From lesson 1, readable now without an account: Frame a bounded improvement problem
When a school team says, “We need to raise attainment,” it has named an understandable ambition but not yet a problem a specialist can investigate. The phrase might refer to difficulty with one concept, a difference between classes, missing pupils, or a change in measurement. Start by turning the ambition into a question that can be checked. The specialist does not need to claim a cause in advance. They need to define the observed pattern, the group, the time window, the available evidence, and the boundary between assigned work and decisions requiring approval.
A problem card is a short tool for this step. Ask: Who is affected? What observable response or outcome changed? Compared with what? Over what period? How was it measured? Which decision is requested? For example: “On two comparable classroom fraction checks over six weeks, 18 of 30 pupils in Year 7 class A did not explain the denominator, compared with 9 of 29 in class B.” That is a reviewable observation, not a judgment that class A's teacher is less capable. Before drawing an inference, check item comparability, marking criteria, and absences. The proportions are 60% and about 31%, but the difference does not establish a cause. A harder second check could make the comparison misleading.
Keep three layers apart. A signal is what the data show. A hypothesis is a possible explanation. A decision is an action selected by someone with authority after examining the hypothesis. Here the signal is a difference in denominator explanations. Possible hypotheses include a different teaching sequence, attendance patterns, or different wording of the question. A suitable early decision may be to review anonymized pupil work and test items with the teachers. It would be premature to prescribe corrective training for one teacher. Separating the layers protects people from an accusation based on one number and allows the hypothesis to change when new evidence arrives.
Authority belongs on the card too. A specialist may be asked to collect anonymized samples, facilitate a short meeting, and prepare options for the school head. That assignment does not itself authorize timetable changes, formal staff appraisal, or disclosure of identifiable pupil information beyond approved channels. Put two columns on the card: “Within my assignment” and “Approval or escalation required.” If authority is unclear, seek clarification before an external action. UNESCO's leadership report links effective leadership to opportunities to decide, while stressing clear decision authority, support, and accountability. We use that as a design principle here, not as a single legal rule for every country.
A useful aim describes a desired change in learning without promising an outcome the team cannot guarantee. For example: “Over four weeks, test whether a recurring short explanation task helps pupils in both classes explain the meaning of a denominator in a new example; compare work samples before and after.” This is a learning test, not a declaration that the task will solve the issue. Choose an outcome measure, such as the proportion meeting a common explanation rubric, and an implementation measure, such as how often the task was used as intended. Read the two together. Better results with poor implementation suggest another influence; faithful implementation without improvement calls for a revised hypothesis or task.
Before approving the card, make a four-column table: observation, measurement quality, alternative explanation, and next check. Give each number a source, date, and denominator. If one of three classroom results tests a different skill, do not average all three and call it a trend. A whole-group mean can also hide a subgroup with a different need. Examine the distribution by class or task when the sample and privacy rules permit, while avoiding a judgment about an individual from a tiny group. The US Institute of Education Sciences recommends making data use part of a continuing cycle of instructional improvement and setting a clear schoolwide vision for it. For this specialist, the card therefore opens a cycle of observation, question, check, bounded action, and fresh reading; it does not close the case.
Ask the assignment owner what evidence would change the proposed diagnosis. If the answer is “none,” the exercise may be justifying a previous decision rather than testing a problem. Specify a check in advance: for instance, two teachers independently apply one rubric to three anonymized samples and discuss disagreements before coding the rest. Agreement between the teachers shows consistency in reading the samples, not improved pupil learning. Then match effort to stakes. A small sample can reveal whether a broader review is needed; a whole-school timetable change in response to one ambiguous item would be disproportionate.