Implementation fidelity, training fidelity, evidence-based practice, implementation monitoring, school implementation, learning outcomes, active ingredients, dosage, quality of delivery and adherence all point toward one deceptively simple question: did the training system that reached the learner still resemble the training system we believed we had designed?

A study plan can look excellent on paper and disappear in practice. A parent may say the child is following a revision timetable while the actual week contains missed sessions, shortened retrieval work, skipped feedback, repeated easy questions and late-night catch-up. A tutor may believe a carefully sequenced intervention is running while school tests, homework emergencies and family logistics continually interrupt the sequence. A learner may appear to have completed the planned dosage while the central mental operation was never performed.

This is why training fidelity matters. Before asking whether a method worked, we need to know whether the method was actually implemented strongly enough to deserve that question. Current implementation guidance uses related terms such as feasibility, acceptability, fidelity, reach and sustainability because outcomes depend not only on what an intervention is supposed to do but on how it is enacted in real settings.

Training fidelity is the degree to which the training system actually delivered to the learner preserves the active ingredients, sequence, dosage, decision rules and quality conditions that the intended plan depended on.

Fidelity does not mean robotic obedience to a script. Good educational implementation needs adaptation. A plan that cannot adapt to illness, examinations, changing school topics or learner response is not robust. The fidelity question is narrower: when we adapt, which parts may flex and which parts must remain intact for the training mechanism still to be the mechanism we think it is?

---

Quick Read: Plan → Delivery → Fidelity Check → Outcome Interpretation

Specify the Training Job → Identify Active Ingredients → Deliver → Record Meaningful Deviations → Check Dosage and Quality → Interpret Outcomes → Adapt Deliberately

When fidelity is high, outcome interpretation becomes clearer. If Mira’s method-selection accuracy improves after a contrast-and-retest cycle that was actually delivered as designed, the training mechanism becomes a plausible explanation. If nothing improves despite high fidelity, the plan itself deserves review.

When fidelity is low, conclusions become ambiguous. If the planned retrieval sessions were replaced by rereading, a poor retention result does not tell us whether retrieval failed. It tells us the intended retrieval intervention was not sufficiently implemented.

That distinction protects both learners and methods from unfair conclusions.

Why Fidelity Is an Educational Problem, Not Merely an Administrative One

Schools and families often talk about implementation as though it were logistical: whether the worksheet was assigned, whether the lesson happened, whether the homework was completed, whether the programme reached Week 6.

But educational fidelity concerns the learning operation.

A retrieval-practice session can occur administratively while retrieval fidelity is low. The learner may immediately open notes whenever recall becomes difficult. A spaced-practice plan can occur on schedule while spacing fidelity is low because every session repeats only the newest material. A feedback loop can exist in name while feedback fidelity is low because errors are marked but never converted into changed practice.

The relevant question is therefore not:

Did the activity happen?

It is:

Did the activity preserve the mechanism that was supposed to create learning?

The Current Apex Implementation Frame

The Australian Education Research Organisation published Staying on track: Monitoring implementation outcomes on 15 September 2026. Its practical framework asks implementation teams to monitor feasibility and acceptability, fidelity and reach, and sustainability. That framing is useful because it refuses to treat implementation as a binary yes-or-no condition. A practice can be present but poorly delivered, well delivered but unacceptable, acceptable but infeasible, or initially successful but unsustainable.

The Education Endowment Foundation’s A School’s Guide to Implementation similarly treats implementation as a process requiring behaviour, context and structured action. Its Explore, Prepare, Deliver and Sustain phases emphasise the fit between an evidence-informed approach and the real setting.

For eduKatePunggol, the translation is local and practical. A tuition plan lives inside a family week. It interacts with school homework, CCAs, transport, sleep, sibling schedules, examinations, illness, motivation, parent availability and the learner’s changing competence. Training fidelity must therefore be measured in the world where the child actually lives.

Fidelity Owner Boundary: What This Article Owns

This article owns the question:

Did the delivered training preserve enough of the intended system that its outcomes can be interpreted as evidence about that system?

It does not re-own several related mechanisms already established elsewhere.

• Training Consistency owns the repeatability of training over time.
• Training Validity owns whether the measure supports the learner-state interpretation being made.
• Training Environment owns how home, school and tuition shape the conditions around learning.
• Training Review owns whether the plan still fits the learner.
• Training Handoffs owns transfer of responsibility among teacher, tutor, parent and learner.

Fidelity sits between plan and enacted reality.

The Five Questions of Fidelity

A useful fidelity check asks five questions.

• What was essential? Which active ingredients carried the mechanism?
• What was actually delivered? What did the learner genuinely experience?
• How much was delivered? Was dosage sufficient?
• How well was it delivered? Was quality adequate?
• What changed? Which adaptations were deliberate, accidental or forced?

These five questions are enough to prevent many false conclusions.

Active Ingredients: What Must Stay Tight?

EEF implementation guidance uses the useful distinction between knowing where to be “tight” and where to be “loose.” In training, the tight parts are the active ingredients: the elements that must be present for the mechanism to remain recognisably the same.

Suppose the intervention is a contrast-based method-selection routine for Mira.

The active ingredients might be:

• two or more plausible competing methods;
• examples selected to expose the discriminating feature;
• Mira making the method decision before calculation;
• an explanation of the discriminating cue;
• a fresh mixed retest where the method is not named.

The following can be loose:

• the exact numbers;
• the order of examples;
• whether the task is handwritten or digital;
• whether the tutor or learner writes the contrast table;
• whether the retest happens at the end of the lesson or next lesson, within reasonable bounds.

If the loose parts change, fidelity can remain high.

If the active ingredients disappear, the mechanism has changed.

Fidelity Failure: The Method Name Gives Away the Decision

Mira is supposed to train quadratic method selection.

The tutor assigns a worksheet titled Factorisation Practice.

Mira completes it accurately.

Administratively, the session happened.

Mechanistically, method-selection fidelity was low because the worksheet title removed the central decision.

The training can still improve factorisation execution.

It simply cannot be used as evidence that the intended method-selection intervention was delivered.

Adherence: Did We Follow the Critical Sequence?

Adherence is often the first fidelity dimension people imagine.

Was the sequence followed?

But educational adherence should focus on critical order, not ceremonial order.

Consider a repair loop:

Attempt → Diagnose → Targeted Feedback → Fresh Reattempt → Delayed Retest

If the tutor gives feedback before the learner attempts anything, diagnosis becomes weaker.

If the learner corrects the same question but never receives a fresh reattempt, verification disappears.

If delayed retesting never occurs, durability remains unknown.

The fidelity issue is not that Step 4 happened before Step 3 by thirty seconds.

The issue is whether the causal logic of the sequence survived.

Dosage: How Much of the Mechanism Reached the Learner?

