Adapt the rubric to the project

1

Research

Why this matters, and what good looks like

A rubric should reveal project learning, including process evidence. PBLWorks warns that an end-product focus can hide that learning. EEF connects useful feedback to the gap between current and intended learning. Ofqual cautions that agreement with human marks alone does not establish validity.

These receipts support pedagogical safeguards, not validated AI descriptors, accuracy or time savings. Ofqual concerns high-stakes qualifications; it does not directly regulate classroom drafting. UK guidance is not Belgian school policy. Teacher validity checks remain necessary.

Preserve approved learning constructs and available evidence.
Keep product, disciplinary and process evidence visible.
Use observable differences between adjacent levels.
Teachers place and annotate permission-cleared anchors.
Keep weighting, fairness and accommodations with the teacher.

Where the evidence comes from

Strong for rubric use; AI benefit unvalidated.

2

Workflow

Brief it, steer it, check it

Audit what the rubric must measure

Confirm the goals and constructs before drafting. Use scratch mode without a rubric; improve mode requires its full text. Stop for missing permissions or unsafe inputs. Never enter personal data, confidential records or identifiable work into AI tools.

Prompt 1 · Rubric inventory

CONTEXT
Mode and existing rubric: [scratch with none, or improve with full rubric].
Approved standards and goals: [paste exact text].
Teacher-approved constructs: [paste complete list].
Product, audience and constraints: [paste exact text with units and scope].
Available product, process and individual evidence: [paste plan].
Permission, privacy, safety and accommodation decisions: [paste teacher confirmations and approved routes, or unknown].

REQUEST
Inventory every approved construct verbatim; map supplied goals and evidence without replacing constructs.
Check permissions, privacy, safety, accommodations, unsupported criteria and evidence gaps separately. Hold descriptor drafting if any is unresolved; ask the teacher to decide.

QUALITY BAR
Invent no goals, criteria, weights, levels or existing rubric. Write absent in current criterion when no rubric is supplied. Flag unmapped goals separately, never create a replacement construct. Preserve feasibility and all site, budget and energy constraints when supplied. AI never scores, including named levels.

FORMAT
Return a complete table: approved construct | exact goal or teacher clarification needed | evidence | exact existing criterion or absent | gap.
Then separate gate lines: permission | privacy | safety | accommodations | unsupported criteria | evidence gaps; give each status, source quote and teacher action.

Draft observable descriptors

Proceed only after teacher resolution of inventory gaps. Draft the whole matrix, not sample rows. The teacher owns validity, weighting and final language. No work samples are placed by AI.

Prompt 2 · Descriptor draft

CONTEXT
Teacher-approved inventory and resolved gate decisions: [paste full record].
Teacher-approved constructs and evidence: [paste complete list and evidence mapping].
Teacher-selected number and names of levels: [paste them].
Age, language, fairness and approved accommodations: [paste requirements].

REQUEST
Draft the full analytic matrix, one row per approved construct and one column per supplied level. For each adjacent pair identify its single main observable distinction. Keep all other demands stable within that pair; revise wording if multiple demands change.

QUALITY BAR
Hold if approval, constructs, evidence, levels or accommodations are missing. Add no criteria, weights or standards. Preserve every construct verbatim and all supplied constraints. Do not reward polish, obedience, fluency or length unless an approved goal requires it. Never assign samples numeric scores or named levels.

FORMAT
Full construct-by-level matrix with evidence needed and accommodation note. Then every adjacent pair: construct | lower descriptor quote | upper descriptor quote | one changed dimension | ambiguity for teacher review. No sample rows or omitted cells.

Audit teacher-placed anchors

Teachers test and place permission-cleared anchors themselves. Record their criterion-specific levels and annotations before the AI audit. Unknown placements stay unknown. Hold reuse without permission; synthetic tests cannot establish real-school permission, fairness or validity.

Prompt 3 · Anchor wording audit

CONTEXT
Draft full rubric: [paste complete matrix].
Permission-cleared non-identifying anchor excerpts: [paste exact excerpts and provenance].
Teacher placements and annotations: [paste criterion-specific levels or not judged, exact evidence quotes and teacher rationale].
Teacher permission, safety and accommodation confirmations: [paste decisions].

REQUEST
Copy the teacher's placements and annotations verbatim into an annotated-anchor record. Audit descriptor ambiguity only against those supplied teacher decisions. Flag overlapping or unclear wording with exact descriptor and anchor quotes. Ask the teacher to resolve concerns; do not revise placements or descriptors here.

QUALITY BAR
If placements, annotations or permissions are missing, HOLD and name the missing teacher decision. Never infer, assign, confirm, contest or reassign a sample's level or score, even a named level. Do not say a sample fits, meets or fails a level. Attribute every reproduced placement to the teacher. Not judged remains not judged. Do not invent a prior revision or subject judgement.

FORMAT
Annotated-anchor record: anchor | teacher placement (verbatim, attributed) | teacher evidence quote | teacher annotation (verbatim).
Separate wording audit: descriptor quote | exact anchor quote | ambiguity relative to teacher annotation | question for teacher. End with teacher revision/approval required; no scoring verdict.

Create student-facing criteria

The teacher resolves wording concerns and approves the full matrix and annotated anchors. Only then translate the criteria. Retain the approved rubric and teacher annotations alongside the student version.

Prompt 4 · Student version

CONTEXT
Teacher-approved full rubric and approval record: [paste them].
Teacher-approved annotated anchors: [paste record].
Known terms and teacher-approved definitions: [paste them, or none needed].
Approved language and access requirements: [paste them].

REQUEST
Draft a student-facing statement for every approved construct, preserving its learning demand and constraints. Retain the supplied teacher-annotated anchors unchanged as a companion record.

QUALITY BAR
HOLD without teacher approval, complete matrix or annotations. Add no criteria, examples, grades, levels for samples or definitions. Do not simplify away learning demand or alter accommodations. The full rubric remains authoritative.

FORMAT
Complete table: exact construct | student-facing statement | full rubric row reference. Then verbatim teacher-annotated anchors, clearly attributed, and any translation concern for the teacher.

Check before you use it

Does every approved construct remain, including feasibility, with no invented criterion?
Are all matrix cells present and adjacent distinctions observable?
Did the teacher supply every anchor placement and annotation before AI auditing?
Has AI avoided assigning or endorsing any score or named level?
Are consent, privacy, safety and accommodations confirmed separately?
Has the teacher approved validity, fairness, weighting and the final student version?

Stop rule

Stop if criteria reward polish/compliance over learning, invent standards or cannot be evidenced; AI never scores.

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 rubric audit, descriptor drafting and sample-testing sequence stays stable across projects.

Not suitable

Build an agent

Chain the steps, with a checkpoint where you approve.

Validity, weighting and fairness need a bounded teacher decision rather than recurring autonomous action.