-
One fictional school reports a rise in low-level classroom disruption, and the same evidence is given to three research teams with different assumptions about what the social world is. Learners declare a realist/positivist, critical realist or social constructionist starting point and then make five decisions - research object, legitimate question, admissible evidence, method, defensible claim - watching the specification and its coherence change as they go. Running a second lens on the same case produces a side-by-side comparison of what each design makes visible, what it hides and what action it points to.
-
Nine messy fictional accounts of daily life, and five stages that turn them into a named condition. Learners select the recurring features, choose a name, set a threshold against 400 simulated respondents, build a five-item screening questionnaire and then decide what to do when a training company, an employer, a manager, a newspaper and a funder come calling. The prevalence figure moves from 41% to 0% without one simulated answer changing, and two of the original nine accounts drop out of their own category.
-
Four fictional colleagues score 6.2 out of 7 on the same attitude questionnaire about voicing disagreement at work, and then behave on 82, 48, 28 and 14 per cent of occasions. Learners predict each colleague before the model answers, then move six conditions one at a time on a single case, then watch the attitude-behaviour correlation across 300 simulated colleagues change with the setting and with the measure. The key structural point is that opportunity multiplies rather than adds: one colleague has every internal condition the best predictor has, and rarely acts.
-
A fictional group judgement task with six things the learner can change about the situation, and a model that splits the resulting conformity into a normative route and an informational one. Six classic conditions can be loaded and are kept in a comparison table, so the private-response manipulation can be seen to collapse one route while leaving the other untouched. A second experiment opens the headline percentage into the 100 simulated people behind it, a substantial minority of whom never conform at all.
-
Nine statements by one fictional person, and seven frameworks for the self applied to all of them. Choosing a framework re-weights the shared evidence board into what is central to it, what it can use indirectly and what it has no concept for; the learner then works out which of four genuine accounts of the same person was written from that framework's premises. After four frameworks a synthesis reports what none of them has been able to make central - including a statement about a fee increase and a cut in hours that six of the seven cannot represent at all.
-
Eight undisputed events in a fictional woman's six months with a community day service, written up five times: as a case note, a risk register entry, an advocacy bulletin, a recovery-oriented discharge summary and a commissioning report. For each account the learner identifies the subject position it produces, the action it makes legitimate, who it authorises to speak and where responsibility lands, with every wrong option being the right answer for a different account. The ledger of events stays on screen throughout, and one entry in it - the decision that halved the programme's hours - appears in only one of the five accounts.
-
Fifteen rounds of a deliberately unanchored judgement: five made alone, five with three simulated others' estimates visible, and five alone again. The three others are generated from anchors drawn around the learner's own opening mean, so the emerging norm is not imposed from outside; the learner's own convergence is never forced or modelled. A final comparison shows three other simulated groups arriving at quite different norms from the identical display, and the historical experiment whose logic the activity borrows is described separately with its criticisms.
-
The learner is sorted into an arbitrary fictional category on a basis the page immediately admits is meaningless, then allocates symbolic points between one anonymous member of each category, six times. Every option prints its own arithmetic, so the option that gives the learner's own category fewer points in exchange for a larger lead is visible rather than disguised. The same six allocations can then be run in a world where the learner's category already holds four times as much, and in one where it holds a quarter - which makes visible the thing a strategy profile cannot record, that an identical choice creates an inequality in one world and widens an existing one in another.
-
One fictional crowd event can be set up five ways - how the police responded, how the crowd read that response, whether a common category was available, what the norm was, and whether people were identifiable. The model reports two things that move independently: how many of six sections of the crowd took part, and how concentrated their action was across four possible targets. A generic deindividuation account predicts the first and is indifferent to the second; the elaborated social identity model predicts the second. Four presets reach the decisive contrasts, including one in which the crowd unites completely and does almost nothing, and one in which the deindividuation description is the better fit.
-
Nine pieces of evidence about a fictional further-education college and the air in its workshop block. The learner works through the same nine items three times, each time asking a different question of them - who decides and who never has to; what counts as knowing; who is believed and what can be said - selecting the evidence each lens brings into focus and naming what it leaves in shadow. Every pass ends with the remedy that lens implies and a note on what that remedy leaves running. No theorist is named anywhere in the task; the exercise is comparative explanatory analysis on shared evidence, not matching names to definitions.
-
Two sequential experiments set in an invented country. In the first, the learner sees one simulated respondent's latency difference score, predicts how many candidates from one region that person shortlisted, and finds out - repeatedly, against a printed base rate, until the difference between a real average association and a useful individual prediction is something they have felt rather than been told. In the second, five instruments (self-report, latency, behavioural choice, informant report and institutional outcomes) are examined one at a time for what each directly observes, what is inferred from it, what alternatives remain and which claim it cannot support. Nothing on the page measures, scores or labels the person using it.
-
One fictional team, eleven pieces of evidence and two sequential experiments. In the first, the evidence arrives in three rounds - individual measurements first, then the team and its supervision, then the rota, the queue, the target and a matched comparison team - and the learner commits to an explanation after each round, so the tool can show them their own trail and point out that nothing about the team changed in between. In the second, six costed interventions across three levels compete for a budget that covers half of them, and every modelled outcome ends by naming which pieces of evidence the combination leaves in force, and by reporting that the rank order of the fourteen staff barely moves whatever is spent.