Dosage is easy to misread because time and learning opportunity are not identical.

A sixty-minute session can contain five minutes of the target operation.

A twenty-minute session can contain twelve high-quality retrieval attempts and be more faithful to the intended mechanism.

For training, dosage should often be counted in meaningful opportunities rather than minutes alone.

Examples:

• retrieval opportunities;
• method-selection decisions;
• fresh transfer attempts;
• explanations generated;
• feedback-repair cycles;
• independent starts;
• delayed returns.

This is more informative than “studied for one hour.”

Dosage Failure: Time Spent Without the Target Operation

Jonas has a forty-five-minute English tuition session.

The target is evidence-bounded inference.

Twenty minutes are spent reviewing vocabulary.

Ten minutes are spent marking old homework.

Ten minutes are spent discussing school announcements.

Five minutes contain one inference item.

The session may be useful overall.

But the dosage of the inference intervention is low.

If inference performance does not improve, the tutor should not immediately conclude the method failed.

Quality of Delivery: The Same Activity Can Be Delivered Differently

Two tutors can run the same worksheet and create different learning.

One waits for Mira to commit to a method before offering help.

The other signals the correct route through tone and leading questions.

One asks Jonas to justify an inference against textual evidence.

The other accepts plausible language without checking evidence.

One asks Nadia to identify variables independently.

The other names them while explaining the question.

Activity fidelity can be high while quality fidelity is low.

The learner completed the planned materials, but the intellectual demand was softened.

Participant Responsiveness: Was the Learner Actually Engaged in the Intended Operation?

Implementation science sometimes distinguishes what was delivered from how participants received or engaged with it.

That matters enormously in education.

A learner can be physically present while cognitively absent.

More subtly, the learner can engage with the wrong operation.

Nadia is asked to explain why a Science pattern occurs.

She searches memory for a memorised sentence rather than connecting evidence to mechanism.

The tutor hears fluent language and assumes the causal reasoning intervention is working.

Participant responsiveness is low to the intended cognitive job even though visible task completion is high.

Programme Differentiation: Is This Intervention Distinct From Ordinary Practice?

Fidelity also requires knowing what makes the intervention different from the baseline.

If the new plan says “use retrieval practice” but the old homework already required closed-book recall, what changed?

If the new tuition strategy says “give targeted feedback” but the tutor already did so, the intervention may not be meaningfully differentiated.

Without differentiation, outcome claims become confused.

A training plan should specify its active mechanism strongly enough that another tutor could distinguish it from normal practice.

The Fidelity Matrix

A simple fidelity matrix can track four dimensions without drowning in paperwork.

• Active ingredient: What must occur?
• Evidence: What observable trace would show it occurred?
• Minimum sufficient dose: What amount is probably enough to test the mechanism?
• Allowed adaptation: What may change without changing the job?

Example for Mira’s method-selection training:

• Active ingredient: mixed cases with no method label.
• Evidence: Mira chooses and justifies a method before calculation.
• Minimum dose: several decisions across contrasting cases plus fresh retest.
• Allowed adaptation: numbers, surface context, question order.

This is compact enough for real tuition.

Fidelity and Adaptation: The False Choice

Fidelity is sometimes treated as the opposite of adaptation.

That is too crude.

Education requires adaptation because learners and contexts differ.

The better distinction is:

• fidelity-consistent adaptation — changes that preserve the active mechanism;
• fidelity-threatening adaptation — changes that remove or alter an active ingredient.

Changing Mira’s quadratic numbers is fidelity-consistent.

Changing a mixed method-selection task into blocked factorisation is fidelity-threatening if method selection is the target.

Adaptation Example: Illness Week

Mira misses two days of school because she is unwell.

The original tuition plan contains four training blocks.

Forcing all four into the weekend would overload the family.

A fidelity-sensitive adaptation asks:

• Which active ingredient matters most this week?
• Which dosage can be reduced without removing the mechanism?
• What can move into next week?
• What should be skipped rather than compressed into fatigue?

The plan may reduce from four blocks to two.

If those two preserve the key decision, feedback and fresh retest, fidelity can remain adequate even though dosage falls.

Adaptation Example: Examination Week

Nadia’s Science training normally includes slow explanation work.

During examination week, time pressure becomes relevant.

The tutor shortens explanation prompts and adds timed mixed questions.

This can be a fidelity-consistent adaptation if the active ingredient—evidence-to-mechanism reasoning—remains inside the faster format.

It becomes fidelity loss if speed simply replaces reasoning.

Adaptation Example: Learner Progresses Faster Than Expected

Jonas stabilises literal comprehension early.

The original plan scheduled two more weeks.

High fidelity does not require wasting those weeks.

The active ingredient was stable literal evidence selection.

If exit criteria are met, the plan should progress.

Blind adherence after the learning job is complete is not fidelity.

It is bureaucracy.

Adaptation Example: Learner Needs More Support

Mira cannot begin a contrast task independently.

The tutor introduces a small cue from Training Cue Hierarchy.

Does this break fidelity?

Not necessarily.

If the active ingredient is discrimination among competing structures, a small cue that restores attention while preserving the method decision may protect fidelity.

If the cue announces the correct method, fidelity collapses.

Fidelity Drift: Small Changes Accumulate

Implementation drift rarely arrives as one dramatic betrayal of the plan.

It often grows through convenience.

Week 1: fresh retests happen.

Week 2: one is skipped because school homework is heavy.

Week 3: the learner sees answers before reattempting.

Week 4: mixed tasks become blocked because they are easier to mark.

Week 5: feedback becomes general rather than mechanism-specific.

Each change looks small.

The training system has quietly become another system.

Drift vs Evolution

Not every change is drift.

Evolution is deliberate and justified by evidence.

Drift is unexamined change that weakens the relationship between intended and delivered training.

A useful test is:

If we asked why this element changed, could we explain the learner evidence and decision rule that justified the change?

If yes, we may be adapting.

If no, we may be drifting.

Fidelity and the Real Family Week

Training plans fail when they are designed outside family reality.

A Punggol student may have school until afternoon, travel, CCA, dinner, homework, tuition and sleep constraints.

A plan requiring ninety minutes of deliberate practice every weekday may be theoretically impressive and operationally impossible.

This is where fidelity connects to the companion article Training Feasibility.

A plan that cannot be enacted will have low fidelity because the family will inevitably modify it.

The correct response is not to blame the family for “poor compliance.”

It is to redesign the plan so the active ingredients survive the real week.

Fidelity and Training Architecture

Training Architecture turns goals into a capability-building system.

Fidelity protects the architecture after it meets reality.

If the architecture says:

diagnose → isolate → repair → recombine → transfer → verify

then fidelity asks whether those stages actually occurred or whether the weekly schedule collapsed into “finish worksheet and mark.”

Fidelity and Session Design

Session Design owns the structure of one good training session.

