# Question-quality review protocol

Status: proposed protocol, no educator assigned and no product benchmark completed. The open practice pack is an original AI-assisted draft, not a sampled output of the Quizverse generation service. Automated arithmetic checks are not educator review.

## To run a genuine product benchmark

Freeze the permissioned source corpus, task/prompt, tool/model version, configuration and random sampling plan before generating. Record all outputs and failures rather than selecting only good questions. Record source rights and exact output hashes, denominator and exclusions. Keep source material and expected answers separately reviewable. Do not include private learner work without permission.

For each sampled question record source trace, correct key, explanation correctness, ambiguity, distractor quality (if applicable), language/accessibility, level fit and unsupported factual statements. Use a named qualified educator; record credentials, scope and conflicts. A second reviewer resolves disagreements, preserving original judgments. Report both question-level error rates and total issue counts with denominators. Do not generalize from this 30-item hand-prepared draft to product accuracy or exam results.

## Editable review record

Question ID: __. Source/tool version: __. Reviewer and relevant expertise: __. Date: __. Correct key? unreviewed. Explanation? unreviewed. Ambiguous? unreviewed. Appropriate level? unreviewed. Accessible wording? unreviewed. Issue/reason: __. Correction: __. Second reviewer/disagreement: __. Approved version/hash: __.

Current record: 30 drafted practice questions; 30 numeric keys programmatically checked; zero numeric mismatches in that check. Educator-reviewed sample: 0. Educator error/ambiguity counts: unavailable, not zero. Actual tool-output sample: 0. Effect on learning: unmeasured.