Blueprints and cut scores

The shape of an exam

```

Blueprint (role track, level)

├── Weighted domains → each a share of the exam, all summing to 100%

├── Form length → how many items a single sitting draws

└── Cut score → the pass mark, from the Angoff panel

```

Weighting the domains

Each domain is a share of the exam. The builder keeps the weights honest — they must total 100% before a blueprint can go to review — and shows, per domain, whether the item bank holds enough live items to fill a form. A blueprint whose bank is too thin is not shippable, and the screen says so.

Setting the cut score (Angoff-lite)

The pass mark is not guessed. For each item, an expert answers one question: what fraction of a just-barely-competent candidate would answer this correctly? The sum of those judgments is the cut score. A standard-error margin can lower it slightly when a panel disagrees — a documented choice, never a silent one.

Why nothing here can be gamed

• Timing is the server's. The deadline is derived from when the attempt started; the browser only displays the time left.
• Scoring is the server's. The client submits answers only — no score, no pass flag.
• Submitting is idempotent. A double submit returns the first result.
• Answer keys never leave the server. The exam client is handed the questions with every correct answer and rubric stripped out.

Publishing

Authors draft; only reviewers publish. A published blueprint is locked so its cut score cannot change under attempts already issued — clone it to a new version to make changes.