De-implementation, low-value practice, stopping ineffective interventions, educational workload, redundant homework, programme withdrawal, intervention exit, implementation science, study efficiency, school improvement, tuition workload and sustainable learning all point toward a problem education rarely names clearly enough: what should we stop doing?
Education is very good at adding. New worksheets. New tuition. New apps. New trackers. New revision routines. New feedback forms. New examination papers. New intervention groups. New parent checks. New prompts. New study blocks. New dashboards. New “must-do” practices.
Education is much worse at subtraction.
A learner repairs a weakness, but the remediation continues. A parent keeps checking every answer long after the child could check independently. A tutor keeps assigning a worksheet because it has always been part of the programme. A school keeps a routine because removing it feels riskier than leaving it. A study habit that once helped a novice becomes friction for an advanced learner. A support scaffold turns into permanent dependency.
Eventually the week fills with educational fossils: routines that had a reason once, but no longer earn the time, attention or adult effort they consume.
Training de-implementation is the deliberate reduction, retirement or removal of low-value, obsolete, redundant or harmful training routines so learner, family and tutor capacity can be reallocated to higher-value learning.
De-implementation is not educational minimalism for its own sake.
It is not “do less” as an ideology.
It is evidence-based subtraction.
If a routine still produces meaningful value at acceptable cost, keep it.
If it no longer does, it should not remain merely because it is familiar.
Quick Read: Training Needs an Exit Path for the System, Not Only the Skill
Identify the Routine → State Its Original Job → Check Current Value → Check Current Cost → Test What Happens If It Is Reduced → Remove, Replace or Retire → Monitor for Loss → Restore Only If Evidence Requires It
A useful de-implementation review asks:
- Why did this routine begin?
- Does that reason still exist?
- What capability is it currently protecting?
- What evidence shows it still adds value?
- What does it cost in time, attention, money, adult support or conflict?
- Could a smaller routine produce the same value?
- Could ordinary curriculum now maintain the capability?
- What would happen if the routine stopped?
- How would we detect harmful loss?
The Current Implementation Frame: Sustain, Adapt, Scale or Stop
EEF’s September 2026 Evaluate and sustain your strategy guidance explicitly asks whether an approach should be sustained, adapted, scaled or stopped. That final verb matters. Good implementation is not defined by preserving every intervention forever. It includes decisions to stop approaches that are not working or no longer deserve continuation.
AERO’s 15 September 2026 guidance on monitoring implementation outcomes complements this by asking teams to monitor feasibility, acceptability, fidelity, reach and sustainability. Those outcomes can reveal not only whether an intervention should continue, but whether continuation is becoming irrationally expensive or poorly aligned.
Implementation science uses the term de-implementation most extensively in healthcare, where low-value or harmful practices may need to be reduced or removed. A 2026 open-access debate article in Implementation Science Communications compares theories and frameworks for de-implementation of low-value care. Education is not healthcare and the specific evidence cannot be transferred directly. The conceptual problem is highly relevant: stopping entrenched practice is a distinct implementation challenge, not simply the reverse of starting something new.
De-implementation Owner Boundary
This article owns one specific question:
How should a training system deliberately remove routines, supports and interventions that no longer earn their cost, without accidentally removing capabilities that still need protection?
It does not re-own:
- Exit Criteria, which owns when a skill can leave active practice.
- Training Maintenance, which owns how stable skills remain available at lower dosage.
- Training Sustainability, which owns what should remain over time.
- The Replanning Trigger, which owns when the current strategy should be abandoned for another.
- Training Adaptation, which owns changing delivery while preserving the mechanism.
De-implementation is subtraction at the system level.
Why Educational Systems Accumulate
Adding feels safer than removing.
If marks fall, add tuition.
If homework is incomplete, add monitoring.
If memory is weak, add flashcards.
If errors recur, add worksheets.
If examinations approach, add past papers.
If parents are anxious, add reports.
Every addition can be individually reasonable.
The combined system can become unreasonable.
