Thread by @StarkWareLtd

Scale 👏 rinse 👏 repeat - Recursion!

The SHARP prover switched on Recursion and hyper-scaling just became a reality!

Every proof of many txs can be rolled up with others into a single proof

What is it and how does it lead to better scaling, L3, and other improvements?
🧵

The usual STARK scaling process includes the following:
Off-chain:
1. Validate transactions’ correctness
2. Generate a proof for this validation

On-chain:
3. Verify this proof

So now we add recursion to the proving part and get magic!🎩
>>

Now, Recursive Proving goes like this:
Off-chain:
1. Validate transactions’ correctness
2. Generate a proof for this validation
3. (Add magic here🎩) Validate the proof and generate a proof for this validation. We can repeat this several times

On-chain:
4. Verify the final proof

We do this Recursive Proving process off-chain, so the costs for this recurring process are low.

Using Cairo to generate Recursion also allows greater flexibility for optimizations.

The result of Recursive Proving?

1⃣ We get hyper-scaling, because we compress the transactions into one proof, and then we compress together several proofs.
>>

2⃣ This sets the stage for L3, meaning an environment built on top of L2 that will send its batched transactions to be proved on L2.
>>
medium.com/starkware/frac…

3⃣ We get better latency, because we don’t need to wait for all transactions to arrive at the prover before we start processing them.

We can start processing smaller batches of transactions, in parallel.
>>

4⃣ The complexity of verifying programs that include builtins remains off-chain. This leaves the on-chain verification to handle simple proofs only.
>>

This is a new in production milestone, that hyper-scales #StarkNet Layer 2 Rollup, as well as #StarkEx-powered apps such as @Sorare, @Immutable, @rhinofi.
>>

Recursion makes scaling more efficient in terms of costs, latency and complexity, and it is live on Ethereum Mainnet today!