A four-week English workshop is designed to give pupils one model example, two guided attempts and a new unaided inference question each week. By the final session, the teacher has taught four lessons, but the guided practice was often replaced with reading model answers aloud and two classes never completed the independent tasks. Can the workshop’s results tell us whether the intended programme works?
Another group reports that a new revision strategy is successful because pupils attended regularly. But being present does not mean the core technique was used, that feedback was accurate or that the learners practised with enough independence. A service can be delivered according to a timetable without delivering the instructional process that was supposed to change learning.
Implementation fidelity is the degree to which an intervention is actually delivered as intended. It may include the content taught, how often and for how long it was taught, the quality of delivery, differentiation from comparison activities and participant responses. These dimensions help explain what was really tested rather than what appeared in a proposal.
Fidelity is not simply rigid obedience to a script. Some adaptations preserve the intended mechanism while improving accessibility; others remove the mechanism and fundamentally change the intervention. The skill lies in identifying essential components, documenting modifications, observing delivery without intrusive surveillance and interpreting outcomes in the context of what occurred.
This article is a focused companion to Intervention Evaluation Skills, which owns the whole evaluation decision, Causal Pathway Mapping Skills, which owns proposed mechanisms, Study Design Skills, which owns the prospective research plan, and Outcome Measure Selection Skills. Here the unique job is checking and describing the *intervention actually delivered* against the designed version.
Before the Top 10: A Programme Title Is Not Its Delivered Content
Start with an implementation blueprint: what components must occur for the proposed learning process to be plausible? For an inference workshop, modelling evidence selection, coached attempts, meaningful feedback and unaided transfer may be core. A worksheet’s colour or the length of a teacher’s opening story might be less consequential.
Different research traditions distinguish adherence to content, exposure or dose, delivery quality, participant responsiveness and programme differentiation. The specific framework and measurement should fit the intervention. A short school study can use simple checklists and documented observations, while a formal multi-site trial may need more extensive data collection.
Fidelity is measured separately from effectiveness. An intervention delivered faithfully may have no effect on the chosen outcome; poorly delivered sessions may produce a weak result even if the idea could work when implemented correctly. The two questions—did the plan happen, and did the plan improve learning—must not be merged.
The ten headings below are a Wintour House / eduKate original teaching synthesis, not official MOE/SEAB inspection rules. Research from Carroll and colleagues and the EEF supports systematic attention to fidelity and context, but no single percentage threshold establishes successful implementation in every education programme.
INTENDED COMPONENTS → ACTUAL DELIVERY → LEARNER EXPERIENCE → MEASURED OUTCOMES → HONEST INTERPRETATION
1. Identify the Core Components Before Delivery Begins
A programme often contains activities of unequal importance. Fidelity should focus on the elements that create the proposed mechanism rather than every decorative detail. Otherwise teachers may appear faithful because they use the same slides while omitting the practice that was meant to produce learning.
For a reading inference clinic, core components may be identifying evidence, explaining the inference, feedback on the reasoning link and independent practice on an unfamiliar passage. Choosing blue rather than green highlighters is not necessarily core. Calling both equally essential would make the fidelity measure uninformative.
List each component and justify how it connects to the target learning outcome. Describe what would count as sufficient delivery and what change would break the intended mechanism. Review the blueprint with appropriate teachers before observing any sessions.
Worth learning because intervention fidelity becomes meaningful when essential instructional ingredients are separated from incidental packaging.
2. Track the Planned Dose and the Dose Actually Delivered
An intervention may prescribe a number of sessions, minutes or opportunities to practise. The amount delivered is one useful implementation measure, though minutes alone cannot describe the quality of teaching. A timetable cannot be substituted for a record of real learning activities.
Suppose the plan specifies four sixty-minute sessions, totalling 240 scheduled minutes. One is cancelled and one finishes thirty minutes early. The maximum delivered time is 180 minutes, or 75% of the planned time, assuming the other sessions run fully. That calculation says nothing by itself about how much purposeful inference practice pupils received.
Record session dates, durations and major activities. If a teacher spends more time helping a struggling group, note the instructional reason rather than simply penalising the deviation. Do not call a time percentage a validated fidelity score for the programme without defining its role.
Worth learning because intended dose and actual opportunity are different facts that affect interpretation.
3. Distinguish Delivery From Learner Exposure and Participation
An educator can deliver an activity that some pupils do not actually experience. Absence, partial attendance, accessibility barriers and lack of engagement can change the received intervention. But exposure does not automatically mean meaningful learning or perfect adherence to instructions.