Fidelity checks whether the live session maintained that structure.

Suppose the design included:

• retrieval;
• targeted modelling;
• independent practice;
• fresh check;
• exit decision.

If school homework consumes the entire session, the planned session did not run.

That may be necessary on a particular day.

Record it honestly so outcome interpretation remains valid.

Fidelity and Training Constraints

Training Constraints identifies what limits improvement.

Implementation itself can become the constraint.

The training method may be sound.

The limiting factor may be that the method reaches the learner inconsistently, at insufficient dosage or with altered quality.

Before changing the cognitive mechanism, check the implementation constraint.

Fidelity and Training Signals

Training Signals measures change before grades move.

Fidelity provides a second signal stream:

• Was the active ingredient delivered?
• How many relevant attempts occurred?
• What support entered?
• What adaptations occurred?

Outcome signals tell us what changed in the learner.

Fidelity signals tell us what training the learner actually received.

Both are needed for interpretation.

Fidelity and Measurement Validity

Suppose an intervention appears ineffective.

Before concluding that the method failed, ask whether the delivered system retained fidelity.

Otherwise the causal story is invalid.

This is not the same as Training Validity, which concerns whether the measure supports the interpretation.

But they interact:

Valid outcome measurement + unknown implementation fidelity = incomplete conclusion.

Fidelity and Responsiveness

A responsive measure can reveal learning change, but only if the intervention was delivered with enough fidelity for change to be expected.

If Jonas receives one inference-training opportunity instead of the planned repeated contrast-and-retest cycle, a flat measure may reflect insufficient exposure rather than an unresponsive instrument.

Implementation and measurement should be reviewed together.

The Fidelity Paradox: High Fidelity to a Bad Plan

High fidelity is not automatically good.

A poor plan can be implemented perfectly.

This is the fidelity paradox.

A family can follow an excessive worksheet schedule with remarkable discipline and still create fatigue without transfer.

A tutor can deliver a narrow drill exactly as planned while neglecting authentic performance.

Fidelity tells us whether we ran the plan.

It does not tell us whether the plan deserved to be run.

That is why Training Review, Validity, Exit Criteria and outcome evidence remain necessary.

The Reverse Fidelity Problem: A Good Outcome From a Different Mechanism

Sometimes outcomes improve while fidelity is low.

Perhaps the adaptation was better than the original design.

Perhaps another part of the environment caused the improvement.

Perhaps the learner matured or school teaching changed.

Do not force the evidence to credit the original plan.

A low-fidelity success can still teach us something—but often it teaches us that the enacted system deserves separate analysis.

Mira Case Study: The Missing Mixed Practice

Mira’s original goal is to improve Additional Mathematics method selection.

The plan:

• Monday: two contrast pairs.
• Wednesday: mixed classification before solving.
• Friday: fresh parallel form.
• Following Monday: delayed retest.

What actually happens:

• Monday runs correctly.
• Wednesday school homework is urgent, so practice becomes a blocked worksheet.
• Friday the tutor demonstrates three examples because time is short.
• Delayed retest is skipped.

At month end, Mira is faster at factorisation but still selects methods poorly.

Did contrast training fail?

No clean conclusion is possible because the central mixed-selection dosage and verification were largely absent.

The fidelity review reveals that execution practice replaced discrimination practice.

Mira Case Study: Fidelity Restoration

The tutor does not add more total work.

Instead, one blocked worksheet is removed.

The recovered time restores:

• three method-classification decisions;
• one contrast explanation;
• one fresh mixed item.

Total minutes remain similar.

Mechanism fidelity rises sharply.

This illustrates a central principle:

Fidelity improvement is often about replacing low-value activity, not adding workload.

Jonas Case Study: Feedback Without Repair

Jonas’s English plan says every recurring inference error should enter a feedback-repair loop.

In practice, homework is marked with comments:

• “too strong”;
• “not enough evidence”;
• “be more precise.”

Jonas reads the comments.

No fresh inference follows.

Feedback technically occurred.

The active ingredient—using feedback to generate a changed response—did not.

Fidelity is low to the repair mechanism.

Jonas Case Study: The Same Amount of Time, Higher Fidelity

The tutor changes the marking routine.

Instead of commenting on six errors, the tutor selects two.

For each:

• Jonas identifies the evidence boundary;
• rewrites the claim;
• answers one fresh parallel item;
• explains why the new wording is justified.

Fewer errors are discussed.

More learning operations occur.

Training fidelity improves even though coverage appears narrower.

Nadia Case Study: Science Explanation Drift

Nadia’s plan targets evidence-to-mechanism reasoning.

During busy weeks, practice drifts into memorising model answer phrases because that feels efficient.

Nadia’s written fluency improves.

On unfamiliar experiments, explanation quality does not.

The fidelity check reveals that the mechanism changed from causal reasoning to language rehearsal.

The remedy is not more memorisation.

It is restoring causal construction from fresh evidence.

Nadia Case Study: Adaptation Without Fidelity Loss

Examination week requires shorter sessions.

The tutor reduces each explanation to a four-line structure:

• evidence;
• pattern;
• mechanism;
• bounded conclusion.

The session is shorter but the active reasoning sequence remains.

This is high-fidelity adaptation.

Evan Case Study: Peer Practice Changes the Mechanism

Evan, Mira’s school friend, sends her completed solutions over messaging.

Mira uses them to “study the method-selection homework.”

The plan required Mira to choose a method before seeing a solution.

Once the completed solution is visible, the decision disappears.

The peer interaction may still help explanation.

But it is no longer faithful to the intended independent selection task.

The family does not need to ban peer learning.

They can simply sequence it:

Attempt first → compare with Evan second.

Fidelity at Home: The Invisible Parent Prompt

Parents often help in tiny ways:

• “Read that line again.”
• “Remember what the tutor said.”
• “Check the denominator.”

These prompts can be useful.

They also change the delivered condition.

If the training target is independent starting, record that a prompt was required.

Supported success and independent success should not be merged.

Fidelity at School: Topic Alignment Can Distort the Plan

School moves to a new topic.

Tutor and family naturally shift attention toward immediate homework.

The training plan’s older weak link disappears from view.

This is a common source of drift.

Immediate school alignment matters.

But a capability-building programme should protect a small amount of time for the persistent bottleneck until exit criteria are met.

Fidelity Across Handoffs

One teacher says “show full working.”

The tutor says “compress routine steps once stable.”

The parent insists every step be written.

The learner receives three versions of the plan.

Training Handoffs matters because fidelity can break at boundaries between people.

The active ingredient should be stated plainly enough that everyone understands what must remain.

A Fidelity Note for Parents

A parent does not need implementation-science terminology.

A useful home note can be:

This week the important part is not finishing every question. Mira must choose the method before seeing help, explain the cue, and do one fresh question later.

