighbouring topics, acceptable execution speed and no recurring high-consequence error. Graduation does not mean deletion from revision. It means the topic moves from active repair to maintenance, where mixed papers, occasional spot checks and scheduled reviews continue to sample it. If errors reappear, the topic re-enters the active queue. Stopping rules therefore prevent two opposite failures: abandoning fragile topics too early and endlessly polishing already-secure topics because success feels rewarding.
active repair → independent success → delayed success → changed cue → mixed discrimination → execution check → graduate to maintenance → periodic resampling → stable / re-enter active queue
Revision Without an Exit Rule Expands Forever
Every topic remains “to revise.”
Every old chapter still appears on the list.
The learner cannot tell whether today’s session is:
- repair;
- maintenance;
- reassurance;
- habit;
- avoidance of harder topics.
Stopping rules give revision a controlled release mechanism.
“Finished the Chapter” Is Not a Stopping Rule
Notes completed.
Worksheet completed.
Video watched.
None proves future access.
Revision should stop active repair because the capability has demonstrated stability—not because the material has been consumed.
One Correct Attempt Is Too Weak
Correct immediately after teaching.
Could reflect recency.
Correct after one familiar flashcard.
Could reflect cue dependence.
Correct in one blocked worksheet.
Could reflect category support.
Graduation needs multiple kinds of evidence.
Delayed Retrieval Is the First Stronger Gate
The topic returns after enough time for immediate activation to fade.
Can the learner still retrieve:
- the core fact;
- the governing relationship;
- the procedure;
- the conditions of use?
Retrieval Strength & Storage Strength explains why this delayed evidence is stronger than end-of-session fluency.
Changed Cues Are the Second Gate
Same idea.
Different wording.
Different diagram.
Different example.
Different question order.
If retrieval collapses, the topic is not ready to graduate.
Retrieval Cues owns the prompt-dependence mechanism.
Mixed Discrimination Is the Third Gate
A method may be stable when practised alone and unstable when its nearest neighbour appears beside it.
Before graduation, test:
- method A versus method B;
- concept X versus concept Y;
- similar vocabulary;
- related historical events;
- nearby formulas.
A topic that cannot reliably beat its nearest competitor still needs active discrimination work.
Transfer Is the Fourth Gate
Can the learner use the knowledge when surface features change?
Retrieval Transfer owns the broader mechanism.
The stopping implication:
do not graduate knowledge that works only inside the practice format that trained it.
Execution Stability Is the Fifth Gate for Exam-Bound Knowledge
A learner can solve correctly with unlimited time.
Under realistic time, the method is too slow.
For an examination goal, active revision may still be needed.
Execution stability can include:
- reasonable response latency;
- clean algebra/writing;
- unit control;
- checking;
- recovery from a small error;
- selection under time.
No Universal Percentage Can Be the Stopping Rule
“80% means mastered.”
Why 80?
On what question set?
With what cueing?
After what delay?
For what consequence?
A percentage can be part of the evidence.
It should not erase the conditions under which the score was produced.
Stopping Rules Should Be Risk-Sensitive
Low-consequence spelling word:
one or two successful delayed checks may be enough for maintenance.
Foundational algebra skill:
requires stronger evidence because many later tasks depend on it.
High-mark examination method:
may require mixed and timed verification before active release.
higher consequence → stronger graduation evidence.
Prerequisites Need Conservative Graduation
If weak manipulation destroys multiple later topics, the cost of false graduation is high.
Keep foundational skills exposed through mixed maintenance even after active repair ends.
Recurring Error History Raises the Bar
A sign error appeared in:
- three homework sets;
- two timed sections;
- one full paper.
One clean question today should not erase the recurrence history.
Require several clean opportunities before graduating the mechanism.
Confidence Must Be Calibrated Before Graduation
A student is correct but believes every answer is a guess.
Another is wrong and highly confident.
Stopping active content repair can still be appropriate for the first learner while calibration work continues separately.
The second learner may need active error discrimination even if occasional scores look acceptable.
How Confidence Works remains the calibration owner.
Graduation Means a State Change, Not Disappearance
Active queue:
frequent targeted repair.
Maintenance:
lower-frequency sampling through:
- mixed questions;
- timed sections;
- full papers;
- spot checks;
- spaced retrieval.
