The Off-chain Compute Map
Making verifiable computation cheap.
✨ TLDR; If you just want to see the map, please scroll down!
It’s been obvious to me that off-chain compute is the next landscape to organize for the past couple of months.
There is so much economic attention going towards restaking and simultaneously various coprocessing solutions are trying to replace entrenched oracle providers.
It’s a fascinating time but one that is very confusing as well.
All the more important to unpack.
Wardley Mapping
To draw the map, we will use Wardley Mapping.
If you’re not familiar with how it works, please check the description I did in the ERC4626 map.
In short, we position elements along the horizontal axis where further to the right means more commoditization and further up means closer to the user (as opposed to infrastructure which would be lower down).
Then we connect value chain components that use each other.
Here it is:
Legend
AVS. Automatically Validated Service (in EigenLayer terms) or equivalent. The service run by a given restaking operator.
Cost. Cost of computation and verification (across all redundant calculations).
Latency. Response time from onchain request for computation.
Intel SGX. Instruction codes that allow applications to run in isolation without trusting the Operating System or other peripheral software.
MPC. Multi-party computation. The ability for multiple parties to collectively agree on a computation output without revealing sensitive inputs from each party.
Privacy. The ability to hide computation inputs.
Pull oracle. Modern (Pyth-style) oracles that are submitted upon onchain or off-chain request from the consumer and are otherwise stale. VRF (Verifiable Random Function) oracles are also pull oracles.
Push oracle. Traditional (Chainlink-style) oracles that periodically update their outputs. The data is “pushed” onchain via a transaction.
Restaking. A protocol that uses slashable commitments of staked Ether or similar assets to bet on the integrity of an operated service in exchange for rewards.
TEE (Trusted Execution Environment) Coprocessor. A service that relies on hardware secure enclave to guarantee computational integrity.
Trustlessness. The ability for a single party or a coalition of parties to manipulate the results of computations.
ZK Coprocessor. A prover service that can submit verifiable computations onchain.
ZK Prover. Software that can generate Zero Knowledge proofs.
ZKML. Machine learning services that can use ZK tech to protect user data inputs and verify what model has been used.
Let me know what you think and reply with any feedback.