The Accumulation Problem
Suppose a Secondary student carries:
- school homework;
- tuition homework;
- vocabulary app;
- flashcards;
- weekly full paper;
- error log;
- parent checklist;
- timed revision block;
- extra grammar worksheet;
- Science notes rewriting;
- Mathematics drill book.
Each routine has a story.
No one has asked which routines should now leave.
The Add-Only Trap
An add-only training system has no natural ceiling.
Every newly identified need becomes additional work.
Eventually the learner’s capacity is consumed by maintenance of the system rather than improvement of capability.
De-implementation Is a Capacity Strategy
Removing low-value work creates space for:
- sleep;
- recovery;
- higher-value practice;
- whole-task integration;
- family time;
- independence;
- the next weak link.
Subtraction is therefore not merely efficiency.
It is reallocation.
Category 1: Obsolete Scaffolds
A scaffold is useful because it is temporary.
Worked steps.
Prompt cards.
Sentence starters.
Method labels.
Parent reminders.
When the learner can perform independently, the scaffold should shrink.
Obsolete Scaffold Example: Mira
Mira originally needs a card listing quadratic methods.
Months later she still consults it before every problem.
Her actual skill may now be strong enough to proceed without it.
Keeping the card could prevent cold-start independence.
De-implement:
- remove on easy cases;
- keep available as emergency backup;
- sample independent performance;
- remove entirely if stable.
Category 2: Permanent Parent Checking
Parent checking can help a young or fragile learner.
It becomes low-value when:
- learner can self-check;
- parent corrections create conflict;
- learner stops taking ownership;
- checking duplicates tutor feedback.
De-implementation should migrate the function:
parent checks everything → parent samples → learner checks → parent only reviews unresolved issues.
Category 3: Redundant Worksheets
Practice volume is often sticky.
A learner improves.
The worksheet quota remains.
Ask:
- Does this worksheet sample something not already covered?
- Does it add transfer?
- Does it add spacing?
- Does it add authentic performance?
- Or is it simply another copy of stable work?
If it adds nothing, remove it.
Category 4: Low-Value Note Copying
Copying notes can organise thinking for some learners.
It can also become activity theatre.
If the learner already has clear notes and copying does not improve retrieval, explanation or application, consider removing the routine.
Category 5: Permanent Error Logs
Error logs are useful during active diagnosis.
They become low-value when:
- errors are no longer reviewed;
- the same resolved errors remain listed forever;
- recording takes longer than repair;
- learner dreads logging and learns nothing from it.
De-implement detailed logging for stable areas.
Keep only recurring high-leverage patterns.
Category 6: Full Papers Used for Every Problem
A full paper is a whole-system sample.
It is not the most efficient repair tool for every weak link.
If one narrow issue is active, reduce full-paper frequency while repairing it.
Return to whole papers for integration and verification.
Category 7: Duplicate Tuition
Families sometimes add a second tuition programme because marks fall.
Before sustaining both, ask:
- Do they own different jobs?
- Do they duplicate materials?
- Do they conflict in methods?
- Does combined homework exceed capacity?
- Is one support now redundant?
More support can create less usable time.
Category 8: Obsolete Diagnostic Monitoring
A target skill stabilises.
The tutor continues measuring cue level, latency and confidence every session.
Monitoring itself becomes low-value.
Move to periodic maintenance sampling.
Category 9: Old Remediation Targets
A learner was weak at fractions in Primary 5.
Two years later, every Mathematics plan still includes fraction drills “just in case.”
If current fresh evidence shows stability, retire the routine.
Category 10: Motivational Systems That Outlive Their Job
Tokens, streaks and rewards can help initiate behaviour.
When routine becomes stable, gradually remove the external system.
Otherwise motivation support becomes permanent administrative work.
Category 11: Over-Specified Study Timetables
A detailed timetable can help during transition.
An older independent learner may later need only weekly priorities and deadlines.
De-implement unnecessary micro-scheduling as self-regulation improves.
Category 12: Model Answers That Become Crutches
Model answers help learners see structure.
Once the route is understood, overuse can replace generation.
Remove models progressively.
Category 13: Chapter Labels
Topic labels support early practice.
If method recognition is now the target, de-implement labels.