That sentence carries the active ingredients.

A Fidelity Note for Learners

The learner needs even simpler language:

Your job is to decide before you calculate. If someone tells you the method first, we lose the practice we actually need.

Fidelity improves when the learner understands the mechanism.

Fidelity Monitoring Should Be Light

Implementation monitoring can become paperwork theatre.

That is especially dangerous in tutoring, where teacher attention should remain on learners.

Track only decision-relevant elements.

For one active target, a fidelity note might contain:

• target operation delivered? yes/no/partial;
• number of meaningful attempts;
• highest cue level used;
• fresh retest completed? yes/no;
• major adaptation and reason.

Five fields can be enough.

The Fidelity Traffic Light

A simple family-friendly system:

• Green: active ingredients delivered adequately.
• Amber: mechanism present but dosage, support or sequence changed materially.
• Red: intended mechanism largely absent; do not interpret outcome as a clean test of the plan.

This does not need numerical precision.

Its value is interpretive honesty.

Fidelity Thresholds Should Match the Decision

Not every deviation matters equally.

If one session shifts by five minutes, ignore it.

If every fresh retest is skipped, investigate.

If one example is demonstrated rather than independently attempted, note it lightly.

If method selection is always cued, the training target is no longer being tested.

Fidelity monitoring should focus on deviations capable of changing the mechanism.

Fidelity and Reach

AERO pairs fidelity with reach.

That distinction matters when a plan covers multiple learners or contexts.

A tuition strategy may be delivered faithfully to Mira but not to Jonas because his school schedule repeatedly removes the practice block.

The programme has local fidelity but incomplete reach.

For an individual child, reach can also mean whether the plan reaches the settings where the capability matters.

A checking routine used only in tuition but never in school examinations has limited reach into authentic performance.

Fidelity and Feasibility

Low feasibility predicts fidelity problems.

If a plan requires too much time, too many materials, too much parental supervision or too many transitions, people will alter it.

This is not a moral failure.

It is a design signal.

The companion article Training Feasibility therefore asks whether the plan can survive the real week before fidelity is demanded.

Fidelity and Acceptability

A technically sound plan can lose fidelity when the learner dislikes, distrusts or rejects it.

Jonas may resist recording confidence after every item because it feels tedious.

Nadia may hate verbal explanations during every Science question because they interrupt flow.

Instead of forcing full compliance, identify the active ingredient and redesign the form.

Acceptability can support fidelity when adaptations preserve function.

Fidelity and Sustainability

Some plans begin with high fidelity because everyone is motivated.

Three weeks later, monitoring fades, sessions shorten and homework pressure takes over.

Initial fidelity is not sustained fidelity.

The companion article Training Sustainability asks whether the active ingredients can continue after novelty, urgency and adult attention decline.

Fidelity Under Scaling

What works for one student may change when applied to three.

What works for three may change when applied to a school.

Scaling increases the number of handoffs, contexts and implementation decisions.

EEF’s scaling framework explicitly includes feasibility and acceptability as dimensions because an approach that works in a controlled setting may change when scale increases.

eduKate’s three-student structure limits scale deliberately. The small group protects visibility, cue calibration and individual branching—active ingredients that could be diluted in a larger class.

Fidelity and the Three-Student Model

Three learners create enough variety for comparison without making individual routes invisible.

That supports fidelity in several ways:

• the tutor can observe whether each learner performs the target operation;
• support can be adjusted without removing the decision;
• fresh retests can be assigned individually;
• peer explanation can occur without turning the room into anonymous mass practice;
• drift can be detected quickly.

Small-group size is not automatically good pedagogy.

Its value appears when it protects active ingredients.

Fidelity and Learner Agency

Long-term fidelity should not depend forever on tutor surveillance.

The learner should increasingly understand the plan well enough to protect it.

Mira notices when she is about to look at a worked solution before attempting.

Jonas notices when he has accepted feedback without reattempting.

Nadia notices when she is memorising a sentence instead of building a mechanism from evidence.

The learner becomes a fidelity monitor.

This is a route toward Training Independence.

Fidelity and Self-Regulated Learning

Self-regulated learners do not simply follow plans.

They monitor whether the current activity still serves the goal.

That is a form of internal fidelity control.

The student asks:

Am I actually practising the operation I need, or am I doing an easier substitute that feels productive?

This question protects against one of the most common study failures: activity replacing mechanism.

Activity Substitution

Planned retrieval becomes highlighting.

Planned writing becomes reading model essays.

Planned mixed Mathematics becomes blocked practice.

Planned transfer becomes repeating the same familiar question.

Planned checking becomes looking at the answer key.

Activity substitution often feels easier and still looks like study.

Fidelity monitoring detects the substitution.

Why Activity Substitution Happens

The intended operation may be uncomfortable.

Retrieval creates failure.

Mixed practice creates uncertainty.

Fresh writing creates blank-page difficulty.

Transfer creates unfamiliarity.

Learners naturally move toward lower-friction activities.

Fidelity is therefore partly about protecting useful difficulty without making the system punitive.

Fidelity and Desirable Difficulty

Difficulty itself is not the active ingredient.

If a learner struggles because the instructions are confusing, fidelity to a retrieval plan does not require preserving confusion.

Reduce irrelevant difficulty.

Preserve the target operation.

This distinction prevents “high fidelity” from becoming “never simplify anything.”

Fidelity and Expertise Reversal

Supports useful to beginners can become unnecessary later.

High fidelity does not require retaining obsolete scaffolds.

The active ingredient may be independent reasoning, not the scaffold itself.

Once the learner can carry the operation, remove support.

Otherwise fidelity to the original novice design can reduce fidelity to the ultimate learning goal.

Fidelity and Evidence Freshness

A plan can remain faithful to a learner model that is no longer current.

Mira improves, but the tutor keeps treating her as fragile.

Jonas repairs inference, but the plan continues to spend most time there.

Nadia’s new bottleneck becomes timing, but tuition stays focused on evidence explanation.

High fidelity to stale evidence is not good training.

Fidelity must operate alongside review.

Fidelity and Replanning Triggers

The site now includes How Tuition Works | The Replanning Trigger.

That article asks when a strategy should be abandoned and another chosen.

Fidelity helps interpret whether replanning is warranted.

If a strategy appears weak but fidelity was low, restore implementation before abandoning it.

If fidelity was high and valid outcome evidence remains poor, replanning becomes more justified.

The Fidelity Decision Tree

When outcomes are disappointing:

• Step 1: Was the target capability measured validly?
• Step 2: Were the active ingredients delivered?
• Step 3: Was dosage adequate?
• Step 4: Was delivery quality adequate?
• Step 5: Did the learner actually engage in the intended operation?
• Step 6: Did adaptations preserve the mechanism?
• Step 7: If fidelity was adequate, review the plan itself.

