Run Tests Checks What Submit Checks

Run Tests on an algorithmic assessment question used to grade only the public cases. The hidden cases were checked for the first time when the candidate finished the assessment. A candidate could see every public case pass, submit feeling sure of the answer, and still lose points on hidden cases they had never run. A check that covers only part of the grading gives candidates the wrong idea of what they are being graded on.

One count for every case

Run Tests now runs every case, hidden ones included, through the same grader that scores the final submission. The results panel shows a single pass count across all of them. Hidden cases add to that count and reveal nothing else. Their inputs and expected outputs stay on the server, and the only cases you can open and read are still the public ones.

A full pass now covers every case the final submission is checked against. When every public case passes but the count comes up short, you know a case you cannot see is failing, and you learn it while there is still time to look for the input you did not handle.

A crash now says it crashed

If a program crashed before printing anything, its row used to say Wrong Answer. Opening the row showed No output, and the error message sat at the very bottom of it. A candidate reading that would go looking for a flaw in logic that never ran.

A crashed test is now labeled Runtime Error, with its traceback at the top of the results panel, so an undefined variable on the first line shows up as exactly that. Wrong Answer now means the program finished cleanly and printed the wrong thing. This works the same in the assessment workspace with AI help and in the one without.

One name, and no endless spinner

The grading button is now called Run Tests in both assessment workspaces. It used to be called Test. In the workspace without AI help, it sat beside a Run button that runs your code once with no input, so it was hard to tell which one gave you a pass count.

Results also come back sooner, because the server no longer saves the whole session record before it runs a single case. That save used to keep candidates waiting on bookkeeping they had no control over, even on a small question. Now the tests run first and the results go straight back, while the save finishes in the background.

If the server never answers at all, the panel no longer spins forever. The elapsed time counts up while grading runs, and a run that gets no reply ends with a note that the tests timed out, your code was not changed, and you can press Run Tests to try again.

Less on screen that does not apply

Algorithmic questions no longer offer a Preview pane. There was nothing to preview, so choosing it gave half the workspace to a note saying so, on the question type where the editor and the test results need the room most.

Checkpoints are gone from assessments too. A checkpoint rolls the code and the chat back to an earlier point. In an assessment, the attempt itself is the record a reviewer reads, so rewinding it would undo work they are meant to see. Checkpoints stay in Builder, Learn labs, and Classroom, where rolling back is just an undo.