AI does
Rehearsal, neutral phrasing and option generation.
You do
Safeguarding, power analysis, fairness and intervention where harm/bullying is present.
1 Research
Why this matters, and what good looks like
A role agreement records what participants freely choose to do and when they will review it. This method covers one low-level task or role disagreement, with teacher-only AI rehearsal. EEF synthesises collaboration research and rates its evidence security low. The frozen Moderate classification concerns structured collaboration, not mediation. MyPBLWorks offers practitioner guidance; only its public description was available, not the login-gated guide. Neither source establishes AI mediation effectiveness. The rehearsal and agreement checks below are authorial process choices.
- EEF distinguishes structured collaboration from unstructured group work: bring the existing task and role expectations.
- EEF stresses support, practice and opportunities to articulate thinking: hear actual voices offline.
- MyPBLWorks describes agreeing contributions and revisiting contracts: retain participants' decisions rather than invent consensus.
Where the evidence comes from
Moderate for structured collaboration; AI benefit absent/very weak.
2 Workflow
Brief it, steer it, check it
Bound the rehearsal
Use your L03 contract, C01 task record and C03 commitments, or equivalent teacher documents. Privately check suitability, harm and coercion first. Bring only generic task constraints, no student profiles or conflict narratives. Stop and re-scope beyond 60 active minutes.
Prompt 1 · Scope rehearsalCONTEXT Generic task and role constraints: [task constraints] Teacher-only gate record with no personal details: [human gate] REQUEST Restate the bounded rehearsal task and list missing materials. Offer two neutral teacher opening questions about the task. This is preparation for the teacher alone, never a live conversation with participants. QUALITY BAR Proceed only with an explicit supplied teacher decision that this is a suitable low-level task disagreement and that harm, bullying, coercion and power concerns have been checked offline. If missing or unresolved, return HOLD and refer to human school procedures without questions or options. Never assess suitability yourself, investigate, profile students, assign blame or request narratives. Do not invent policies, facts or approvals. FORMAT Status; supplied gate; task constraints; missing materials; two rehearsal questions only if the gate is supplied. Label questions proposals, not participant voices.
Compare neutral options
Check the first output against your materials. Supply your accepted scope and exact edits. Rehearse two task arrangements privately; do not ask AI to infer fairness or choose people. Take only your selected questions into the human conversation.
Prompt 2 · Compare arrangementsCONTEXT Exact approved step 1 output: [approved scope] Teacher acceptance and edits, with renewed gate: [teacher edits] REQUEST Compare two possible arrangements within the supplied task constraints. Give one neutral question the teacher could ask about each arrangement. Leave all people, choices and commitments unassigned. QUALITY BAR HOLD if approved scope or explicit edits/acceptance are missing, stale or the human gate is unresolved. No actual mediation, simulated student replies, fairness verdicts, safeguarding or disciplinary decisions. Identify practical trade-offs only. Never infer participant agreement or recommend a winner. Keep unsupported claims HOLD. FORMAT Table: proposal ID | task arrangement | supplied constraint | practical trade-off | teacher rehearsal question. Follow with unresolved inputs and a reminder that participants choose offline.
Record the human agreement
Facilitate offline. Hear each perspective without AI; stop for harm, coercion or unresolved suitability. Participants may decline agreement. Keep their exact voices and named role assignments locally. Supply only a nonidentifying agreement summary and explicit teacher decisions after checking them.
Prompt 3 · Organise agreementCONTEXT Exact approved step 2 comparison: [approved comparison] Teacher selection and edits: [selection] Minimised human conversation record: [human agreement] REQUEST Organise only the supplied facts, perspectives, shared goal, revised role tasks, chosen action and check-in date. Mark each as supplied, disputed or missing. Use a local-record pointer for exact voices and named role assignments kept outside AI. QUALITY BAR HOLD if any participant agreement, renewed teacher suitability gate, chosen action or check-in date is missing or disputed. Never turn proposals into agreement, resolve disagreements, assign students or invent consensus. No identifying details or sensitive narrative. Ask the teacher to resolve gaps offline; never continue a live conflict through AI. FORMAT Table: agreement field | supplied content | provenance/status. Then unresolved items and HOLD or organised for human checking. Distinguish participant decisions from AI proposals.
Audit the selected record
Compare the organised record with your private originals and participants' decisions. Supply corrections and your selected record for a completeness check. Resolve flags with people; record your final use, revise or stop decision offline. AI cannot authorise an agreement.
Prompt 4 · Check agreement recordCONTEXT Exact approved step 3 output: [approved agreement] Teacher-selected record, edits and original checks: [selected record] Teacher final decision after participant confirmation: [final decision] REQUEST Audit the selected record against supplied decisions without rewriting voices or choices. Check facts, perspectives, goal, role tasks, chosen action, check-in date and local pointer. Identify unsupported or stale statements. Report the teacher's final decision verbatim. QUALITY BAR No certification of safety, fairness, consent or truth. If human gate, participant confirmation, original checks or final decision are missing or unresolved, HOLD. AI does not release anything. Preserve dissent. Any harm, coercion or suitability concern returns to human school procedures. Actual people work is outside your observation. FORMAT Table: field | supplied supporting record | gap or mismatch. Then unresolved flags, supplied final decision verbatim, and separate AI authority: none. Mark real-world verification unobserved by AI.
Check before you use it
- Confirm suitability, power and harm checks yourself; stop for unresolved concerns and use human school procedures.
- Keep identifying narratives, profiles, exact participant voices and named assignments outside AI.
- Check every recorded fact and perspective against your private originals; preserve dissent.
- Confirm participants freely chose the goal, role changes and action; never treat a proposal as consent.
- Confirm the check-in date and record your final decision locally.
- Treat AI wording as teacher rehearsal only; never use AI for live conflict, safeguarding or discipline.
Stop rule
Never delegate live conflict, safeguarding or disciplinary decisions to AI.
3 Reuse it
So the next one takes minutes
Build a skill
Save this as a reusable instruction you can run on any project.
Conflict suitability and participant agreement require context-specific human judgement.
Build an agent
Chain the steps, with a checkpoint where you approve.
Do not automate participant interaction, decisions or release.
Use cases are starting points with one perspective. Disagree with the process, that's expected. Make it yours.
Co-funded by the European Union under Erasmus+ KA210-SCH. No student data. No teacher material stored.