This order prevents method changes from being triggered by implementation failure.

The Fidelity Decision Tree When Outcomes Are Strong

When outcomes improve:

• Was fidelity high?
• Did the improvement occur on fresh tasks?
• Could school teaching or another intervention explain the gain?
• Did the learner become less support-dependent?
• Did the active ingredient plausibly contribute?
• Which parts should now be faded, sustained or transferred?

Strong outcomes do not remove the need for fidelity analysis.

They help identify what deserves preservation.

Fidelity and Counterfactual Thinking

The Counterfactual Check asks whether tuition actually caused improvement.

Fidelity provides necessary context.

If the intervention was not delivered, it is a weak candidate cause.

If it was delivered consistently and the targeted process changed first, the causal story becomes more plausible—though still not proven by informal observation alone.

A Fidelity Audit for Parents

• What was the plan actually trying to train?
• Which parts were essential?
• Did those parts happen this week?
• How many meaningful attempts occurred?
• Was the work independent or heavily prompted?
• Did homework emergencies replace the planned training?
• Were fresh retests completed?
• What was adapted, and why?
• Are we blaming the method for a plan we did not really run?
• Is the plan too demanding to implement faithfully?

A Fidelity Audit for Tutors

• Can I state the active ingredients in one paragraph?
• Can I distinguish tight from loose elements?
• What observable trace shows each active ingredient occurred?
• Is dosage counted in relevant learning opportunities rather than time alone?
• What support entered?
• Did any support remove the target decision?
• Which adaptations were deliberate?
• Which changes were accidental drift?
• Did school demands displace the mechanism?
• Did the learner actually perform the intended cognitive operation?
• Was a fresh or delayed verification included?
• Can I interpret the outcome without knowing fidelity?

A Fidelity Audit for Learners

• What skill am I trying to train?
• What do I actually have to do for that training to count?
• Am I avoiding the difficult part by switching to an easier activity?
• Did I attempt before looking at help?
• Did I use feedback to change another attempt?
• Did I test myself on something fresh?
• Did I keep practising after the skill was already stable?
• What can I now do without the tutor?

Fidelity Failure Mode: Completion Theatre

Every planned box is ticked.

Worksheet finished.

Video watched.

Notes copied.

Homework submitted.

But the active operation never occurred.

This is completion theatre.

The visible system is compliant.

The learning system is weak.

Repair:

track the target operation, not the paperwork surrounding it.

Fidelity Failure Mode: Dosage by Minutes

“One hour of Mathematics” is treated as a dose.

But thirty minutes may be spent waiting, copying or checking answers.

Repair:

count meaningful attempts, decisions or repair loops when possible.

Fidelity Failure Mode: Hidden Prompting

The learner succeeds repeatedly.

The tutor quietly signals the route.

The parent reminds at home.

The worksheet labels the method.

Performance looks strong.

Independence is untested.

Repair:

record support and use fresh unprompted verification.

Fidelity Failure Mode: Over-Scripting

Every sentence and timing interval is prescribed.

The tutor becomes afraid to respond to the learner.

Repair:

specify functions and active ingredients rather than unnecessary surface details.

Fidelity Failure Mode: Adaptation Without Documentation

The plan changes every week.

No one remembers why.

After a month, outcomes cannot be connected to a coherent system.

Repair:

record material adaptations in one sentence: what changed, why and which active ingredient was protected.

Fidelity Failure Mode: Protecting the Plan From the Evidence

The learner is clearly ready to progress.

The tutor insists on finishing the scheduled unit because “that is the programme.”

Repair:

remember that the plan serves the learner, not the reverse.

Fidelity Failure Mode: False Blame

A family cannot complete an unrealistic plan.

The system interprets this as low motivation.

Repair:

check feasibility before attributing implementation failure to character.

Fidelity Failure Mode: Fidelity to Materials Instead of Mechanism

The tutor believes fidelity means finishing Book 2.

But the learner’s active ingredient was mixed method selection.

The book is only one delivery vehicle.

Repair:

protect the mechanism even if materials change.

Fidelity Failure Mode: Endless Monitoring

The tutor records so much implementation data that teaching quality falls.

Repair:

use the minimum fidelity indicators required for the decision.

Fidelity Failure Mode: Ignoring Learner Adaptation

Learners modify tasks too.

Mira looks at the answer before attempting.

Jonas uses a memorised template for every paragraph.

Nadia skips explanation and jumps to keywords.

These are learner-level implementation adaptations.

Repair:

teach the learner why the active ingredient matters.

Fidelity Failure Mode: Assuming One High-Fidelity Week Is Enough

The plan runs beautifully during a holiday.

School restarts and the system collapses.

Repair:

evaluate fidelity under the normal week, not only ideal conditions.

The Real-Week Fidelity Test

Before trusting a plan, run it through an ordinary week.

Monday school ends late.

Tuesday CCA.

Wednesday homework surge.

Thursday tuition.

Friday fatigue.

Weekend family commitments.

Where do active ingredients survive?

Where do they disappear?

This test connects directly to Training Feasibility.

The Low-Energy Fidelity Test

A sustainable system needs a reduced mode.

When energy is low, what is the smallest version that preserves the mechanism?

For retrieval:

five closed-book questions instead of thirty minutes.

For writing:

one paragraph reconstructed from a plan instead of a full essay.

For Mathematics:

three mixed classification decisions instead of a full worksheet.

For Science:

one evidence–mechanism–conclusion chain instead of a full chapter review.

This is fidelity-preserving compression.

The High-Pressure Fidelity Test

Near examinations, plans tend to collapse into paper volume.

Ask:

• Are feedback loops still occurring?
• Are errors still diagnosed?
• Are fresh retests still present?
• Is mixed recognition still practised?
• Is recovery still protected?

Pressure should change emphasis, not erase mechanism.

The Tutor Absence Fidelity Test

What happens if the tutor is unavailable for a week?

If the system disappears completely, fidelity depends on one person.

Long-term training should gradually move active ingredients into learner routines.

The learner should know how to retrieve, check, reattempt and review without needing the tutor to initiate every loop.

The Parent Absence Fidelity Test

What happens if the parent cannot supervise homework?

If the child immediately stops the training routine, support has not yet migrated.

This does not mean the learner is irresponsible.

It identifies a handoff still incomplete.

The School Disruption Fidelity Test

A school project, camp or examination changes the week.

Can the training system reduce dosage while preserving a minimal active ingredient?

If yes, the design is robust.

If no, the system may be too brittle for real education.

Fidelity During Re-entry

After illness or holidays, the old plan may no longer fit.

Training Re-entry should sample what survived before full fidelity to the old sequence resumes.

Otherwise we may faithfully deliver a programme for a learner state that no longer exists.

Fidelity During Transitions

Primary to Secondary changes workload and independence.