Imagine a worksheet is taught to twenty pupils, sixteen attend the full lesson and twelve independently attempt the follow-up question. All three counts are relevant. The teacher delivered the lesson, but the learning experience and independent practice differed among pupils. An attendance rate alone cannot show whether the mechanism reached every learner.
Maintain separate numbers for scheduled, delivered, attended and independently attempted activities, using consistent denominators. Avoid collecting unnecessary personal reasons for nonattendance. Report reach accurately without pretending the absent pupils failed the learning task.
Worth learning because what was offered, what was delivered and what students experienced are distinct implementation dimensions.
4. Observe Quality of Delivery, Not Just Presence of Materials
A checklist saying that feedback occurred does not establish whether it was accurate, timely or actionable. A teacher can read a model answer word for word while pupils remain passive, or teach the same principle through questions that reveal and repair misconceptions.
In Mathematics, two tutors may both distribute practice sheets. One asks why a step is valid, identifies an incorrect representation and fades hints before independent work. The other provides correct answers immediately. The materials and duration match while the quality of instructional interaction differs markedly.
Use a small set of observable quality indicators tied to the programme’s intended mechanism. Calibrate observers with examples, make feedback constructive and avoid reducing teacher quality to one superficial tally. A judgement should point to instructional evidence, not personal style preferences.
Worth learning because fidelity includes the quality with which a meaningful activity happens, not simply whether a box was ticked.
5. Record Learner Responsiveness Without Mistaking Enjoyment for Impact
Participant responsiveness may include attending, attempting tasks, asking questions or using feedback. Positive attitudes can matter to feasibility, but they are not identical to learning outcomes. A programme could be popular and still fail to improve the intended skill.
A short English clinic receives favourable responses from most pupils, while fresh passages show continued difficulty explaining evidence. The satisfaction finding is real among respondents; the pathway from participation to independent inference appears weak. Conversely, demanding practice may be less enjoyable while producing meaningful learning under appropriate conditions.
Choose a few respectful observations of how pupils engage with core tasks. Separate learner experience from performance. Where a survey is used, preserve response rates and avoid implying that nonrespondents agree with the people who answered.
Worth learning because participation information helps explain implementation without being promoted into proof of attainment.
6. Keep the Intervention Distinct From Ordinary Teaching and Comparators
Fidelity can be difficult to interpret when the supposed intervention group and comparison group receive much the same activity. Programme differentiation asks which meaningful components actually distinguish the intervention from existing practice or the alternative being tested.
Suppose a trial compares a new retrieval quiz routine with ordinary revision, but teachers in the comparison group independently begin using similar retrieval quizzes. The contrast between groups becomes smaller than originally planned. A weak effect estimate might not assess the intended difference, even if both sets of pupils benefited from practice.
Record the core activities provided in each condition and any spillovers across groups. Do not assume that an assignment label guarantees actual differentiation. In a simple classroom improvement project, state what changed relative to the usual lesson.
Worth learning because a comparison cannot assess a distinctive programme ingredient if both groups receive it in essentially the same way.
7. Distinguish Planned Adaptation From Omission of a Core Mechanism
Teachers routinely adjust examples, timing and materials for learner needs. Some changes make a programme more accessible while preserving its essential function. Others remove the component the intervention relies on. Both should be documented, but not treated as equivalent failures.
A teacher translates an instruction into simpler language while keeping the requirement that pupils explain their own inference; the change may support access without abandoning the skill. If the teacher instead writes every inference for pupils and removes all unaided attempts, that adaptation risks eliminating the independence mechanism.
For each change, note what was altered, why and whether the intended causal component remained. Seek the programme designer’s guidance when formal treatment fidelity matters. Avoid rigid enforcement of cosmetic features that do not serve the learning goal.
Worth learning because thoughtful implementation protects the active ingredient while allowing necessary contextual adaptation.
8. Check Context, Resources and Staff Preparation
Delivery can vary because classrooms have different timetables, pupil needs, staff training, spaces and competing school priorities. These contextual factors can explain implementation differences without establishing that individual teachers lack effort.
An intervention assumes pupils can access one device each, but a class shares a few functioning tablets. The planned individual feedback cycle cannot occur at the same rate. Alternatively, a substitute teacher may receive the slides but not the training needed to recognise the target misconception.
Record context relevant to core delivery: staffing, materials, timetable interruptions, training, technical access and permission constraints. Distinguish unavoidable constraints from preventable process failures. Use the information to plan feasible adjustments rather than to assign blame.
Worth learning because implementation is a real-world process shaped by conditions, not simply a document that teachers are expected to obey.