The topic remains in the learning system, but stops demanding focused repair time.
Revision Scheduling Owns the Maintenance Return
Once a topic graduates, Revision Scheduling decides how long it can remain away before another sample is needed.
Stopping does not terminate scheduling.
It lowers its frequency and changes its form.
Revision Triage Benefits Immediately From Graduation
A secure topic leaves the active queue.
Time becomes available for the next weak prerequisite or recurring error.
Revision Triage can now reprioritise.
This is why stopping rules are not merely about efficiency.
They improve the whole allocation system.
Phase Transition and Stopping Are Different
A topic can stop Phase 3 blocked practice and enter Phase 4 mixed practice.
That is not graduation.
Revision Phase Transitions changes the test form.
Stopping Rules change the topic from active repair to maintenance after enough phases have demonstrated stability for the intended use.
Do Not Over-Revise Secure Topics Because Success Feels Good
Students naturally prefer:
- questions they can complete;
- topics that look familiar;
- revision that produces high scores;
- activities with quick closure.
This can create a competence trap: secure topics consume time because they reward the learner emotionally while weak topics remain avoided.
Stopping rules protect the learner from their own success bias.
Do Not Under-Revise Because One Good Paper Created Relief
The opposite trap:
one unusually easy paper goes well.
The learner declares the topic finished.
Use multiple evidence sources before graduation:
- different question sets;
- delayed checks;
- different representations;
- mixed contexts.
A Topic Can Graduate at One Level and Remain Active at Another
Vocabulary definition:
secure.
Contextual use in writing:
still weak.
Formula recall:
secure.
Method selection:
still weak.
Graduation should attach to a defined capability, not a vague chapter name.
Use Capability-Specific Exit Criteria
Instead of:
Algebra mastered.
Use:
Linear rearrangement: independent, delayed, mixed accuracy stable; move to maintenance. Algebraic fractions: still requires supported repair. Method selection in mixed word problems: amber.
Maintenance Frequency Should Decline With Stable Evidence
First maintenance check after graduation:
relatively soon.
If clean:
longer gap.
If repeated clean paper exposures continue:
isolated retests may become unnecessary because mixed examination practice is already sampling the skill.
Re-Entry Triggers Must Be Explicit
A maintenance topic returns to active repair when:
- the same error recurs twice;
- delayed retrieval fails;
- a new related topic creates interference;
- exam format changes;
- timing collapses;
- confidence becomes badly miscalibrated;
- teacher feedback reveals a hidden misconception.
The exact trigger can vary.
What matters is that graduation is reversible.
One Error Does Not Always Require Re-Entry
A stable topic produces one random slip.
If the learner immediately detects and explains the error, active repair may be unnecessary.
Re-entry should respond to mechanism recurrence or meaningful instability, not every isolated imperfection.
Near the Examination, Stopping Rules Tighten
A topic that is probably stable might be allowed a long maintenance gap in January.
Two weeks before a major examination, a final verification can be rational because the cost of unnoticed drift is now higher and there is little time for another cycle.
But Last-Minute Overchecking Can Be Counterproductive
Secure topics are repeatedly tested the night before an exam.
Sleep is lost.
Anxiety rises.
No meaningful new learning occurs.
Stopping rules should include a final boundary where further revision has lower expected value than rest and execution readiness.
Revision Completion Is a Portfolio of States
At any moment, the syllabus may contain:
- active red repairs;
- amber discrimination topics;
- green maintenance topics;
- full-paper execution issues;
- recently graduated topics awaiting first maintenance check.
“Revision complete” therefore does not mean every topic has vanished from the system.
It means no important capability remains in an unmanaged failure state.
A Practical Graduation Rule
For a typical school topic, consider moving from active repair to maintenance when the learner has demonstrated:
- independent retrieval;
- accuracy across more than one attempt;
- success after delay;
- reduced cue dependence;
- changed-cue or changed-representation success;
- discrimination from likely confusions;
- acceptable execution for the intended task.
Then schedule a maintenance check instead of continuing focused repetition.
This is a template, not a universal numerical standard.
A Better Revision-Stopping Model
active repair → repeated independent success → delay → changed cue → mixed competitor test → execution check → risk-sensitive graduation → maintenance schedule → mixed/paper sampling → stable / re-entry trigger → targeted repair
A 30-Lens Revision Stopping Audit
- Capability: what exactly is being graduated?
