[Super-Finality: High-Safety Confirmation for Casper - Consensus - Ethereum Research](https://ethresear.ch/t/super-finality-high-safety-confirmation-for-casper/16429) Can boost safety resilience of Casper beyond the 33% stake threshold provided by finality confirmation rule Implemented client-side with no changes to protocol itself ## Overview *Safety* means that once a transaction is considered confirmed, teh transaction will remain in the chain forever in the same position Cannot be removed, reordered, or altered [[Gasper]] is a combo of [[LMD-GHOST]] and [[Casper]] FFG Users who want low latency and can tolerate reorgs follow the tip of the LMD GHOST chain or use a confirmation rule for LMD GHOST Users who require safety adopt the confirmation rule of Casper With 33%+ stake held by Lido alone, there's a real risk ## Super-finality rule for Casper Client can choose the level of safety it desires and apply the rule on a per-transaction basis Higher safety comes at the cost of higher latency Can be as high as 99%, meaning xx% of staked ETH has to confirm and finalize the transaction before it is considered finalized![[Pasted image 20230921135819.png]] If participation for validators dips too low, users may see chain stalls **Super-finality rule:** Super-finalize block A (and its ancestors) iff: 1. We see a block B whose state commits to A or its descendants as the finalized checkpoint 2. We see votes in favor of B (FFG votes with target B or descendant B) from validators totaling q fraction of stake![[Pasted image 20230921140058.png]] ### Higher Confirmation Quorum for Higher Safety Resilience ![[Pasted image 20230921140224.png]] ### A Validator's Vote Reveals what Finalization it is Locked on Once a validaator sees a block as finalized with 2/3 votes, it will never vote for any blocks that are inconsistent with that If a validator votes for B which commits to A as the finalized checkpoint, then the validator is saying it confirms the finalization of A. It'll never vote for a block that conflicts with A Thus, when a client sees validators with q fraction of stake voting for B, super-finalizing A, it knows that super-finalizing C would require at least 2q-1 fraction of stake to be malicious Cost of a high safety like 90% is that if 10% of the total stake goes offline, then no block gets super-finalized and the client loses liveness