---
name: certification-items
description: Write and revise certification questions, scenarios, practical exercises and scoring rubrics against a blueprint and source versions.
license: MIT
---

# Certification Items

Read [item development](../certification-workshop/references/method.md#item-development) and [records](../certification-workshop/references/records.md). Author to one primary domain and explicitly linked competencies. Each item needs a source location/version, prompt, type, expected response or rubric, rationale, maximum score and revision. Mark new work draft. Source-based question creation builds on the user's Bloom workflow without treating cognitive level as empirical difficulty.

For selected response, make distractors plausible and comparable, establish the keyed answer from the supplied situation, and explain why alternatives fail. Avoid hidden assumptions, wording cues, irrelevant trivia, unsupported absolutes and generic 'best algorithm' questions without constraints. For practical tasks, specify environment/setup, starting state, candidate deliverables, observable scoring anchors, partial credit and scoring boundaries. Do not provision paid environments or use real production data to stage a task.

Before handing off selected-response items, compare all options for plausibility, parallel structure, and cues. If the key alone contains a long checklist, move shared context into the stem and make each alternative a plausible decision at the same level. Do not pad distractors to hit an arbitrary word count. For practical rubrics, split overloaded criteria into separately observable decisions while retaining the intended total score. Explain each point and any conflicting-response rule so two raters can apply it independently. Record changes and remaining judgment calls in the review packet.

Keep instructional feedback and answer keys restricted. Public practice material should be separate from live items. Increment revision for substantive changes and obtain fresh review; never rewrite history or carry approvals across revisions. Run the helper's `validate` command, then generate `review-packet` for substantive review. Automatic checks find structural problems; they do not approve content or prove measurement quality.