# Assessment Model and Readiness Scoring How OpenAI Academy assessments work, how our independent mock exams differ, how question anatomy is designed, and how the readiness indicator should be read. import { Accordions, AccordionItem } from '@prosefly/astro-components'; Read this before your first mock exam. It explains what the real assessments do, what ours do differently and why, and how to interpret a score without fooling yourself. ## How an Academy assessment works | Property | Value | | --- | --- | | When | At the end of every Academy course; taking it is optional, passing it is required for the badge | | Length | **10–20 questions** per attempt | | Source | Randomised selection from a **50-question bank** per assessment | | Pass mark | **80%** or higher | | Retakes | Allowed, with a newly randomised selection | | Result | An OpenAI Academy badge issued through Accredible | | Pathway effect | Every course in a pathway must be passed at ≥ 80% for the pathway certificate of completion | Two consequences follow from the bank being small and the selection random. **First, coverage beats cramming.** With 10–20 items drawn from 50, any single narrow topic may or may not appear, but across retakes everything appears. Studying "what is likely on the test" is not a strategy when the sample changes each attempt. **Second, 80% is unforgiving on short forms.** On a 10-item form you can afford **two** wrong answers. On a 15-item form, three. That is why our mocks are longer and our internal target is higher than 80%: a long mock at 85% is a much better predictor of a short assessment at 80% than a short quiz at 80% is. ## How our mock exams differ | | Academy assessment | Our mock exams | | --- | --- | --- | | Items per sitting | 10–20 | 50–60 | | Bank | 50 items per assessment | 100–120 items per track, across two exams, plus 60–150 in-page questions | | Timing | Untimed in practice | Timed, defaulting to the track's recommended duration; 30/60/90-minute drills and untimed mode available | | Scoring | 80% to earn a badge | Independent readiness bands, with the same 80% line called out | | Feedback | Pass or fail | Per-domain breakdown, full correction with distractor analysis, retry-incorrect-only | | Status | Official | **Independent practice. Not official OpenAI questions.** | Our items are written from published learning objectives and product documentation, not from anyone's recollection of a real assessment. That is both an ethical and a practical choice: an item written from the objective tests the skill, whereas an item written from a leak tests memory of a leak. ## Anatomy of a well-built item Every question in this section carries the same structure, whether it lives on a domain page or in a mock bank: | Component | Purpose | | --- | --- | | **Scenario** | A short, realistic work situation – who you are, what you have, what is at stake | | **Stem** | One precise question, usually with a qualifier: *FIRST*, *BEST*, *MOST cost-effective*, *TWO* | | **Four options** | One correct, three plausible-but-weaker; all similar in length and register | | **Correct answer** | Marked, with the reasoning that makes it correct rather than merely acceptable | | **Distractor analysis** | Why each wrong option is wrong – the part that actually teaches | | **Domain and objective** | So a wrong answer maps to a page you can re-read | :::tip[How to read a qualifier] *FIRST* asks for sequence, not quality: several options may be good, only one comes first. *BEST* asks for a trade-off judgment against the stated constraint. *MOST cost-effective* means the cheapest option that still meets the requirement – not the cheapest option. *TWO* means an item is scored all-or-nothing: both right, no extras. ::: ## Readiness bands Our results screen reports an **independent readiness indicator**. It is not an OpenAI score, and it has no standing with OpenAI. | Score | Band | Interpretation | Next action | | --- | --- | --- | --- | | under 70% | **Keep learning** | Gaps are structural, not incidental | Re-read your two weakest domains, redo their in-page questions | | 70–79% | **Building confidence** | You know the material but not reliably | Drill weak domains; retry incorrect items only | | 80–89% | **Assessment ready** | At or above the Academy badge threshold | Take the Academy course, then its assessment | | 90%+ | **Strong readiness** | Comfortable margin on a short randomised form | Move to the next track | Per-domain scores matter more than the total. A 3 of 8 in one domain and 95% overall still means a real chance of failing a short form that happens to sample that domain twice. ## Practice modes and how to use them | Mode | Length | Use it for | | --- | --- | --- | | Domain questions | 10–24 per page | Learning while reading; immediate feedback | | Short drill | 30-minute timed run through a mock | Warm-up, or a focused re-test after revision | | Full mock 1 | 50–60 items, recommended duration | Diagnostic before you study hard | | Full mock 2 | 50–60 harder items, timed | Readiness gate before the real assessment | | Retry incorrect only | Varies | Closing a specific gap; available after any submission | Progress is saved in your browser, so a closed tab offers to resume the attempt. ## Common self-deception patterns The explanation is the highest-value part of the item, and it is worthless if you read it while your answer is still undecided – recognition feels like knowledge. Commit, then expand. If you were unsure, mark the item and revisit it a day later. The second sitting of the same bank measures recall of items, not competence. Use mock 1 for diagnosis and mock 2 for readiness, and put at least one revision cycle between them. Domain weights tell you where the marks are, not where your gaps are. Weight your revision by *weight × your error rate*, not by weight alone. The certification programme is explicitly built around demonstrating skills, with ChatGPT acting as tutor, practice environment and feedback loop. Every track here has a hands-on checklist for that reason. An hour of doing beats three hours of reading about doing. ## Honest limits of this material - Item counts, durations and weights on our mocks are **our design choices**, not published OpenAI parameters. The published parameters are exactly two: 10–20 items from a 50-item bank, and 80% to pass. - Product facts drift. Model names, prices and feature availability were verified in September 2026; the appendix pages tell you where to re-check each one. - Nothing here guarantees a badge, a certificate, or eligibility for any future OpenAI certification.