---
name: marking
description: "Experimental rubric-first marking and feedback workflow for testing with synthetic work only. Validate a test rubric, record evidence, calculate provisional scores and review feedback. Not suitable for live assessment, real student submissions or official grades."
---

# Marking — testing only

*Author: Cam Anderson (hello@camanderson.au). Licensed CC BY 4.0, adapting MIT-licensed workflow ideas from Clyde Rankin's essay-skills.*

**Experimental: suitable only for testing with synthetic work. Do not use this skill for actual student assessment, official grades, assessment decisions or feedback released to students.** Every output must be labelled `TEST ONLY — simulated marking; not an official result`.

This skill builds on others' work, including ideas from an earlier six-stage essay workflow. Read [references/provenance.md](references/provenance.md) for the recorded design lineage and attribution limitations.

## Test inputs and boundaries

Use fictional test submissions, test rubrics and task instructions that staff are permitted to share. Do not request, read, upload, retain or score real student submissions, including de-identified actual work. Do not include student names, identifiers, grades or identifiable feedback. Use a synthetic case label and neutral greetings. If real student material is supplied, stop that intake and request a purpose-created synthetic example.

The supplied rubric is the source of truth for the simulation. Do not score until the test assignment instructions and rubric are available and readable. If the rubric is missing, contradictory, unweighted when weighting is required, or too vague to support a defensible judgement, request clarification. Do not invent criteria or silently apply general notions of quality. A test rubric is not evidence of institutional approval.

## Workflow

1. Confirm that inputs are permitted synthetic test material, then read the task instructions, rubric, fictional submission and supplied test policy/context.
2. Validate the rubric: criteria, descriptors, bands, weights, maximum scores, total marks, grade boundaries, thresholds and permitted deductions. Record its version; a changed rubric requires a separately versioned test run.
3. Read the fictional submission for overall understanding, noting evidence locations and uncertainty without converting first impressions directly into scores.
4. Evaluate each criterion independently. Record evidence, descriptor match, provisional score, rationale and confidence.
5. Run [references/fairness-checks.md](references/fairness-checks.md).
6. Calculate the weighted test result and independently check the arithmetic and supplied grade boundary. Record the actual method and result; do not claim an independent check that was not performed.
7. Draft simulated feedback using [references/feedback-guide.md](references/feedback-guide.md).
8. Complete the test-output checks and report limitations. Borderline, ambiguous, policy-sensitive or low-confidence results require test moderation rather than false certainty. No test status authorises release or production use.

## Fairness principles

- Score only what the rubric assesses.
- Do not let one weakness contaminate unrelated criteria unless the rubric explicitly connects them.
- Do not penalise the same issue twice.
- Do not infer missing evidence where the task did not require a feature or the material cannot fairly establish it.
- Separate observed evidence, interpretation against the descriptor and judgement.
- Treat writing style, sophistication, formatting preferences and apparent effort as assessable only when the rubric makes them relevant.
- Do not let the warmth of feedback influence a score or invent strengths to balance criticism.

## Test record and feedback

For each criterion preserve `criterion → weight → evidence location → descriptor match → provisional level/score → rationale → confidence`. Use [references/marking-record.md](references/marking-record.md). Do not award a total first and retrofit criterion scores afterwards.

Simulated feedback should be accurate, respectful, encouraging and useful. Begin and end with a brief, proportionate acknowledgement, using a neutral greeting rather than a real name. Link strengths and next steps to observable evidence. Encouragement must not imply a stronger result than the rubric supports. Never present the simulated feedback as ready for delivery to a student.

Use one test status:

- `test complete`: rubric validated, evidence and calculations recorded, simulated feedback checked; the simulation is complete, not validated for real use.
- `requires clarification`: rubric, task, synthetic material or test policy is insufficient.
- `requires test moderation`: a provisional simulation exists but ambiguity, borderline scoring or low confidence needs review.

Report what was checked, what remains uncertain and whether observed behaviour supports further controlled testing. This package makes no claim of validated fairness, reliability, institutional approval or suitability for operational marking.