possible bug?

This morning I tried to solve acm-shortcut and made 5 attempts at 8:46, 9:03, 9:26, 10:06 and 10:15. My first two attempts were wrong, my third one was (probably) correct but too slow. Then I completely re-wrote my code, but still got TLE on my fourth attempt. I carefully looked at my code, but could see no reason for it to be too slow, so in my last attempt I resubmitted excactly the same code. Result: Accepted with runtimes an order of magnitude faster then on my fourth attempt.
Can it be that on my fourth attempt the grader used the (cashed) compiled code from my third attempt? The runtimes are comparible. Strange...

Thanks for this.

You pretty much guessed correctly:
Your 4th submission was actually graded, but the code waiting for the queue got the results from the 3rd submission, This happened because the 3rd submission wasn't cleared from the cache (queue results).

Should be fixed now.