Stop the privacy risk before scoring

Published 2026-10-13 · Version 1.0

Reject a superficially anonymous upload, then inspect an unsupported criterion annotation using wholly invented work.

Proposed estimate: 25 minutes. Proposed practice estimate. This exercise is not operational grading, legal review, or approval to use real learner data.

Existing workflows

• Calibrate LLM-assisted scoring with synthetic work
• Write feedback that leads to a revision
• Turn formative evidence into a next step

Synthetic proposal: where to stop

All details in this case are invented. A fictional teacher proposes uploading a real worksheet after replacing its name with “Student C.” Its story would still mention a unique local competition, a specific date, and a three-person club. The teacher also proposes uploading that club’s three-row results as an “anonymous aggregate.” No upload has occurred.

Your first task is to choose: proceed, remove one more field, or hold the proposed disclosure. Identify the missing authorization and privacy checks before reviewing the scoring exercise. Do not create a similar example by changing details in actual student work.

Worked privacy decision

Hold the disclosure. A pseudonym does not remove distinctive content, and linked details or a tiny group can make identification possible. A table is not safe merely because it is called an aggregate. U.S. Department of Education guidance explicitly includes information identifiable through combinations of details; this exercise does not determine legal compliance.

Keep actual work in its approved institutional location. Any proposed real-work use requires a separate institution-approved environment and process covering the specific purpose, legal authority, permitted data, vendor terms, access, retention, human review, and correction procedures. A school-approved product is not blanket permission for every data type or use. Ask the responsible institutional reviewer; do not self-certify de-identification. For this practice, use only the invented inputs below.

Invented task, rubric, and independent reference

Task: compare assembly times A = 4, 6, 8 minutes and B = 2, 6, 10 minutes using means and ranges; limit the conclusion to these observations. Checked answer: both sums are 18, both means are 6 minutes; ranges are 8 − 4 = 4 and 10 − 2 = 8 minutes.

Use three separate criteria. C1: 2 for both means with valid methods, 1 for correct values without methods, 0 otherwise. C2 uses the same structure for ranges. C3: 2 for connecting A’s smaller range to less observed variation with an explicit sample limit; 1 for the link without that limit; 0 for an absent interpretation or a claim guaranteeing future results. For demonstrably truncated or unreadable input, abstain rather than invent evidence.

Anchor A: “Each sum is 18; 18 ÷ 3 = 6 each. Ranges: 8 − 4 = 4 and 10 − 2 = 8. A varies less in these three observations only.” Reference: 2, 2, 2. Anchor B: “Both means are 6; ranges are 4 and 8.” Reference: 1, 1, 0.

Held-out C: “Each sum is 18; dividing by 3 gives 6. A: 8 − 4 = 4; B: 10 − 2 = 8. A is always more predictable.” Record your judgment before revealing this reference: 2, 2, 0. “Always” overstates these observations. An imagined model answer of 2, 2, 2 on the grounds that the writer “understands the sample limit” invents evidence. Reject that annotation; do not average it with yours.

Feedback, adaptation, and stopping rule

A usable comment is: “Your calculations match the table. Revise the claim about every future attempt so that it says only what these observations support. Connect your revised claim to the ranges.” Reserve a short revision opportunity. A synthetic revision could say: “In these three observations, A varies less because its range is 4 minutes rather than 8; future times may differ.”

Fresh check: A = 5, 7, 9; B = 3, 7, 11. Both means are 7; ranges remain 4 and 8. Inspect the explanation, not just the numbers. A terse response containing the same evidence should receive the same criterion annotations; polished prose cannot supply missing reasoning. No scores belong to actual learners or a gradebook. This small rehearsal establishes neither reliable scoring nor permission for automatic consequential grading.

Self-check

• Did you stop before disclosure rather than relying on a pseudonym?
• Did you distinguish an approved product from an approved data use?
• Did C3 follow the words actually present?
• Did the feedback leave the revision’s reasoning to the writer?

Evidence and limits

The privacy sources explain risks and institutional review considerations. The synthetic scoring references and feedback are authored for practice, not a validated scoring instrument or legal determination.

What constitutes de-identified records and information? — U.S. Department of Education, Student Privacy Policy Office. Reviewed 2026-10-13.

The Department’s FERPA guidance includes information that can identify a learner in combination with other details. This supports the warning that removing names alone is insufficient.

Limit: The brief workflow cannot establish that a record is de-identified or that disclosure is authorized. Institutional privacy and legal review remain necessary for any proposed real-data use.

Privacy and Education Technology — U.S. Department of Education, Student Privacy Policy Office. Reviewed 2026-10-13.

The Department links guidance and terms-of-service review resources for evaluating online educational services and their handling of information.

Limit: A source link or completed exercise does not approve a vendor, create a lawful basis for disclosure, or certify FERPA compliance.

Published October 13, 2026. All time budgets are proposed planning estimates, not measured results. Public sources inform stated principles but do not evaluate or endorse these paths, prompts, worked answers, or exercises. No learning gains, workload savings, or historical classroom outcomes are claimed.