Final project: the defence
Written knowing the result — the calls justified against the distribution they faced rather than the result they got
Due: 18 June 2027 · Weight: 20%
The brief#
You know how the race went. Defend the calls anyway.
The defence is written after the result is in, and that is the point of it. Anybody can justify a call that worked. This document asks you to justify each call against the distribution it faced rather than the result it got — including the call that lost, and including the one that won for a reason you cannot take credit for.
One draw is one draw. Your race was a single sample from the range of races your calls could have produced, and a sample of one says almost nothing about the decision that met it. That is why the finishing position carries no marks and this document carries a fifth of the course.
What it contains#
One document, not two. The account of how your instrument works is the evidence your defence stands on rather than a section in front of it: each call you defend points into the model that recommended it, and a call with nothing to point at has no defence.
It covers, in whatever order serves the argument:
- each call worth arguing about, against what your instrument was saying at that timestamp and what the race could still have done next
- where your instrument was wrong in this race, and what being wrong there cost you
- whether each supported departure matched an actionable limit you declared at submission and the contemporaneous evidence recorded when its observable condition applied; unsupported departures are named as such
- the objective your calls served, and whose interest it serves
A general limitation can be honest and useful here without having been recognisable during the race. Only an actionable limit could license a live departure, and only when the log also holds contemporaneous evidence that its clear, observable condition applied.
Where your instrument was wrong is the part students leave out. An instrument you cannot fault after racing it is an instrument you were not reading.
What it is not#
It is not a reflection. Nothing here is marked on what you learned about yourself, and a paragraph on how the pressure felt displaces a paragraph that argues.
It is not a defence of the result either. If your race went well, the document is the same document: the same calls judged against the same distributions, and the same admission of where the instrument was wrong. Honest, rigorous analysis of an unsupported departure can earn defence marks. It cannot retrospectively justify that departure or restore race conduct marks.
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.
- one document, in which the account of the instrument is the evidence rather than a section before it
- every call it defends cites the recommendation standing at that call's timestamp in your log
- every supported departure cites its submitted actionable limit and the contemporaneous trigger evidence in the log; unsupported departures are analysed as unsupported
- it names at least one thing the instrument got wrong in the race, and what that cost
- it states the objective the calls served
How it is marked
| Criterion | Weight |
|---|---|
| The call against the distribution it faced rather than the result it got | 40% |
| Departure evidence — each supported departure traced to a submitted actionable limit and its contemporaneous trigger, or analysed honestly as unsupported | 25% |
| What the instrument got wrong in this race, and what it cost | 20% |
| The objective, and whose interest it served | 15% |
The heaviest criterion is the one the obvious document fails. A call is defended against what your instrument said was likely at that moment and what the alternatives were worth, not against what happened next. An argument that starts from the result and works backwards scores low even when the call was good, because it defends the outcome instead of the decision.
Convincing evidence is your own instrument quoted against itself: the recommendation standing at the timestamp, what it was weighing, and why your call was the one that reading supported. A supported departure needs both the actionable limit you submitted and the evidence recorded at the call showing that its observable condition applied. An unsupported departure admitted and analysed can score here, but cannot regain race conduct marks or become justified afterwards. For what the instrument got wrong, a marker is looking for one specific failure in this race and the cost attached to it, rather than a general admission of imperfection.