Build the class inquiry and question system

1

Research

Why this matters, and what good looks like

An inquiry board keeps questions, evidence needs and next actions visible. PBLWorks professional guidance describes sustained inquiry as posing questions, finding resources and applying information. It also places student voice and decisions within project design. This is guidance, not a study of AI question systems. Board fields and the checking sequence below are implementation choices. The retained sources establish no AI benefit.

Begin with questions students actually pose and retain their own words (project design).
Keep decisions about inquiry with students and teachers (student voice and choice).
Organise resources and checkpoints while preserving open-ended inquiry (teaching practices).

Where the evidence comes from

Strong for sustained inquiry and need-to-know systems; AI benefit unvalidated.

2

Workflow

Brief it, steer it, check it

Collect original questions first

Use optional D02, L01 and L02 outputs, or equivalent teacher documents. First invite actual teacher and student questions. Keep verbatim originals and minority questions offline. Limit this instance to twelve questions from one small class; transfer only safe exact text.

Prompt 1 · Cluster originals

CONTEXT
Approved project context and boundaries: [paste safe context]
Actual teacher/student-generated questions with question IDs and verbatim text, no learner identities: [paste safe original question register]
Teacher-confirmed available evidence routes: [paste routes]
Existing board: [paste safe board or none]
REQUEST
Suggest topic clusters by question ID and flag possible duplicates, broad, leading or closed questions. A closed question can be answered briefly; a leading question suggests the answer. Quote each original exactly and keep every ID, including singleton or minority questions.
QUALITY BAR
Do not generate replacement questions, preferred answers, priorities or owners. Do not merge or delete originals. Question IDs identify text only, never people. If originals, context or evidence routes are absent, return HOLD. If text risks identification, do not reproduce it; HOLD that item for offline handling. All cluster labels are proposals, not decisions.
FORMAT
Table: question ID, verbatim original, proposed cluster, flag with quoted evidence, evidence-route gap. Then HOLD items and a count reconciliation.

Review flags with students

Check every original against the offline register. Reject misleading clusters and mark your edits. Students discuss the flags; AI may identify evidence gaps only. Keep minority questions visible and let students revise their own wording outside AI.

Prompt 2 · Check routes

CONTEXT
Exact step 1 output checked by teacher: [paste approved output]
Teacher edits after checking originals: [paste explicit edits or none]
Verbatim original register and teacher-confirmed evidence routes: [paste safe originals and routes]
REQUEST
Check each proposed cluster against the originals and available evidence. Flag questions needing a narrower scope or a feasible evidence route. Give concise reasons referencing IDs; do not offer answers or rewritten questions.
QUALITY BAR
Preserve all original IDs and text. No owner assignment, priority ranking, deletion, new questions or preferred answers. Missing originals or routes, unsupported claims or stale route confirmation mean HOLD. Teachers and students decide revisions and priorities outside AI. Do not infer learner characteristics from questions.
FORMAT
Table: question IDs, supplied route evidence, limitation or HOLD, issue for human discussion. Include singleton/minority IDs and count reconciliation.

Record student revisions and priorities

Students revise and prioritise questions, agree evidence needs and choose owners outside AI. Record dissent and every original beside revisions. Keep owner identities offline. Submit the approved decisions for comparison only; an undecided question stays visible and on HOLD.

Prompt 3 · Check supplied board

CONTEXT
Exact step 2 output checked by teacher: [paste approved output]
Teacher edits to route flags: [paste explicit edits or none]
Original register and student/teacher decision log: [paste safe originals and versioned revisions, priorities, evidence needs, status and review triggers]
Human-built board with original text and revision history: [paste safe board with owner-confirmation status only, no people or identifying team codes]
REQUEST
Compare the supplied board with the originals and decision log. Flag missing IDs, changed originals, unrecorded revisions, lost minority questions or unsupported priorities, routes, statuses and owner confirmations.
QUALITY BAR
Do not author or select revisions, priorities, answers or owners. Keep exact originals and all dissent; a majority choice does not erase a minority question. Missing decisions or infeasible routes mean HOLD for affected rows. The actual owner mapping remains offline. No personal data or grading.
FORMAT
Discrepancy table with question ID, quoted source evidence, board difference and human action. Include count reconciliation and HOLD rows. Do not rewrite the board.

Audit the board students selected

Review flags with students and record explicit corrections. Select the board together, then audit it. Check evidence access and actual owners offline before teacher release. If participation or agreement cannot fit within 60 active minutes, stop and re-scope.

Prompt 4 · Audit selected board

CONTEXT
Exact step 3 output checked by teacher: [paste approved output]
Teacher edits and subsequent student decisions: [paste versioned decisions or none]
Original register, decision history and teacher/student-selected board: [paste safe originals, history and selected board]
Teacher offline checks of evidence access, actual owners and student participation: [paste completion categories and unresolved items only]
REQUEST
Audit whether every original remains verbatim and traceable, minority questions survive, revisions and priorities match human decisions, and evidence needs, status, review trigger and offline owner confirmation are present.
QUALITY BAR
Do not replace questions, embed preferred answers, choose owners, rank priorities or approve release. Missing decisions, unresolved evidence feasibility, stale claims or absent human checks mean HOLD. Do not infer participation or consensus from tidy text. Human release and any revisions remain outside AI.
FORMAT
Two lists: HOLD issues with ID, exact evidence and human action; matched items with source evidence. Reconcile original and final ID counts. State that textual comparison cannot verify real participation, evidence access or ownership.

Check before you use it

Are actual teacher and student questions preserved verbatim beside every revision?
Are all original IDs and minority questions still visible, including deferred or disputed questions?
Did students revise and prioritise questions and decide owners outside AI?
Have you checked evidence routes yourself and kept infeasible inquiries on HOLD?
Do board categories, priorities, evidence needs, status and history match recorded human decisions?
Have you kept identifiable material and owner mappings offline and completed participation checks?
Have you rejected preferred answers and stopped or re-scoped any unfinished agreement?

Stop rule

Stop if AI replaces student questions, embeds preferred answers or creates inquiries with no feasible evidence route.

3

Reuse it

So the next one takes minutes

Suitable

Build a skill

Save this as a reusable instruction you can run on any project.

The bounded comparison and checking sequence can repeat with new non-sensitive materials and explicit human decisions.

Not suitable

Build an agent

Chain the steps, with a checkpoint where you approve.

Live student negotiation and teacher release require direct participation; unattended monitoring is outside this task.