Category 14: Excessive Confidence Ratings
Confidence tracking may help calibration during diagnosis.
Do not require a rating after every trivial item forever.
Use selective high-value sampling.
Category 15: Redundant Feedback
Three adults correct the same homework.
School teacher.
Tutor.
Parent.
The child receives duplicate commentary and less time to repair.
Clarify ownership.
Category 16: Permanent Worked Examples
Worked examples should fade as expertise grows.
If learners continue to read solutions before attempting, independent problem solving may remain underdeveloped.
Category 17: Permanent Vocabulary Lists
A word moves from active acquisition to stable use.
It does not need equal review forever.
Increase spacing or remove from active deck.
Category 18: Obsolete Flashcards
Flashcard systems become bloated when stable items never leave.
Use graduation rules.
Category 19: Low-Value Homework Rituals
Some routines continue because they signal seriousness:
- rewrite every correction three times;
- copy definitions repeatedly;
- redo entire worksheets after one error.
Ask what learning mechanism they serve.
If the answer is unclear, test removal.
Category 20: Data That No Longer Changes Decisions
Every tracked metric consumes attention.
If a field never changes the plan, archive it.
De-implementation Type 1: Stop Completely
Use when:
- routine is ineffective;
- routine is harmful;
- routine duplicates another system;
- capability is stable and maintenance does not require it.
De-implementation Type 2: Reduce Dosage
Full weekly practice becomes monthly maintenance.
This is the most common safe form.
De-implementation Type 3: Narrow Scope
A routine once applied to every subject.
Keep it only where the learner still needs it.
De-implementation Type 4: Remove Support, Keep Task
Parent checking stops.
Learner still completes the task.
De-implementation Type 5: Remove Task, Keep Function
Dedicated checking worksheet disappears.
Checking becomes embedded in ordinary Mathematics.
De-implementation Type 6: Replace With Lower-Cost Equivalent
Full error log becomes three recurring error codes.
Learning function remains.
De-implementation Type 7: Pause and Monitor
Stop a routine temporarily.
Watch for performance drift.
Restore only if needed.
De-implementation Type 8: Transfer Ownership
Tutor no longer runs the routine.
Learner continues it independently.
External intervention is de-implemented; internal capability remains.
De-implementation and Exit Criteria
Exit Criteria provides evidence for skill-level de-implementation.
Once a learner can perform:
- independently;
- on fresh tasks;
- after delay;
- under relevant variation;
dedicated practice may leave active training.
De-implementation and Maintenance
Stopping active intervention does not always mean stopping contact.
Training Maintenance provides the lower-dose alternative.
De-implementation and Sustainability
Training Sustainability requires subtraction.
If nothing ever leaves, no system can remain sustainable.
De-implementation and Feasibility
Removing low-value work creates feasibility.
A plan may become workable without reducing ambition if redundant activity is removed.
De-implementation and Acceptability
Learners accept systems better when they see that solved problems actually leave the plan.
Permanent remediation undermines trust.
De-implementation and Fidelity
Be careful.
Some routines look expendable but carry an active ingredient.
Do not remove fresh retesting from a repair loop simply because it takes time.
That would reduce fidelity, not low-value burden.
De-implementation and Adaptation
Training Adaptation asks how to change form while keeping function.
De-implementation asks whether a function should continue to receive dedicated system resources at all.
De-implementation and Reach
When dedicated reach is no longer needed, contract it.
Move from direct intervention to ordinary curriculum.
De-implementation and Replanning
If an intervention remains high cost and ineffective despite adequate fidelity, stopping it may be better than endlessly adapting it.
De-implementation and Training Review
Every review should include one subtraction question:
What can leave the system now?
De-implementation and Training Priorities
Priorities are impossible if every old priority remains active.
Subtraction makes priority real.
De-implementation and Training Load
Total training load can fall without losing capability when mature routines move to maintenance.
De-implementation and Proxy Failure
The site’s Proxy Failure article warns that visible metrics can improve while capability does not.
De-implement proxy-driven tasks that no longer contribute to authentic performance.
De-implementation and Path Dependence
The site’s Path Dependence article matters because early routines can become embedded institutions.
