8.1. Test-Taking Strategy
The GH-300 is about 60 scored questions (plus unscored pretest items) in roughly 100 minutes — a little over 90 seconds per question on average. That's comfortable if you don't get stuck. A few strategies tuned to this exam:
Manage the clock with flag-and-return. If a question isn't yielding in ~90 seconds, pick your best candidate, flag it, and move on. The exam lets you review flagged items; a question you skip now often becomes obvious after later questions jog your memory. Don't let one hard scenario eat five easy ones.
Read scenario questions twice — once for the situation, once for the ask. GH-300 scenarios ("a team needs… what should you do?") often bury the real question in the last line. The setup may include details that are true but irrelevant to what's being asked. Identify the actual decision before scanning answers.
Eliminate on principle, not on recall. When unsure, fall back to the durable principles: responsible answers add oversight (never remove it); accountability stays with the human; the lightest capability that solves the task is usually right; "always/never/guaranteed" claims about AI output are usually wrong. These eliminate distractors even on unfamiliar features.
Watch for the responsibility trap. Any answer that resolves a problem by trusting Copilot more (skip review, disable a filter, assume security) is almost always wrong. The exam consistently rewards validation and human oversight.
Match feature to surface and autonomy. For "which feature" questions, ask two things: which surface is the developer on (IDE / CLI / GitHub.com / Mobile), and how much autonomy does the task need (inline → Chat → Edit → Plan → Agent). The answer usually falls out of those two axes.
Don't over-memorize volatile specifics. Pricing, exact credit figures, and per-editor feature grids change; the exam (frozen at January 2026) tests concepts and judgment far more than today's price list. Spend memory on what features do and who controls them, not dollar amounts.
