I was also researching SGX for the purpose of a permissionless blockchain design but gave up due to issues with security, particularly side-channel attacks and others that will cause an issue undoubtedly when money is on the line. One interesting model I found that would potentially make SGX go was the antarctica model as described here: https://www.idc.ac.il/en/schools/cs/research/Documents/jackson-msc-thesis.pdf

which relies on timing analysis to skip the need for RA back to Intel and rely on metrics to understand if an enclave is running code it shouldn’t be. But for this a specific instruction for this timing needs to be included in the boot sequence of the device for it to work and thus again relies on the authoritative decision of Intel to make it happen which doesn’t seem like they want. Peter Todd has been pretty harsh aswell on Intel as being dis-honest about intent of design and not designing it with trustlessness in mind (they are a for-profit company afterall).