# Exam Day Checklist

A practical, timed checklist for the day of any Claude certification — logistics, OnVUE setup, in-exam strategy, distractor triage and post-exam steps.

import { Steps } from '@prosefly/astro-components';

All four exams are 120 minutes, scaled 100–1000, pass 720, multiple-choice and multiple-response, delivered via Pearson VUE OnVUE or a test centre, **no guessing penalty**. This page is the operational runbook for the day.

:::tip[No guessing penalty]
Because there is no penalty for a wrong answer, **never leave an item blank**. Flag-and-move, then return; make a best guess on anything still open before time expires.
:::

## The evening before

- [ ] OnVUE system test passed on the **same** machine and network you will use.
- [ ] Government photo ID ready; name matches Pearson VUE registration exactly.
- [ ] Quiet, private room booked; desk cleared of notes, phones, extra monitors.
- [ ] Webcam, mic, stable internet, power adapter confirmed.
- [ ] VPN / extensions that could block the proctor disabled.
- [ ] Start time and time zone double-checked.
- [ ] Current Exam Guide re-downloaded; weights unchanged.
- [ ] Sleep — no new studying.

## Thirty minutes before

<Steps>
1. Close all applications except the OnVUE browser.
2. Launch check-in early; complete the room scan and ID capture calmly.
3. Have water in a clear container if permitted; use the restroom first.
4. Do a 60-second breathing reset — nerves cost more points than any single fact.
5. Recall your first-move heuristics (below) so they are loaded before question 1.
</Steps>

## Time budget (120 minutes)

| Exam | Items | ~ Seconds/item | Reserve for review |
| --- | --- | --- | --- |
| CCAO-F | 60 | ~110 | ~10 min |
| CCDV-F | 53 | ~125 | ~10 min |
| CCAR-F | 60 | ~110 | ~10 min |
| CCAR-P | 63 | ~105 | ~8 min |

Rule of thumb: at the halfway clock mark you should be near the halfway item. If you are behind, speed up on recall items and flag the analysis-heavy ones for the reserve.

## In-exam strategy

<Steps>
1. **Read the qualifier first** — BEST, FIRST, MOST cost-effective, NOT, TWO. It changes the answer. Circle it mentally.
2. **Find the binding constraint** — cost, latency, compliance, reliability, irreversibility. The constraint usually decides among plausible options.
3. **Eliminate distractor patterns** — cross out constraint-blind, over-engineered, prompt-as-enforcement, self-report, silent-failure, recall-only, aggregate-metric, sentiment-as-complexity options.
4. **For "Select TWO"** — each choice must be independently correct; beware one-right-plus-one-tempting. You need both.
5. **Flag and move** — never burn 4 minutes on one item; flag it, answer your best guess, return in the reserve.
6. **Answer everything** — no blank items (no guessing penalty).
7. **Trust the principle over the wording** — a familiar-sounding phrase is often a distractor; match the underlying decision rule.
</Steps>

## First-move heuristics (load these before you start)

| Stem cue | First move |
| --- | --- |
| "How does the loop know it's done?" | `stop_reason: end_turn` |
| "Must never / under no circumstances" | Deterministic hook / permission |
| "When should it escalate?" | Explicit request or beyond-capability; not sentiment/self-report |
| "Overall accuracy is X%" | Demand per-segment |
| "Confident but wrong after a doc update" | Suspect retrieval/indexing first |
| "Forced tool call" + Fable 5.1 | 400 — use `auto`/`strict`/structured output |
| "MOST cost-effective, latency doesn't matter" | Batch + caching + smallest sufficient model |
| "Reused across apps/agents" | MCP server |
| "Irreversible / regulated / external" | Human-in-the-loop gate |
| "Are you sure?" same chat | Fresh session / different model |
| "Too many tools" | 4–8; tool search + `defer_loading` |
| "Multi-file / architectural change" (Claude Code) | Plan mode |

## Distractor triage table

| Pattern | Tell-tale wording | Why it's wrong |
| --- | --- | --- |
| Constraint-blind | Ignores a stated cost/latency/compliance limit | Fails the actual requirement |
| Over-engineered | Multi-agent where a chain suffices | Complexity without benefit |
| Prompt-as-enforcement | "Add a rule to the system prompt" for a hard rule | Prompts guide, not enforce |
| Self-report reliance | Routes on "the model's confidence" | Unreliable signal |
| Silent failure | Empty result / treats truncation as done | Hides errors |
| Recall-only | Restates a fact without applying it | Not the decision asked for |
| Aggregate metric | "Overall accuracy is fine" | Masks per-segment failure |
| Sentiment-as-complexity | Escalate because the user is angry | Tone ≠ difficulty |

## If something goes wrong

- **Proctor disconnect / tech issue** — stay calm; the platform usually resumes and preserves your progress. Follow the proctor's chat instructions; contact Pearson VUE support if it does not recover.
- **Blanking on a topic** — flag, guess, move on; do not let one item cost you five.
- **Running out of time** — in the last two minutes, answer every remaining item (no blanks).

## After the exam

<Steps>
1. Note your provisional result and, if provided, the per-domain breakdown.
2. **If passed** — claim the Credly badge; set a renewal reminder at month 10; the credential is valid 12 months with free on-time renewal.
3. **If not** — read the per-domain scores; use the retake strategy (target the weakest domain, respect the 14-day minimum, verify with a timed practice exam before rebooking). Max 4 attempts per rolling year.
4. Log what surprised you while it is fresh — it sharpens your notes for a retake or the next certification in the track.
</Steps>

## Key takeaways

- 120 min, pass 720/1000, no guessing penalty — **answer every item**.
- Read the qualifier (BEST/FIRST/MOST/NOT/TWO) and find the binding constraint before choosing.
- Eliminate by distractor pattern; match the underlying decision rule, not familiar wording.
- Flag-and-move; keep a review reserve; never blank an item.
- Sort logistics (OnVUE, ID, environment) the night before so exam day is calm.