Week 6 Lab: When the rival moves

Build a bounded response function, precommit the falsifier and defend the stop returned when Moss answers

29 March 2027

Two hours, worth 2%, completed and submitted inside the session. The codebase is standalone. It supplies Week 5’s correct position-aware candidate outputs and paired traces; it does not import your earlier code.

What leaves the room is a defended stop against a supplied Moss response function. The record includes the falsifier you committed before the response columns opened, so the standard that changed the call cannot be rewritten after the result arrives.

Before the session#

Install the supplied codebase and run its tests. Read the candidate, paired-outcome and report schemas, including the objective: finishing position first, adjusted time as the tie-break. Lower position is better for SlopU; Moss chooses the reply that makes that result worse.

Run the micro-fixture validator. Hand-select SlopU’s best response in both Moss columns and Moss’s worst reply in both SlopU rows. Bring those four selections and one sentence explaining why a conditional P1 is not guaranteed. The full response columns remain locked.

What is supplied#

Every race value is fictional. The bounded response table is not a reproduction of a research algorithm or an FIA rule.

In the session#

1. Make the objective executable (20 minutes). Validate every paired outcome and implement the lexicographic position-first comparison. Reject a table with a missing action, inconsistent units or a duplicate pair.
2. Select conditional replies (25 minutes). Implement SlopU’s best result for a fixed Moss action and Moss’s worst result for a fixed SlopU action. Match the two-action micro-fixture and explain why P1 depends on Moss’s action.
3. Commit the falsifier, then unlock (20 minutes). Before releasing the response columns, record this rule: retain the fixed-rival stop-13 recommendation unchanged if every supplied Moss reply leaves SlopU in P2 or better; replace it if any reply leaves SlopU in P4; otherwise qualify it as vulnerable but retain it for this bounded task. Preserve the checkpoint ordering in the submission record.
4. Build the response function (35 minutes). For each SlopU stop, emit Moss’s worst reply and its returned position/time result. Then select the SlopU stop with the best returned result. Reproduce the fixed-column undercut and overcut, Moss’s earlier reply and the full response mapping from the unlocked table.
5. Validate and defend (20 minutes). Match the full 3×3 table, make a position-first objective fixture fail, and write four sentences naming the objective, the permitted response set, the returned result and the evidence that would reopen the decision.

The five time boxes total two hours.

What you hand over#

A marker checks:

• the position-first comparison and both conditional selection functions;
• the micro-fixture and full 3×3 response table;
• the pre-reveal falsifier and whether its replacement threshold was crossed;
• all three Moss replies, the returned row results and the response-aware selected stop;
• the rejected objective-error fixture and the four-sentence conditional defence.

This lab practises ordering with a smaller gate: the fixed recommendation is already visible, and only the response columns stay locked. Assignment 1 is stricter — its first stage closes before the race, rival history and recommendation appear. The contest and falsifier habits transfer; the release mechanics are not the same. This fixture reveals none of Assignment 1’s evidence or answer.