Secondary 2 to Secondary 3 changes subject complexity.

Secondary 4 to post-secondary changes control and self-management.

The active ingredients may remain, but delivery must adapt.

Training Transitions therefore requires re-specifying what fidelity looks like in the new stage.

Fidelity in English Training

English is especially vulnerable to low-fidelity substitution because the subject has many plausible activities.

Reading can replace writing.

Vocabulary can replace inference.

Model essays can replace generation.

Grammar can replace idea development.

Comprehension practice can replace evidence calibration.

All are legitimate English activities.

Only some match the current training job.

Fidelity in Mathematics Training

Mathematics is vulnerable to page-count substitution.

A learner completes fifty routine questions.

The target was method selection under mixed conditions.

Execution improved.

Selection did not.

The volume is real.

Fidelity to the target is low.

Fidelity in Science Training

Science is vulnerable to content-recall substitution.

A learner memorises notes and definitions.

The target was evidence integration in unfamiliar experiments.

Recall improves.

Application remains fragile.

Again, study happened.

The intended training did not.

Fidelity in Examination Training

Examination preparation can become paper accumulation.

Past paper completed.

Score recorded.

Next paper.

But the intended system may have been:

simulate → analyse errors → identify source → repair → retest → return to simulation

If repair and retest disappear, the exam-training mechanism changes.

Fidelity in Home Study

Home study has fewer external controls.

This makes fidelity a learner skill.

The child needs to know:

• what the session is for;
• what operation must occur;
• what counts as completion;
• what help is allowed;
• what fresh check ends the session.

A home plan that only states “Science 45 min” gives weak fidelity guidance.

From Time Blocks to Training Contracts

A more faithful plan writes the job:

Science — 20 minutes: retrieve five concepts without notes, check, repair two errors, answer one unfamiliar evidence question without model answer.

This is a training contract.

It specifies the operation rather than merely the duration.

The Fidelity Contract

For any active training target, write:

• Job: what capability changes?
• Active ingredient: what must the learner do?
• Minimum dose: how many meaningful opportunities?
• Support rule: what help is allowed?
• Verification: what fresh evidence closes the loop?
• Adaptation rule: what may change?

This six-line contract can align tutor, parent and learner.

Worked Fidelity Contract: Mathematics

• Job: recognise which quadratic method fits.
• Active ingredient: choose before calculating and justify structural cue.
• Minimum dose: six mixed decisions across at least two sessions.
• Support rule: attention cue allowed; method name not allowed before choice.
• Verification: fresh mixed set after delay.
• Adaptation: coefficients, contexts and order may change.

Worked Fidelity Contract: English

• Job: keep inference strength within evidence.
• Active ingredient: select evidence, propose claim, compare claim strength.
• Minimum dose: several inferences across more than one passage type.
• Support rule: evidence-location cue allowed; adjective not supplied initially.
• Verification: fresh passage without cue.
• Adaptation: genre and topic may change.

Worked Fidelity Contract: Science

• Job: connect evidence to mechanism.
• Active ingredient: state evidence, pattern, mechanism and bounded conclusion.
• Minimum dose: several unfamiliar contexts.
• Support rule: apparatus terminology may be clarified if not the target.
• Verification: new experimental setup.
• Adaptation: topic surface may change.

Fidelity and the Learning Dispatcher

The existing How Tuition Works | The Learning Dispatcher asks where the next learning job should be routed.

Fidelity asks whether the routed job is actually performed once it arrives.

A perfect dispatcher cannot help if every routed intervention mutates into generic worksheet completion.

Fidelity and the Flight Recorder

How Tuition Works | The Flight Recorder owns progress history.

A fidelity log complements it by recording implementation history.

Outcome history says what happened to performance.

Fidelity history says what training conditions preceded it.

When both are available, review becomes much more informative.

Fidelity and the Evidence Threshold

The Evidence Threshold asks when there is enough evidence to change the plan.

Fidelity should be checked before using weak outcomes as change evidence.

If fidelity was red, the outcome may be evidence to change implementation rather than the underlying method.

Fidelity and Proxy Failure

The site’s Proxy Failure article asks when the metric improves while capability does not.

Fidelity asks the mirror question: did the active training mechanism occur while the visible proxy activity was completed?

A learner can hit worksheet targets while mechanism fidelity falls.

Fidelity and Distribution Shift

The site’s Distribution Shift article asks when practice and performance come from different worlds.

Fidelity should not be confused with repeating one distribution forever.

A high-fidelity training system may deliberately broaden task conditions so the active capability survives distribution shift.

Fidelity and Graceful Degradation

Graceful Degradation offers a useful implementation principle.

When the week becomes difficult, preserve the core mechanism at reduced scale.

That is often better than attempting full dosage for three days and then collapsing entirely.

Fidelity and the Cold Start Test

The Cold Start Test asks whether the learner can begin without the tutor.

If the plan’s active ingredient is independent initiation, tutor-initiated success should not be counted as faithful delivery of that target.

Fidelity and Prompt Migration

Prompt Migration tracks the movement from tutor cues to learner self-cues.

Fidelity changes across that progression.

Early fidelity may include an external cue.

Later fidelity may require the learner to generate the cue internally.

The active ingredient evolves as responsibility migrates.

Fidelity and the Help Gradient

The Help Gradient protects the smallest-help principle.

If a training plan depends on independent problem solving, full explanations at the first hesitation reduce fidelity.

If the learner lacks the model entirely, refusing help can also reduce fidelity to the intended learning progression.

Help should match the current state.

Fidelity and Error Detectability

Error Detectability asks whether learners can notice mistakes before external feedback.

If the intervention includes self-checking, fidelity should record whether the learner actually performed the check rather than whether the final answer happened to be correct.

Fidelity in AI-Assisted Study

AI changes fidelity risks.

A learner plans to practise generation but asks an AI system for the answer first.

A learner plans to self-explain but requests a polished explanation and then reads it.

A learner plans to write independently but uses generated paragraphs as scaffolds.

These tools may still support learning.

But the cognitive operation changes.

AI use should therefore be sequenced according to the active ingredient.

AI Fidelity Rule: Attempt Before Assistance When Generation Is the Job

If the target is retrieval, generation, planning, method selection or writing, protect an unaided attempt first.

Then use AI for:

• feedback;
• comparison;
• alternative examples;
• explanation;
• error checking.

This preserves learner generation while still using AI productively.

AI Fidelity Rule: Verify the Feedback Loop

Receiving AI feedback is not the same as using it.

High-fidelity repair requires:

• identify the change;
• understand why;
• generate a revised response;
• test on a fresh case.

Copying the AI revision directly removes the learner repair step.

AI Fidelity Rule: Keep the Human Decision Visible

If the target is decision-making, the AI should not silently make the decision.

Mira can ask for two candidate methods and compare them.

