Everything below assumes code similarity scoring for grading at scale is already running somewhere and the argument has moved on to who reads the output and what they are allowed to do with it.
Run it on the first assignment of the term, not the last. Finding reuse in week two changes what the student does for the rest of the course; finding it in week twelve changes only what goes on a form. The detection quality is identical in both cases and the outcome is not remotely comparable, which makes timing the highest-leverage variable in the whole exercise and the one least often discussed.
Put the evidence somewhere durable and boring. Screenshots in a chat thread and a spreadsheet on one laptop are how findings get lost between the decision and the review of the decision. A plain directory of case files, one per matter, with the two sources and the date of retrieval, is unglamorous, needs no maintenance, and is the version that still exists when someone asks two years later.
Boilerplate is not noise to be tuned out — it is signal about the assignment, and the right place to remove it is the specification, not a threshold. If forty per cent of every submission is scaffolding the brief supplied, then every pairwise score starts at forty per cent and the interesting variation is compressed into the top half of the range. Subtract the supplied code first and the same detector suddenly discriminates.
A code plagiarism checker earns its place the first time it saves a reviewer from reading two files side by side by hand.
comments (0)