Calibrated quizzes
Ten short quizzes where you declare how sure you are, and being wrong while confident costs more than being wrong while hedging.
Due: 13 May 2027 · Weight: 15%
Ten quizzes, one in each of weeks 2 to 11, ten to fifteen minutes each. Your best eight count, so two can be missed without a reason or a form.
They are marked with certainty-based marking, which is not a gimmick and not invented for this course. Gardner-Medwin developed it at UCL, where it has run across seventeen summative medical examinations.
How the scoring works#
With each answer you declare a certainty level. What the answer is worth depends on that declaration.
| Declared certainty | If correct | If wrong |
|---|---|---|
| Low | +1 | 0 |
| Mid | +2 | −2 |
| High | +3 | −6 |
So a confident error costs you six marks, and a hedged error costs you nothing.
Why you cannot game it#
This is a proper scoring rule. That is a technical property, and it means reporting your actual confidence maximises your expected score — there is no clever strategy that beats honesty. The thresholds work out at roughly 67% for mid and 80% for high: below those, declaring high is a bad bet even though it pays more when you are right.
Which removes the dilemma that ordinary negative marking creates, where you have to decide whether answering at all is worth the risk. Here the question is never whether to answer. It is only how much to stake.
What to expect in the first fortnight#
Your score will look bad, and this is normal rather than a sign that you are doing badly.
Students who answer 80% of questions correctly typically collect around half the available marks. The gap between those two numbers is not a scoring quirk. It’s the difference between how right you were and how right you thought you were — which is the subject of week 12, arriving early and in your own results.
Most people misuse the scale for two weeks and then it clicks. Tuyet runs a practice quiz in week 1 that does not count, and it is worth doing.
A note on reading#
Certainty-based marking puts a premium on having reasons to be sure. Answers memorised from a summary tend to attract high confidence without the reasons that would justify it, and the scoring punishes exactly that. Reading the papers is the cheaper path.
The rules that apply to this task#
Extensions, late penalties, academic integrity and what you may use a language model for are on the policies page, which governs all four assessments. Nothing on this page overrides it.
The spec
The brief poses the problem and leaves room for a response. The spec is the fixed part: what a marker checks. Some lines can be checked mechanically; the rest call for judgement.
- at least eight of the ten quizzes are attempted
- each answer carries a declared certainty level
- the best strategy available to you is to report your actual confidence