You asked. Here's what came up in the replies this week. "Can you prove a specific model ran on a specific input while keeping the prompt private?" Yes. The prompt goes in as a private input, only its commitment is published, and the proof binds that commitment to the run. Nobody sees the prompt. Everybody can check it was the one used. "Does that proof say anything about answer quality?" No, and this is the part worth remembering. The proof covers faithful execution: this model, this input, this output. A bad prompt still produces a perfectly valid proof of a bad answer. "If blockchain is trustless, why is it so hard to track an address that stole funds?" Tracking it isn't hard — every hop is public. Linking it to a person is. Trustless means nobody has to be trusted to validate transactions. It never meant anonymous, and it never meant reversible. Keep them coming. Every question in the replies gets read💜