Assignment 1: the audit

The audit — a recommendation the team bought, tested against the race it was written for and the rival who responds to it

Due: 30 April 2027 · Weight: 25%

The brief#

Slop University Racing did not write the strategy for this race. It bought one. Your audit says whether the team should run it.

The recommendation arrives from outside the team. It names a strategy — which laps to pit on, and which tyre compound goes on the car at each stop — with a written justification and a claimed margin at the flag. It is confident. Nobody at the team has checked it.

You are not asked for a better strategy. You are asked what the team should do with this one, in a form that survives being wrong: a verdict with its conditions attached. Retain unless this; reject if that. Retain, qualify and reject are all available to you, and which one you write is the assignment. A verdict that names no condition is one you cannot be held to.

The objective is in the brief from the first day, fixed and stated. This is not an assignment about what Slop University Racing wants from the weekend; it is an assignment about whether the strategy it bought delivers it.

The archive this race comes from is synthetic. The timing data was generated, so the quantities you estimate have true values and your error bars can be checked against them rather than admired.

What you submit#

Submission is in two stages, and the first closes before the second opens.

Stage one — the falsifier. Before you see the race, the rival’s history or the recommendation, you submit one page fixing the standard the recommendation will have to meet: what evidence would make you retain a bought strategy unchanged, and what evidence would make you reject one outright. Submitting that page unlocks the rest of the material. It is timestamped on submission, so written before looking is a property of the record rather than a claim you make about yourself.

Stage two — the audit. A codebase and a short document.

The codebase is new. It combines what the labs of weeks 2 to 6 left you and extends it, and it is deterministic and position-aware: no hazards and no sampling, because nothing in this assignment needs them. Run it twice and it says the same thing twice.

Classification in this sport is by finishing position, not by elapsed time — track position is a resource you may not be able to overtake your way out of.

The document has three parts, in this order.

• The verdict, in one sentence, with its conditions attached.
• The evidence beneath it: what you measured and how uncertain the measurements are, whether the strategy is legal under the regulations, what the strategy does to the race in finishing position rather than in elapsed time alone, and what the rival does in response.
• The falsifier, reproduced exactly as you submitted it at stage one, and not as you would write it now.

The standard#

A verdict is marked at the level of the evidence beneath it, not the confidence above it.

Correct recomputation contributes evidence to an audit. It is not sufficient on its own: the criteria also ask for model fitness, for a rival treated as responding rather than as a fixture, and for a verdict with its conditions stated.