Every answer is checked server-side against the verified key. The score you see is exactly correct ÷ total — nothing weighted, nothing massaged.
The diagnostic covers all twelve areas real tests draw from. Each question carries named wrong-answer patterns (the specific mistake each wrong option encodes), so when you miss one we can tell you the area — and on free questions, the exact error — behind it.
Real percentiles need real test-takers. Until our own attempt volume is large enough, your percentile is computed against a seeded baseline cohort (n=200) — a fixed, documented prior — and it is always labelled as such next to the number. As real attempts accrue, the baseline switches to actual test-taker data and the label updates with the live n. We never display a percentile against a population that doesn’t exist.
Each question’s difficulty tier (1–5) is computed from its structure — steps, lookups, number awkwardness, time pressure — and recalibrated from real solve data. It is never set by hand.