But if the system simply says “use factorisation,” method-selection training disappears.

Fidelity and Data Dashboards

Dashboards can show completion and scores.

They may not show whether the active ingredient occurred.

A 100% completion rate can coexist with low fidelity.

Implementation monitoring should therefore include at least one mechanism-level indicator where it matters.

Fidelity and Privacy

Implementation monitoring should remain proportionate.

Families do not need invasive tracking to know whether a study plan ran.

Simple self-report, tutor notes and selected artefacts are often enough.

Collect only what improves the educational decision.

Fidelity and Motivation

Low fidelity is often labelled low motivation.

That can be unfair.

A plan may be:

• too long;
• too repetitive;
• poorly timed;
• misaligned with school demand;
• unclear;
• socially unacceptable;
• dependent on adult support unavailable at home.

Before diagnosing motivation, inspect implementation design.

Fidelity and Responsibility

At the same time, design quality does not remove learner responsibility.

Once a feasible, acceptable and clear plan exists, the learner increasingly needs to carry it.

The aim is not to explain away every missed session.

It is to make responsibility fair: the system asks the learner to do something realistic, meaningful and understood.

Fidelity and Trust

Families trust training plans more when they can see the logic.

“Do thirty questions” is brittle.

“Make six method decisions without labels, then retest two later” explains the learning job.

Understanding creates better implementation because people can adapt the form without destroying the function.

Fidelity and Tutor Professional Judgement

Fidelity does not replace expertise with checklists.

It sharpens professional judgement.

The tutor still decides:

• which ingredient is essential;
• which adaptation is safe;
• when dosage is sufficient;
• when the learner is ready to progress;
• when fidelity should yield to a better plan.

The checklist is there to make the decision traceable.

Fidelity and Parent–Tutor Communication

A good update does not say:

We covered Chapter 6.

It says:

We trained method recognition across mixed cases. Mira can now classify four of six independently; two still require a structural cue. Next session we retest fresh before adding speed.

This communicates both outcome and fidelity.

Fidelity and Learner–Tutor Communication

The learner should hear:

We are not doing this question because the worksheet says so. We are using it because you need to practise noticing which structure changes the method.

Purpose protects fidelity.

Fidelity and School–Tutor Alignment

Tutors should not reproduce school blindly.

They should understand school demand and preserve the training job.

If school is doing blocked practice, tuition may provide mixed discrimination.

If school is introducing a new representation, tuition may pre-train the notation.

Alignment means complementary implementation, not duplication.

Fidelity and Home–Tutor Alignment

Parents can accidentally remove productive difficulty by helping too soon.

Tutors can protect fidelity by giving one home rule:

Give Mira thirty seconds. If she is still stuck, ask what the question is asking. Do not name the method.

One shared cue policy keeps the home environment aligned without turning parents into tutors.

Fidelity and the Family Week

eduKatePunggol owns the family-life layer because educational systems run inside real households.

Fidelity therefore belongs in calendars.

Which sessions are protected?

Which can move?

Which collapse under CCA or exam weeks?

Which require parental presence?

Which can be learner-run?

The answers determine whether the designed programme will ever exist outside the document.

Fidelity and Sleep

A plan that preserves active ingredients by sacrificing sleep is not a successful implementation.

Sleep is part of the environment that makes learning possible.

If dosage requires repeated late nights, feasibility is weak and fidelity pressure is misallocated.

Fidelity and Recovery

Recovery can be an active ingredient.

If the training architecture includes rest between demanding blocks, removing rest to “increase fidelity” actually decreases fidelity.

The plan is a system, not a list of work tasks.

Fidelity and Spacing

Spacing fidelity is not merely “practice happened on Tuesday and Friday.”

The content returned after a meaningful interval.

If Friday simply repeats the newest content, cumulative spacing is weak.

Record what returned, not only that a session occurred.

Fidelity and Interleaving

Interleaving fidelity requires genuine method or category choice.

Alternating pre-labelled sections—fractions, algebra, geometry—may create topic alternation without requiring recognition.

If selection is the active ingredient, conceal the labels.

Fidelity and Retrieval

Retrieval fidelity requires an attempt to recover information before external support supplies it.

If the learner opens notes immediately, the activity becomes review.

Review can be useful.

It is not retrieval practice in that moment.

Fidelity and Worked Examples

Worked examples are not faithful if the learner merely reads them repeatedly while the intended progression requires prediction, explanation, completion and independent transfer.

The canonical Worked Example State on eduKateSengkang owns that progression.

Punggol fidelity simply asks whether the intended stage of that progression actually occurred.

Fidelity and Self-Explanation

Self-explanation fidelity requires learner-generated reasoning.

If the tutor supplies the explanation and asks the learner to agree, the surface looks similar while the mental operation changes.

Fidelity and Generation

Generation fidelity requires the learner to produce before being shown.

Examples can be imperfect.

Predictions can be wrong.

The attempt is part of the mechanism.

Fidelity and Contrast

Contrast fidelity requires comparison.

Showing two examples sequentially without asking what changes may not produce the same discrimination work as placing them side by side and making the critical difference explicit.

Fidelity and Nonexamples

Nonexample fidelity requires the learner to see why the near-miss fails.

Simply showing a wrong answer without boundary analysis may create confusion rather than concept sharpening.

Fidelity and Invariants

Invariant training requires surface changes while the underlying relationship remains.

If every example keeps the same surface too, the learner may never need to separate invariant from incidental detail.

Fidelity and Perturbation

Perturbation fidelity requires changing one meaningful feature at a time.

If five features change together, the learner cannot tell which feature changed the decision.

Fidelity and Ambiguity

Ambiguity training should preserve multiple plausible interpretations long enough for discriminating evidence to matter.

If the tutor signals the “intended answer” too early, ambiguity disappears and so does the decision skill.

Fidelity and Distractors

Distractor training requires plausible alternatives linked to real misconceptions or near-neighbour rules.

Random nonsense options do not preserve the discrimination mechanism.

Fidelity and Case Families

Case-family fidelity requires relationships among examples.

A random pile of worksheets is not a case family simply because the questions share a topic.

Fidelity and Reconstruction

Reconstruction fidelity requires external structure to disappear.

If the notes remain open, the learner is not reconstructing.

They may still be studying.

Again, the distinction is mechanism, not morality.

Fidelity and Pre-training

Pre-training fidelity requires teaching the components without pre-solving the main relationship.

If the full answer is effectively given in advance, the later task loses the decision it was supposed to train.

Fidelity and Branching

Branching fidelity requires the learner response to influence the next task.

If every learner follows the same fixed sequence regardless of evidence, the system is not truly branching.

Fidelity and Baselines

Baseline fidelity requires sampling before the intervention substantially changes the learner.

If teaching occurs during the baseline and the final supported answer is recorded as the starting state, interpretation is distorted.

