Create knowledge quiz — check learning progress

A knowledge quiz with the same ten questions for all participants is worthless after the first round — the answers circulate in internal Slack within two days. A question pool of thirty to fifty questions from which ten are randomly drawn keeps the quiz robust against sharing and allows repetitions without identity detection.

For the build: every question gets a difficulty level (easy, medium, hard) and a topic category. At quiz start the calculation engine randomly draws a mix of categories and difficulties — e.g. "three easy, five medium, two hard, evenly distributed across four topics". This way all participants get a comparable quiz but not an identical one.

Question quality matters. Unambiguously phrased questions with a clearly correct answer are the goal — questions where "it depends" would be a plausible answer belong in a discussion thread, not the quiz. Before going live all questions should be reviewed by two to three domain experts. An ambiguous question distorts the score distribution and undermines trust in the certificate.

A quiz without time limit is actually a research test — anyone with Google open answers everything correctly. A too tight limit, on the other hand, tests typing speed instead of knowledge. The balance is typically 30 to 60 seconds per question, depending on answer type. Multiple-choice needs less time than free text.

Technically this can be solved with a client-side countdown plus server-side validation. The frontend timer shows the participant the remaining time, on expiry the question is auto-submitted (with the so-far selected answer or empty). Server-side the timestamp per answer is stored — answers arriving after expiry are ignored. This prevents manipulated browser time.

Communicate the time limit in advance. "30 seconds per question, the quiz takes about 5 minutes total" provides expectation clarity. Reduce stress through clear visualization — a slowly draining progress bar is less intimidating than a large red seconds display. For accessible quizzes the time limit should be optionally extendable, e.g. via configuration "Need more time? Click here" — acceptance rises noticeably.

A bare score "8 out of 10" tells the participant little. Was that good, average, bad? Score bands give the result meaning: 0-4 points "repetition recommended", 5-7 "solid foundation", 8-9 "advanced", 10 "expert". This grouping must be defined before the quiz and ideally based on an empirical distribution — not gut feeling.

For training certificates a special logic applies: there is a clear pass threshold (e.g. 70 percent correct), often regulatorily required in compliance contexts. Anyone below has not passed — no "almost passed". This hardness is needed because otherwise the threshold becomes arbitrary. Conditional logic in the calculation engine implements this automatically: at score X result page A opens, otherwise result page B.

For learning quizzes (without pass threshold) softer score bands are better. Here it is about self-assessment, not filtering. A result page with personalized hints ("You were strong in topic X, weak in topic Y, here are three learning resources") is more valuable than a pure score display. AI can pull matching recommendations per topic weakness from a pool so the result page feels contextual.

A passed quiz without a visible certificate is wasted value — especially in compliance contexts or professional development. A PDF certificate with name, quiz title, date, score and a unique verification ID is the expected standard. This PDF is generated automatically right after passing and delivered by email plus offered for download.

Technically this can be implemented via server-side PDF generation — for example with an HTML template that fills in variables and a conversion via headless Chrome or a PDF library. The template should have room for a logo, a signature (or digital signature) and a QR code for verification. The QR code points to a public URL where the certificate authenticity can be checked — important against forgeries.

For legally binding certificates (e.g. for TÜV-relevant trainings) a pure PDF is not enough — here you need a qualified electronic signature (QES) under eIDAS. For internal trainings or marketing quizzes the standard PDF is fully sufficient. The verification page should show at least name, date, score and an activity-log entry of the quiz participation — this effectively deters manipulation.

May a participant repeat the quiz? This question has more impact than it initially appears. For learning quizzes repetition makes sense — learning works through repetition. For compliance tests it is problematic — anyone trying three times eventually guesses the right answer without understanding it. The configuration must be deliberately chosen.

Three typical models: unlimited repetition (for learning quizzes, score stored as best attempt). Limited repetition (e.g. three attempts per day/month — prevents brute force). One-time per person (for official certificate tests, identification via email or login). With limited repetition the question pool should be large enough that no question set repeats per attempt — otherwise repetition becomes a memorization test.

For compliance repetitions after longer pause a cooldown logic is sensible: on failure the next attempt may only happen after 24 hours or a week. This forces learning instead of guessing. The activity-log entry records every attempt with timestamp, score and drawn questions — important for audit trails in regulated industries. A webhook after passing can automatically update the learning status in the LMS so the employee does not have to enter it manually.