The longer a low-value routine remains, the harder it becomes to remove.
Why Low-Value Routines Become Sticky
Several forces protect them.
- Habit.
- Sunk cost.
- Fear of regression.
- Visible completion.
- Parent expectation.
- Teacher convenience.
- Commercial packaging.
- Data systems built around the routine.
- No one owns the decision to stop.
The Sunk-Cost Trap
“We bought the workbook, so we should finish it.”
Past cost does not justify future low-value work.
The Completion Trap
“We are already on Unit 8 of 10.”
Finishing a sequence is psychologically satisfying.
It is not an educational reason if the learner no longer needs it.
The Fear-of-Regression Trap
Adults worry that stopping practice will cause loss.
Use maintenance testing rather than permanent high dosage.
The Visibility Trap
Worksheets are visible.
Independent thinking is harder to see.
Systems often keep visible activities because they reassure adults.
Evidence should decide.
The Commercial Trap
A programme may be sold as a fixed package.
That does not mean every component remains necessary for every learner.
Training architecture should follow the learner state.
The “More Is Safer” Trap
Adults assume extra practice cannot hurt.
But extra practice consumes:
- sleep;
- attention;
- family time;
- another subject;
- recovery;
- motivation.
Low-value work has opportunity cost.
Mira Case Study: The Permanent Algebra Workbook
Mira was weak at algebra six months ago.
Her tuition plan still includes a weekly algebra workbook.
Fresh evidence shows:
- execution stable;
- mixed method selection remains weak;
- graph interpretation emerging.
The workbook consumes forty minutes.
De-implementation decision:
- remove weekly blocked algebra workbook;
- embed algebra in mixed Mathematics;
- use freed time for method selection and graphs.
Nothing important was abandoned.
Capacity was redirected.
Mira Case Study: Parent Checking
Mira’s parent has checked every Mathematics answer since Primary school.
At Secondary level, this now delays independence.
De-implementation sequence:
- Week 1: parent samples two questions.
- Week 2: Mira self-checks; parent reviews one unresolved error.
- Week 3: no routine parent checking.
- Month end: parent reviews school result only.
Checking function migrates rather than vanishes.
Mira Case Study: Formula Card
The formula card once reduced working-memory load.
Mira now retrieves formulas accurately.
Keep the card in the bag but remove it from normal practice.
Cold-start performance confirms readiness.
The support is de-implemented.
Jonas Case Study: Full Essay Twice a Week
Jonas’s writing intervention began with two full essays weekly.
Later evidence shows his weakness is narrow: paragraph development on unfamiliar topics.
Two full essays now create fatigue and avoidance.
De-implement one full essay.
Replace with two targeted paragraph sessions.
Keep one weekly whole composition for integration.
Jonas Case Study: Grammar Corrections
Grammar was once a major weakness.
It is now stable.
The tutor still marks every minor grammar choice in detail.
De-implement intensive grammar annotation.
Return deep feedback attention to argument and development.
Jonas Case Study: Vocabulary Flashcard Bloat
Jonas has 900 flashcards.
Many stable common words still appear every week.
De-implementation:
- archive stable words;
- widen spacing;
- keep active only unstable high-value vocabulary;
- test archived items occasionally through reading.
Nadia Case Study: Science Note Rewriting
Nadia rewrites Science notes because the routine began during a period of weak organisation.
Her notes are now clear.
Retrieval and application remain the actual needs.
Stop rewriting.
Use that time for closed-book recall and unfamiliar evidence questions.
Nadia Case Study: Permanent Model Answers
Nadia reads model explanations before every new Science question.
This support once taught structure.
It now blocks independent generation.
De-implement pre-exposure.
Move model answers to post-attempt comparison.
Evan Case Study: Peer Answer Sharing
Evan and Mira developed a habit of sending completed solutions before discussing homework.
The routine is socially efficient and educationally costly for independent method selection.
De-implement answer-first sharing.
Keep the peer relationship.
Replace sequence with:
attempt → compare method → discuss → fresh individual check.
De-implementation in a Three-Student Class
Different learners should leave supports at different times.
