We Analyzed Our Own 780 SQL Questions. Here's What the Data Shows.

SQLMentor's five certification practice exams and general quiz add up to 780 original multiple-choice questions. We ran the whole bank through the same topic-tagging logic that already powers the "topics to revisit" feature on exam results — and found something we didn't expect in our own answer key. (Update: we found it, disclosed it, and fixed it — see below.)

The corpus

Everything below is computed directly from the JSON files that power SQLMentor's practice exams and quiz — content/cert-exams*.json and content/quiz.json — not sampled or estimated.

Every question in every set has exactly four answer options — no five-option trick questions, no "all of the above." That consistency is what made the next finding possible to measure cleanly.

Finding 1: our own answer key leans hard toward option B

If four answer options were placed in a random order, each letter would be correct roughly 25% of the time. We counted the actual distribution across all 720 certification questions:

Option B is correct nearly two and a half times as often as chance predicts, and option D is correct in only 1 out of every 27 questions. This isn't a fluke in one exam — it shows up, to varying degrees, in every one of the five sets:

What this means in practice: a test-taker who guessed "B" on every question they didn't know would score noticeably better than random chance on four of these five exams — which defeats the point of a practice exam. This was a real bias in our own question-authoring process (correct answers were written in place rather than the option order being randomized afterward), not a property of the real certification exams.

All four options now land within a couple of points of the 25% you'd expect from chance — the largest per-exam skew remaining is DP-300 at 32%/20%/27%/22% (a smaller question set, 60 questions, so more sampling noise is expected). Question text, explanations, and which answer is correct are all unchanged — only the on-screen order of the four options was reshuffled. We're leaving the original numbers above rather than editing them out, because a corrected line item is worth less than a visible before/after.

Finding 2: what the Oracle certification questions actually emphasize

We tagged all 600 questions across the three Oracle exams (1Z0-071, 1Z0-082, 1Z0-149) using the same keyword-bucket rules already live in production for the "topics to revisit" feature on cert results pages — reused here across the whole bank rather than just one person's wrong answers. About 9% of questions didn't match any bucket clearly enough to tag and are excluded rather than force-fit.

The concentration in procedures/functions makes sense given the bank spans a pure-SQL exam (1Z0-071), a DBA exam (1Z0-082), and a PL/SQL exam (1Z0-149) — PL/SQL constructs show up across all three whenever a question touches stored logic. If you're studying for 1Z0-071 or 1Z0-082 specifically and want the SQL-only breakdown, the 1Z0-071 exam guide and 1Z0-082 exam guide break the official syllabus down by domain.

Finding 3: the general quiz skews toward fundamentals — on purpose

Unlike the certification exams, the 60-question general quiz is hand-balanced across three explicit difficulty tiers — 20 easy, 20 medium, 20 hard, exactly even. Its topic spread is intentionally front-loaded toward core SQL:

That's deliberate: the quiz is meant as a general SQL-literacy check, not certification prep, so fundamentals (filtering, aggregates, basic SQL) get proportionally more coverage than in the Oracle-specific exams above.

Methodology, in full

• Data source: the live JSON files behind SQLMentor's practice exams and quiz (content/cert-exams*.json, content/quiz.json), read directly — no sampling.
• Topic tagging for certification questions: a first-match keyword/regex ruleset already in production (used to power "topics to revisit" on exam results screens), applied here across the entire question bank instead of one user's incorrect answers. It's a heuristic, not a hand-labeled ground truth — about 9% of questions matched no rule and are reported as uncategorized rather than guessed at.
• Topic tagging for the quiz: quiz questions carry an explicit, hand-assigned topic field in the source data, so that breakdown is exact, not heuristic.
• Answer-letter distribution: a direct count of the stored answer index per question, converted to a letter (A–D). No interpretation involved — this one is exact.