I was scanning the OpenGradient ($OPG , #OPG ) Python SDK reference, expecting verification to be the headline feature the project leads with — @OpenGradient frames its network as designed from the ground up for verifiable AI inference. The LLM completion function takes an inference_mode argument with two options, VANILLA or TEE, and the default sitting in the function signature is VANILLA. The docs describe that default plainly: standard execution that returns on-chain results but provides no hardware attestation. No enclave, no cryptographic proof of execution, just an answer and a payment hash. Getting the guarantee the vision is actually built around means a developer has to explicitly pass TEE as the argument; otherwise every call runs in the mode the documentation itself describes as skipping that step. It's one keyword in a function call, easy to leave untouched while copying a quickstart example, and it quietly inverts the order I expected — verification reads like the foundation, but ships as something you opt into after the fact. I keep wondering what share of inference traffic on the network is actually running under the unattested default, and whether that number shows up anywhere a user could check it.