Mira no longer needs method cards.
Jonas still does.
Nadia has moved to transfer tasks.
Do not keep all learners on the same support merely for administrative convenience.
Group De-implementation
A shared routine can leave the whole class when:
- all learners meet exit criteria;
- ordinary whole-task work now maintains the capability;
- the routine consumes time better used elsewhere.
Individual De-implementation
One learner may leave while peers continue.
Normalise this:
Different next jobs.
De-implementation and Fairness
Removing work from one learner can look unfair.
Explain that volume follows current need.
Fairness is not identical homework.
De-implementation and Learner Motivation
Seeing routines disappear after mastery improves trust.
The learner learns:
If I stabilise this skill, I graduate from the intervention.
De-implementation and Adult Anxiety
Parents can fear that reduced practice means lower standards.
Show fresh evidence and maintenance plan.
De-implementation and Tutor Identity
Tutors can become attached to favourite methods.
A method’s past success does not justify indefinite use.
The learner state decides.
De-implementation and School Requirements
Some low-value-feeling tasks are school requirements.
They cannot simply be removed.
Instead:
- complete efficiently;
- avoid duplicating them in tuition;
- add only what school work does not cover.
De-implementation and Mandatory Practice
Distinguish:
- non-negotiable requirement;
- optional supplementary training;
- obsolete legacy routine.
Only the latter two are under direct de-implementation control.
De-implementation and AI
AI creates new routines quickly.
Prompt templates.
Daily quizzes.
Automated feedback.
Generated flashcards.
Every new tool should have an exit rule.
AI Routine Exit Questions
- Does the tool still save time?
- Does it still improve learning?
- Has the learner internalised the supported function?
- Is verification cost too high?
- Does the tool now create dependence?
De-implementing AI Support
Move from:
- AI generates → learner edits
toward:
- learner generates → AI critiques → learner revises
Then eventually:
- learner generates and self-checks → AI used selectively.
De-implementation and Digital Platforms
Do not keep a platform because data history is impressive.
If usage is burdensome and learning value low, migrate or stop.
De-implementation and Data Dashboards
Metrics can outlive decisions.
Archive dashboards that no longer influence action.
De-implementation and Notifications
Reminders can help launch routines.
Later they become noise.
Fade reminders as initiation becomes self-controlled.
De-implementation and Streaks
Streaks can create pressure after routine is established.
Remove them when they no longer add useful behaviour.
De-implementation and Rewards
External rewards should reduce as behaviour becomes internalised.
Do not keep rewarding a mature routine forever.
De-implementation and Study Plans
Detailed plans can become lighter.
Primary learner: daily checklist.
Older independent learner: weekly priorities.
Support should mature.
De-implementation and Monitoring
Close monitoring is useful during uncertainty.
Once stable, sample rather than track continuously.
De-implementation and Privacy
Less monitoring can also restore learner privacy and autonomy.
The De-implementation Evidence Test
Before stopping a routine, ask:
- What evidence says the original problem is solved or changed?
- What evidence says the routine itself adds little value?
- What lower-cost mechanism can maintain the capability?
- What indicator will reveal harmful regression?
The Trial Withdrawal
When uncertain, use a reversible test.
Reduce or pause the routine.
Monitor fresh performance.
If stability remains, continue removal.
If meaningful decline appears, restore at the smallest effective dose.
The Withdrawal Window
Choose a window long enough for loss to become visible if the routine was actually necessary.
Duration depends on the skill.
Do not impose one universal interval.
De-implementation and Measurement Noise
One bad result after stopping does not prove the routine was necessary.
Use representative evidence.
De-implementation and Regression to the Mean
The site’s HPL Regression to the Mean owner warns against over-crediting interventions after extreme outcomes.
The same caution applies when deciding whether stopping caused a dip.
De-implementation and Counterfactual Thinking
If a routine stops and performance remains stable, that supports the view that the routine may no longer be necessary.
It does not prove the routine never contributed historically.
Historical Value vs Current Value
A routine can have been valuable and now be obsolete.
Do not rewrite history to justify removal.
Say:
This helped then. It is no longer needed now.
