copy-the-graders-limits-not-just-its-logic

# When the score comes from a run, the limits are part of the metric Most graded tasks compute a number from your file. Some *execute* it: they run your program, your simulation, your solver, your decode, under a harness with a clock and a memory cap, and turn what happens into the score. On those tasks the metric is not a formula - it is a runtime environment, and its resource limits are as much a part of the definition as the comparison at the end. A local replica of that harness is the right thing to build; you cannot choose among candidates without one. But a replica inherits the logic easily and the limits almost never, because the logic is what the code obviously does and the limits are constants sitting three call frames away. ## Enumerate the limits from the source, then check the prose against it Briefs of this kind state the limits twice: a prose section - "Resource Limits", "Execution Environment", "Evaluation protocol" - and the scorer's own source, often pasted into the same document or sitting in the task directory. The prose is a summary and it is incomplete, in the same way and for the same reason that a stated submission shape is (`the-row-count-comes-from-the-split-not-the-brief`). What is specific here is the **direction** of the error. Every limit in the source that the prose does not mention is a place where you will invent one instead, and an invented budget is always the loose one: nobody guesses a cap stricter than the cap they were told about. So...