9. Compare Fidelity Records With Actual Outcomes Without Assuming Causation
Knowing how an intervention was delivered helps interpret an outcome pattern, but high fidelity and high achievement occurring together does not automatically prove one caused the other. Better classrooms may have both stronger implementation and other advantages.
Imagine two fictional groups: one completes nearly all planned inference practice and reports higher fresh-passage scores; another completes little practice and performs less well. The pattern is compatible with the intended mechanism, but baseline attainment, attendance and other teaching might also differ. A credible causal claim still needs suitable design.
Link implementation dimensions to outcome measures, then ask what alternative explanations remain. Interpret fidelity as process evidence that clarifies the treatment received, not as a replacement for counterfactual comparison.
Worth learning because fidelity evidence explains what happened during delivery without proving an intervention’s causal effect by itself.
10. Report the Delivered Programme and Use the Findings to Improve It
A useful implementation report states what was planned, what actually happened, the major adaptations, who was reached, what was measured and what remains uncertain. The next decision may be to repair scheduling, train teachers, revise materials or test a different approach.
For a four-week clinic that delivered three sessions and no meaningful unaided practice, a clear conclusion is that the intended mechanism was only partially implemented. A decision to retest after restoring independent practice may be sensible. It would be misleading to declare that the full intended intervention was ineffective from the altered delivery alone.
Keep dated records and distinguish fidelity findings from outcome findings. Choose one important implementation problem for a next trial and record the predicted effect of correcting it. Test with changed examples and independent student performance rather than merely a smoother timetable.
Worth learning because fidelity becomes valuable when it makes a programme more transparent, more feasible and more accurately evaluated.
The Top 10 Implementation Fidelity Skills as One System
The practical sequence is CORE COMPONENTS → DOSE → PARTICIPATION → DELIVERY QUALITY → RESPONSIVENESS → DIFFERENTIATION → ADAPTATION → CONTEXT → OUTCOME INTERPRETATION → TRANSPARENT REVISION. Each link matters for a different question; not every small classroom routine requires a ten-page process report.
The quality test is whether another educator could understand what the intended intervention was, what pupils actually received and which implementation limits affect the meaning of the results.
Fidelity Is Not the Same as Student Adherence or Effectiveness
A programme may be delivered faithfully by educators while pupils receive different amounts because of attendance or task completion. Some formal research distinguishes provider fidelity from participant adherence. The precise definition should be stated, rather than calling every attendance gap an instructional failure.
Intervention Evaluation asks whether the programme merits continuation or revision overall. Fidelity provides evidence about its delivery. A consistently delivered ineffective programme is still ineffective for its chosen outcome; an effective-looking but poorly documented programme still needs methodological scrutiny.
Worked Example: A Four-Week Inference Workshop
Plan: four weekly sixty-minute sessions; each includes modelled evidence selection, guided explanation, corrective feedback and an unseen independent question. Actual delivery: one cancellation, one shortened lesson and repeated substitution of teacher-read answers for pupil attempts.
The planned total is 240 minutes, and actual scheduled delivery might be only 180 minutes under the stated fictional timetable. But the main fidelity issue may be removal of independent inference practice, not the lost hour alone. A simple report should describe both content adherence and actual task opportunities.
If fresh unseen assessments reveal weak transfer, the implementation record suggests a specific hypothesis: insufficient independent practice may have limited learning. That hypothesis should be tested rather than asserted as a proven cause.
Worked Mathematics Example: Calculating Delivered Dose
A programme intends to deliver four sessions of 60 minutes each, or 240 minutes. Three sessions run fully, while one is cancelled. Delivered time is 180 minutes; the proportion of planned time delivered is 180/240 = 75%.
Now assume each session was supposed to include twenty minutes of independent practice. Only two delivered sessions included it, giving forty actual minutes against eighty planned. That component’s dose is 40/80 = 50%. The two percentages tell different stories.
The arithmetic is illustrative and does not establish a universal threshold that 75% is acceptable or 50% unacceptable. Valid interpretation requires the component’s role in the intended mechanism and pre-specified standards appropriate to the programme.
Worked Science Example: A Fair-Test Teaching Routine
A new Science teaching routine asks pupils to predict a result, identify a controlled variable, perform a safe approved investigation and explain the observation. If a teacher performs all experiments as demonstrations while pupils copy the conclusions, the intended hands-on reasoning activity has changed.
An observer could document whether pupils made independent predictions, participated in relevant measurement and gave their own explanation. Student science performance should then be evaluated on a fresh comparable task, without treating observation of the routine as evidence of achievement by itself.
