Runtime Footprint

CoreTex has two runtime profiles:

Profile What it does Hardware expectation
Client/verifier Decode state, verify bundle/root bindings, replay patches, inspect corpus artifacts Ordinary CPU machine, slower is acceptable
Coordinator evaluator Serve API, run bounded CoreTex evaluation, sign eligible receipts AWS CPU server sized for expected miner throughput
Calibration host Run long corpus generation, determinism checks, Qwen/BGE calibration, exhaustive E2E Larger CPU box; memory bandwidth matters more than parallel workers

The canonical scoring path is CPU-only. GPU execution is not part of the deterministic launch path. The client bundle and model file hashes are what make results reproducible across hosts.