- Initial failure: why was it active?
- Repair: was the mechanism corrected?
- Independent success: can it perform without notes?
- Repeat success: was it more than one lucky attempt?
- Delay: did it survive time?
- Cue dependence: how much support remains?
- Wording: can the prompt change?
- Representation: can form change?
- Transfer: can the idea apply in a variant?
- Competition: can nearest alternatives be rejected?
- Interference: does a similar topic still intrude?
- Execution: is performance accurate enough?
- Latency: is access fast enough?
- Checking: can ordinary slips be detected?
- Recurrence: how often did this error historically return?
- Prerequisite: what depends on it?
- Consequence: how costly is false graduation?
- Threshold: is a percentage being used without context?
- Confidence: is self-belief calibrated?
- Phase: has the topic survived the relevant practice phases?
- Maintenance: what lower-frequency exposure will remain?
- Paper sampling: will full papers naturally retest it?
- Rare topic: could it disappear too long?
- Graduation date: when did active repair stop?
- Next check: when is maintenance due?
- Re-entry trigger: what evidence reactivates repair?
- Exam horizon: is final verification needed?
- Opportunity cost: what weakness receives the freed time?
- World return: after leaving active revision, does the capability continue to perform when later tasks unexpectedly call it back?
Laboratory 1: Graduate One Topic
Choose one apparently secure topic. Test it after delay, under changed wording and beside its closest competitor. Decide whether it truly belongs in maintenance.
Laboratory 2: Build Re-Entry Triggers
For three graduated topics, define what evidence would send each back to active repair. Make the trigger specific enough to act on.
Laboratory 3: Find Over-Revision
Audit last week’s revision time. Identify topics that received repeated work despite stable delayed/mixed performance. Reallocate that time to higher-priority active repairs.
For Primary Readers
When you can spell a word correctly today, tomorrow and later in a sentence, you do not need to practise it ten times every night. Put it in the “check sometimes” pile and spend more time on the words you still miss.
For Secondary Readers
Create graduation criteria that require delayed retrieval, changed questions and mixed discrimination before moving a topic from active revision to maintenance.
For Advanced Readers
Model revision stopping as a risk-sensitive state transition from intensive intervention to sparse monitoring. Graduation requires sufficient posterior evidence that future failure probability has fallen below the action threshold for the intended consequence, while reversible maintenance and re-entry preserve responsiveness to drift.
Common Misconceptions
- “Mastered means never revise it again.” Graduation means maintenance, not permanent deletion.
- “One high score means the topic is done.” Recency, cueing and an easy question set can inflate one result.
- “Use the same mastery percentage for everything.” Consequence, delay, transfer and task conditions matter.
- “Every mistake after graduation means the topic was never mastered.” One isolated slip can occur without systemic instability.
- “More revision is always safer.” Over-revision has opportunity costs and can crowd out sleep or genuinely weak topics.
Research Corridor
- US Institute of Education Sciences — Organizing Instruction and Study — spacing, active retrieval and metacognitive study allocation.
- Bjork Learning and Forgetting Lab — retrieval/storage strength, desirable difficulties and delayed performance.
- eduKateSG — Revision Scheduling.
- eduKateSG — How Revision Works.
Frequently Asked Questions
When can I stop revising a topic actively?
When it has repeatedly performed independently, after delay, under changed cues and relevant mixed conditions at a level appropriate to the task’s consequence. Then move it into maintenance rather than removing it entirely.
Is 80% enough to call a topic mastered?
Not by itself. A score needs context: question difficulty, delay, cues, transfer, recurrence history and the cost of failure.
What if a graduated topic starts failing again?
Use the pre-defined re-entry trigger: diagnose the new failure, return the relevant capability to active repair, then require fresh delayed/mixed evidence before graduating it again.
Final Thought: Revision Needs the Discipline to Stop
Stopping is not giving up.
It is releasing secure knowledge from expensive active repair so the next unresolved weakness can receive attention.
A good stopping rule says: this capability has earned a quieter place in the system—but we know exactly what evidence would call it back.
REVISION · FOUR PILLAR LEGS
Return to How Revision Works, or continue through Revision Scheduling, Revision Triage and Revision Phase Transitions. Return to the How X Works Hub.