Adjustments for safety and access may be essential and should be assessed by whether they preserve the learning purpose rather than by superficial similarity to the original worksheet.
Worked English Example: Feedback Quality and Independent Writing
Two classes use the same writing rubric. In Class A, pupils receive comments identifying which evidence does not support the claim and then revise a new paragraph. In Class B, teacher comments simply say “good” or “needs improvement” and pupils copy a model. Both classes can report using a rubric and giving feedback.
The planned mechanism may rely on actionable explanation and independent revision. Fidelity should therefore record the quality of feedback and the opportunity to act on it, not merely the presence of a printed rubric.
A valid assessment must still check independent writing on a suitable new prompt, and not simply count the number of comments teachers produced.
For Primary Students: Did We Actually Practise the Skill?
Young pupils can reflect on a classroom routine through simple questions: did we attempt the new problem, did we explain our answer, did we receive feedback and did we try again? The goal is not to grade a teacher but to recognise what kind of learning opportunity occurred.
Adults should keep records proportionate, privacy-preserving and focused on improving access to meaningful practice.
For Secondary Students: Materials Are Not Mechanisms
Secondary learners can distinguish a handout distributed, a demonstration observed and an independent task completed. In Maths, Science and English, each activity gives different evidence about exposure to a learning mechanism.
Ask pupils to compare a programme description with a fictional session log and decide which core element was actually missing. Their explanation should refer to the intended learning outcome rather than preference for an attractive format.
For JC Students: Distinguish Fidelity From Impact Estimates
JC learners can examine adherence, dose, teacher quality, programme differentiation, contextual adaptation and outcome measures. They should recognise that fidelity–outcome correlations can be confounded and that changes to an intervention can complicate the causal question being estimated.
A carefully reasoned implementation report describes what was done and how it might affect interpretation, without presenting the report as a complete substitute for randomisation or other credible impact designs.
A Four-Week Fidelity Clinic
Week 1: identify core programme components and propose simple observable indicators. Week 2: practise documenting delivered activities and pupil exposure, including permissible adaptations. Week 3: examine a fictional case where implementation and outcome results disagree. Week 4: propose one delivery improvement and a suitable fresh outcome check.
This timeline is an editorial teaching example, not a research-validated implementation dosage. A real education programme may require much longer preparation, staff training and appropriate evaluation safeguards.
Implementation Fidelity With AI Systems
AI can generate a fidelity checklist quickly, but it may mistake cosmetic script compliance for meaningful teaching, invent session records or wrongly infer why a pupil did not attend. An automated summary of logs also cannot show everything about the instructional quality of an interaction.
Use AI to organise already verified delivery records and identify possible differences from a documented blueprint, then check the interpretation with responsible educators. Do not upload identifying pupil records or recordings without appropriate permissions. Reassess with independent learning evidence, not simply AI-generated narrative.
The useful achievement is an accurate description of the programme actually received by pupils.
The Wintour House Test: Did the Intended Instruction Ever Happen?
An impressive intervention report can attribute outcomes to a programme name while the real teaching differs substantially from that name. Before arguing about impact, identify whether the intended active ingredients actually reached learners.
The Gold Standard is a programme description honest enough that another educator can understand its core components, actual delivery, necessary adaptations and the limits on conclusions drawn from it.
Research Anchors and Limits
Carroll and colleagues’ A Conceptual Framework for Implementation Fidelity (2007) describes adherence, moderators such as participant responsiveness and intervention complexity, and why measuring delivered intervention content matters for interpreting programme results. It is a conceptual research framework, not validation of this ten-heading school checklist.
The Education Endowment Foundation’s What Can We Learn from Implementation and Process Evaluations? (2022) explains why process evidence helps understand how education trials operated, especially when outcomes are difficult to interpret.
The SPIRIT 2025 explanation and elaboration distinguishes intervention delivery fidelity from participant adherence and calls for transparent planning of how both are measured when relevant. Its clinical trial context is much more formal than ordinary school teaching.
The CDC Program Evaluation Framework (2024) connects implementation and process information with credible evidence and justified conclusions. This provides a broader model of programme transparency.
The ten headings, hypothetical dose calculations and clinic are original Wintour House editorial synthesis. They are not an MOE/SEAB inspection standard, formal trial protocol or promise that a specific proportion of delivered lessons produces a particular grade gain.
Final fidelity check: Can the learner distinguish planned and delivered teaching, identify which core components pupils actually experienced, record adaptations accurately and separate fidelity from evidence of learning impact?
