
Thirty Spokes is an experimental Bittensor competition for cost-aware model-routing heads whose benchmark runs are intended to be verified through Intel TDX attestation.
Miners build a small routing model that chooses where to enter an owner-defined ladder of remote frontier models, run benchmark tasks in a confidential VM, and submit hardware-attested proof artifacts for validator verification.
The competition and attestation stack are implemented, but the official documentation says a reproducible measured image is still required before secure enforce-mode mining is complete. The planned OpenAI-compatible multi-model gateway is explicitly not live and has not been built yet.
Bittensor miners and routing researchers willing to operate Intel TDX infrastructure and train cost-aware routing heads over the subnet's pinned model pool.
There is no end-user gateway pricing: the official economics documentation says the gateway, pricing and revenue path are not yet built. Miners pay for their own confidential VM and model calls.
The current design moves benchmark execution into miner-controlled TDX environments and lets validators verify hardware-attested proofs instead of paying to run model inference themselves.