Teaching and seminar activities
All scenarios below are constructed. They are not excerpts from private PISA conversations, student work or the internal seminar deck. Activities introduce a way to reason about proposals, evidence, authorised claims and contribution traces; they do not validate the framework.
Suggested 35-minute workshop
Illustrative facilitation timings, not a tested delivery duration. Adapt the pace, accessibility and disciplinary level to the group.
| Time | Activity | Participant output |
|---|---|---|
| 5 minutes | Skim the field guide, especially the pre-use permission gate. | One permitted and one prohibited use of research material. |
| 8 minutes | Scenario A: guided practice using the illustrated workflow. | A conditional claim and the check that could overturn it. |
| 9 minutes | Scenario B: interpret a qualitative record and ambiguous contribution history. | A warranted interpretation, an attribution limit and a confidentiality decision. |
| 8 minutes | Scenario C: independent transfer to an archival study. | One verification plan, one rejected claim and the minimum relevant trace. |
| 5 minutes | Compare answers and identify upstream stop rules. | One reason to revise, abandon or defend a consequential move. |
A compact response format
For each scenario complete four lines: (1) proposed move; (2) evidence still needed; (3) strongest defensible claim at present; (4) contemporaneous record worth keeping, if any. Add one data-sharing restriction if the case contains third-party material. Participants need not fill out every resource template.
Example four-line response for Scenario A (not the only acceptable answer):
- Proposed move: Compare the job insecurity–wellbeing association among respondents in different employment-contract groups.
- Evidence still needed: Check group coverage, questionnaire measures, survey design, suitable uncertainty estimates and whether the group patterns actually differ.
- Strongest defensible claim now: No empirical difference can yet be reported; the group comparison is only a proposed analysis. Even if supported later, the cross-sectional data alone would not establish that contracts cause mental-health damage.
- Record worth keeping: If the proposal changes the design, retain a permitted dated note of the suggestion and the researcher’s conditional decision; append any later verified results separately.
The response shows the level of specificity expected. Alternative checks and interpretations may be appropriate when justified by the study design. Scenarios B and C remain independent exercises.
Scenario A — Guided practice: a causal rewrite
Constructed situation. A researcher is considering an AI-proposed comparison of insecurity–wellbeing associations by employment-contract category. A writing assistant substitutes “precarious contracts damage mental health” for a conditional descriptive statement. No identification strategy or checked output has been provided.
Task: identify the exact comparison awaiting verification, the words that exceed the evidence, a possible failed-check outcome and what an admissible eventual sentence would have to preserve.
Facilitator checks: participants must not assert that a difference was observed; must explain that contract categories are the central proposal; must distinguish self-reported wellbeing from clinical mental health; and must make the claim conditional on real design-appropriate checks. The full illustration provides the reasoning, not simulated data.
Scenario B — Qualitative analysis: a dated codebook is not an origin certificate
Constructed situation. An AI assistant and researcher considered possible interview themes. Months later, the researcher claims the assistant discovered a theme about trust. A dated codebook contains the theme label, but no preserved record establishes who suggested it first. Interview extracts are covered by confidentiality undertakings. The researcher proposes uploading raw extracts to an external assistant and then publishing selected quotations.
Tasks: identify what the codebook establishes; name two plausible alternative contribution histories; explain the interpretive work needed for a defensible thematic claim; and distinguish permission for AI processing from permission to quote publicly.
Facilitator checks: the dated codebook establishes a theme label in a particular version, not origination, adequate interpretation or consent. Possibilities still include a researcher proposal, an AI suggestion or joint development. Suitable interpretive challenge may involve analytic memos, contextual extracts, attention to discordant accounts and reflexive decisions, as appropriate to the chosen qualitative approach—not an automatic demand for statistical agreement. If disclosure is unapproved or unclear, raw interviews must not be uploaded. Publication requires its own rights and disclosure assessment even if analytic use was authorised.
Scenario C — Independent transfer: archival interpretation
Constructed situation. An AI assistant proposes that three surviving local-government circulars show a council-wide consensus in favour of a housing policy. The researcher has only verified that the circulars exist; their dates, intended recipients, authorship and relationship to a disputed committee decision have not yet been investigated. A subsequent AI-assisted draft says: “All councillors supported the policy.”
Tasks: apply the four-line format without copying the survey or interview answer. Decide which originals and contextual materials should be inspected, what narrower claim the three circulars might warrant, why institutional communication is not equivalent to unanimous belief, and which consequential change deserves a contemporaneous trace.
Facilitator checks: verification of document existence is weaker than historical interpretation. Participants should propose source criticism including authorship, provenance, timing, intended audience, contrary records and survival/selection bias where relevant. At most, appropriately checked documents may support a claim about what particular circulars communicated; they cannot by themselves prove unanimity. A trace is useful if the AI suggestion or subsequent rejection changes the research trajectory, not because each source was read. Different historiographic approaches may require different assessments.
Debrief and transfer
- Did any team mistake a proposed method or theme for evidence?
- Which new source, check or interpretation would genuinely reverse the favoured claim?
- Which contribution history remains underdetermined even when the analytical result is sound?
- Was restricted material protected before any AI use, rather than only before publication?
- When does repeated revision erase a warranted distinctive argument rather than correct a weakness? Consult Risks & Failure Modes.
Facilitators may use the protocol crosswalk and resources as references. The PISA cases are bounded documented illustrations and should not be presented as validation data for the constructed exercises. No permission to circulate the separate internal seminar slides is implied; see the seminar policy.