The Sunset Clause
Every new intensive routine should include a review date.
Example:
Run for three weeks, review fresh transfer, then continue, reduce or stop.
Sunset clauses prevent accidental permanence.
The Exit Budget
Each term, identify at least one routine that can:
- stop;
- shrink;
- move to maintenance;
- transfer to learner control.
This keeps system load bounded.
The One-In, One-Out Principle
When adding a substantial new training demand, ask what can leave.
This is not a rigid law.
It is a capacity discipline.
When One-In, One-Out Is Wrong
During genuine transition, temporary load may increase.
But temporary should have an end point.
De-implementation and Scaling
At scale, low-value routines become expensive quickly.
A ten-minute unnecessary step multiplied across hundreds of learners is substantial system cost.
The companion Training Scale article owns how expansion changes implementation economics.
The De-implementation Ladder
Before full removal:
- reduce frequency;
- reduce duration;
- reduce scope;
- transfer ownership;
- embed in normal work;
- pause and monitor;
- stop.
Use the least disruptive subtraction that achieves the purpose.
The De-implementation Traffic Light
- Green to remove: stable capability, low added value, high redundancy, maintenance available.
- Amber: value uncertain; reduce and monitor.
- Red to remove: active ingredient still necessary, fresh performance unstable, no maintenance route.
The Parent De-implementation Audit
- What routine are we still doing because it used to be needed?
- What can the child now do independently?
- Which parent checks are duplicated by tuition?
- Which worksheets repeat stable work?
- Which apps or subscriptions are no longer useful?
- Which routine causes conflict without clear learning value?
- What can move to maintenance?
- What can stop?
- How will we know if stopping was a mistake?
The Tutor De-implementation Audit
- Which active targets have met exit criteria?
- Which scaffolds should fade?
- Which monitoring fields can be removed?
- Which homework tasks duplicate school work?
- Which favourite routines no longer fit the learner?
- Which full-paper tasks can be replaced by targeted repair?
- Which targeted repair can now return to whole-task maintenance?
- What is taking tutor time without changing decisions?
- What can the learner own?
The Learner De-implementation Audit
- What help do I no longer need?
- What routine feels repetitive because the skill is stable?
- What tracker no longer helps?
- What study activity feels productive but does not improve performance?
- Which old weak area is no longer weak?
- What can I maintain through normal schoolwork instead?
The De-implementation Field Manual: Eighty Questions
- Why did this routine begin?
- What problem was it solving?
- Does that problem still exist?
- What evidence says yes?
- What evidence says no?
- What active ingredient does the routine carry?
- Can another lower-cost routine carry it?
- Is the skill stable?
- Is it stable on fresh tasks?
- Is it stable after delay?
- Is it stable independently?
- Is it stable under relevant variation?
- Has support dependence fallen?
- Has the learner cold-started?
- Has the learner self-cued?
- Is the routine duplicated at school?
- Is it duplicated at tuition?
- Is it duplicated at home?
- Is three-way feedback occurring?
- Is the routine high burden?
- Is it high parent burden?
- Is it high tutor burden?
- Is it high learner burden?
- Is it high financial cost?
- Is it high travel cost?
- Is it consuming sleep?
- Is it consuming another subject?
- Is it creating family conflict?
- Is it creating avoidance?
- Is it still acceptable?
- Is it still feasible?
- Is it implemented faithfully?
- Does high fidelity produce useful outcomes?
- Has the metric improved while capability has not?
- Is the routine measuring more than teaching?
- Does the data change decisions?
- Can tracking shrink?
- Can dosage shrink?
- Can frequency shrink?
- Can scope shrink?
- Can the task embed in normal work?
- Can the learner own it?
- Can the parent step back?
- Can the tutor step back?
- Can the prompt migrate?
- Can the tool be removed?
- Can the model answer disappear?
- Can the formula sheet disappear?
- Can the checklist disappear?
- Can the reward disappear?
- Can the notification disappear?
- Can the leaderboard disappear?
- Can the second tuition block disappear?
- Can the extra workbook disappear?
- Can the full-paper frequency fall?
- Can the error log simplify?