Fidelity and Retests

Retest fidelity requires fresh enough items, appropriate independence and comparable difficulty.

Repeating the exact correction is not the same measurement job.

Fidelity and Parallel Forms

Parallel-form fidelity requires the blueprint to remain comparable while surface details change.

If Form B is much easier, apparent improvement cannot be cleanly attributed to learning.

Fidelity and Anchor Tasks

Anchor-task fidelity requires stable reference conditions.

If the anchor changes difficulty every time, it stops functioning as an anchor.

Fidelity and Triangulation

Triangulation fidelity requires genuinely different evidence windows.

Three worksheets from the same template are not three independent windows.

Fidelity and Observability

Observability fidelity requires preserving the route we need to inspect.

If learners erase all working, the process becomes less observable.

Fidelity and Measurement Resolution

Measurement-resolution fidelity requires using the granularity the decision needs.

If every problem is coded merely correct/wrong, a fine-grained repair plan cannot be implemented reliably.

Fidelity and Error Propagation

Error-propagation analysis requires preserving the sequence so the first invalid state can be traced.

If only final answers are available, the diagnostic mechanism cannot run.

Fidelity and Exit Criteria

Exit criteria must actually be tested.

If a skill leaves active training because the calendar moved on, the exit mechanism was not implemented.

The Fidelity Review Meeting

A monthly five-minute review can ask:

• What were our two most important active training jobs?
• Were their active ingredients delivered?
• What repeatedly displaced them?
• Which adaptations helped?
• Which adaptations removed the mechanism?
• What should be redesigned?

This is enough to detect drift before it becomes the new normal.

Fidelity Evidence Can Be Qualitative

Not everything needs a percentage.

A note such as:

Fresh retest repeatedly skipped during exam weeks because tuition time is consumed by urgent homework

is actionable evidence.

Do not force false numerical precision onto implementation.

When Numbers Help

Simple counts can help when they represent the mechanism:

• number of independent starts;
• number of fresh retests;
• number of method decisions before cueing;
• percentage of planned retrieval opportunities actually attempted closed-book;
• number of sessions preserving the target active ingredient.

Use numbers to clarify, not decorate.

Fidelity Trend vs Single Week

One low-fidelity week may be harmless.

Repeated low fidelity is a system problem.

Look for trends.

Does the training plan degrade whenever school workload rises?

Does home practice become supported whenever the parent is present?

Do fresh retests disappear near examinations?

Patterns reveal design weaknesses.

The Fidelity Threshold for Interpretation

There is no universal percentage at which a training intervention becomes interpretable.

Different mechanisms tolerate different deviations.

Instead of imposing a fake universal threshold, ask:

• Were the active ingredients present often enough to plausibly create the expected change?
• Were missing elements peripheral or central?
• Did the learner actually experience the relevant operation?
• Were outcome measures aligned with what was delivered?

This keeps fidelity judgement mechanism-based.

Fidelity and Scientific Humility

In research, causal claims require designs stronger than ordinary tuition monitoring.

In practice, we should still think scientifically.

If fidelity is unknown, say so.

If an adaptation changed the mechanism, say so.

If several influences changed at once, avoid pretending one caused the outcome.

Honest uncertainty improves training decisions.

Research Foundations

The strongest current practice anchor is AERO’s 15 September 2026 guide Staying on track: Monitoring implementation outcomes, which explicitly organises monitoring around feasibility and acceptability, fidelity and reach, and sustainability. The guide positions implementation outcomes as part of a deliberate, structured approach to embedding evidence-based teaching practice rather than treating implementation as a one-time launch.

EEF’s A School’s Guide to Implementation provides a complementary framework: behaviours, contextual factors and a structured Explore–Prepare–Deliver–Sustain process. Its implementation-planning resources emphasise specifying active ingredients and knowing where to be tight and where to be loose. EEF’s contextual guidance also highlights time, roles, data systems and enabling people, which map directly onto real training fidelity.

EEF’s September 2026 Evaluate and sustain your strategy guidance reinforces that implementation and final outcomes should be evaluated together and that strategies may need to be sustained, adapted, scaled or stopped.

These sources operate mainly at school and programme level. This article translates their implementation logic cautiously into individual and small-group educational training. It does not claim that a family study plan is equivalent to a school-wide intervention. The transfer is conceptual: outcomes are difficult to interpret without knowing what was actually delivered, and implementation quality depends on more than whether an activity appeared on the timetable.

A Fidelity Field Manual: Fifty Questions

Use these selectively. Their purpose is to help adults and learners interrogate the enacted plan without turning learning into paperwork.

• What capability is this training supposed to change?
• What is the essential mental operation?
• What would a superficial substitute look like?
• What evidence shows the operation actually occurred?
• What was the planned dosage?
• What meaningful dosage was actually delivered?
• What was skipped?
• What replaced it?
• Was the replacement intentional?
• Did the replacement preserve the active ingredient?
• What support was allowed?
• What support actually entered?
• Did any support remove the target decision?
• Was the task fresh?
• Was the learner shown the answer too early?
• Did the learner attempt before receiving feedback?
• Was feedback followed by changed performance?
• Was a fresh reattempt included?
• Was a delayed return included?
• Did the learner perform under mixed conditions?
• Was the method named by the worksheet?
• Did school homework displace the target operation?
• Did parental help change independence?
• Did peer help reveal the answer route?
• Did AI assistance remove generation?
• Was the learner engaged in the intended cognitive job?
• Was the session long enough to deliver the mechanism?
• Was the session so long that fatigue changed quality?
• Did the tutor rush the final verification?
• Did the plan adapt because of evidence or convenience?
• Was the adaptation recorded?
• Which parts were tight?
• Which parts were loose?
• Could another tutor identify those parts?
• Could the learner explain the purpose?
• Could the parent support without accidentally changing the mechanism?
• Did the plan remain feasible this week?
• Did the learner accept the plan?
• Did the plan survive a low-energy day?
• Did the plan survive an exam week?
• Did the plan survive tutor absence?
• Did the plan survive parent absence?
• Was fidelity equally strong across home and tuition?
• Did the active ingredient reach authentic school performance?
• What fidelity problem recurred?
• What redesign would solve that problem?
• Are we blaming the method for an implementation failure?
• Are we protecting a weak plan because fidelity was high?
• What should remain next cycle?
• What should be deliberately changed?

Final Principle

A training plan does not exist because it is written.

It exists only when a learner experiences its active ingredients.

That sounds obvious, but education is full of proxy completion: hours logged, worksheets finished, notes copied, programmes purchased, tuition attended.

Fidelity asks the harder question beneath all of them.

Did the learner actually receive the training system whose outcomes we are now trying to judge?

If the answer is yes, outcome evidence can speak more clearly.

If the answer is no, repair implementation before declaring the method a success or failure.