- Can the vocabulary deck shrink?
- Can stable flashcards archive?
- Can note copying stop?
- Can the learner use retrieval instead?
- Can the support move to maintenance?
- What is the lowest-cost replacement?
- What is the withdrawal risk?
- What indicator will monitor that risk?
- How long should the withdrawal test run?
- What would trigger restoration?
- What would trigger full removal?
- What would trigger a different intervention?
- Are we protecting the routine because of sunk cost?
- Are we protecting it because completion feels good?
- Are we protecting it because adults are anxious?
- Are we protecting it because no one owns stopping?
- What capacity would removal free?
- Where should that capacity go?
- What new bottleneck deserves the time?
- What sleep or recovery would be restored?
- What family friction would reduce?
- What independence would increase?
- What is the sunset clause?
- When is the next stop-or-continue review?
The De-implementation Design Template
Routine: what are we considering removing?
Original job: why did it begin?
Current learner state: does the original problem remain?
Current value: what benefit does the routine still add?
Current cost: time, burden, money, support, conflict?
Active ingredient: what function must be preserved if the routine changes?
Removal mode: reduce, pause, transfer, embed, stop?
Maintenance route: what lower-cost contact remains?
Risk indicator: what would show harmful regression?
Review date: when will the withdrawal be evaluated?
Worked De-implementation Design: Mathematics
Routine: weekly blocked algebra workbook.
Original job: stabilise algebraic execution.
Current state: execution stable; method selection weak.
Value: low incremental value.
Cost: forty minutes weekly.
Active ingredient to preserve: algebra maintenance.
Removal: stop workbook.
Maintenance: algebra embedded in mixed work.
Risk indicator: execution errors recur across fresh tasks.
Review: four weeks.
Worked De-implementation Design: English
Routine: two full compositions weekly.
Original job: improve writing fluency and structure.
Current state: whole-task structure stable; idea development remains weak.
Value: one full composition still valuable, second is redundant.
Cost: high time and avoidance.
Removal: reduce to one full composition.
Replacement: two short development drills.
Risk indicator: whole-task integration declines.
Worked De-implementation Design: Science
Routine: weekly note rewriting.
Original job: organise concepts.
Current state: notes clear; recall and application more important.
Value: low.
Cost: thirty minutes weekly.
Removal: stop rewriting.
Replacement: retrieval + one unfamiliar evidence question.
Risk indicator: concept organisation becomes confused.
Research Foundations and Evidence Boundaries
The education-specific practice anchor is EEF’s September 2026 Evaluate and sustain your strategy, which explicitly frames ongoing implementation decisions as whether an approach should be sustained, adapted, scaled or stopped. That “stopped” category is crucial because educational improvement cannot be add-only.
AERO’s 15 September 2026 Staying on track guide supports the monitoring logic needed before stopping: feasibility, acceptability, fidelity, reach and sustainability provide evidence about whether a practice remains viable and useful.
The formal de-implementation literature is more developed in healthcare than in education. The 2026 open-access article De-implementation of low-value care: contrasting theories, models, and frameworks to generate methodological guidance shows that removing low-value practice is itself an implementation problem with distinct theories and strategies. The specific domain and recommendations should not be transferred mechanically into education. The conceptual transfer used here is limited: low-value practices can become entrenched, stopping requires deliberate implementation, and withdrawal should be monitored rather than assumed harmless.
The claims in this article are therefore bounded. There is no universal formula for deciding which educational routine should be removed or exactly when. De-implementation should rely on current learner evidence, opportunity cost, implementation burden, maintenance alternatives and fresh monitoring after withdrawal. It should not be used as a shortcut to remove difficult but valuable practice.
Final Principle
A mature training system is not defined by how much it contains.
It is defined by how well it distinguishes what still earns a place.
Some routines are essential.
Some are temporary scaffolds.
Some are maintenance.
Some are legacy.
Some are simply habits with educational language around them.
The job is to tell the difference.
Good training does not only know what to add. It knows what to retire, what to reduce, what to hand over to the learner and what to stop before yesterday’s solution becomes tomorrow’s unnecessary workload.
