00:05:44 We could have each block mined specify a voting key. Once x% of blocks over some n block period include it, where x >= 51, the system activates. Each mined block is one vote usable in a BFT consensus protocol. Votes expire after 24h, and the protocol stops if blocks with voting keys drop below 51%. 00:05:45 Nodes which opt-in only follow descendants of the finalized chain. If a hash power spike by an adversary occurs, causing a reorg, so long as the participating honest nodes maintain 51% of hash power *over time*, they'll reorg the chain back. If an adversary can maintain 51% for an extended period of time, that's its own issue and worth halting over. 00:05:47 The main issue is there's no cost to equivocate and finalize two distinct blocks. You'd need to have some time delay to probabilistically detect equivocations and halt upon it. 00:06:25 Or we hardfork in slash mechanics and build a proper consensus layer. 00:07:51 I only through out the off-chain ideas because they're likely better received. 00:07:53 *threw 01:11:04 I like the light design, but PoS ain't PoS without slashing, and definitely not without stakes :) (I'm still not convinced by Ouroboros) 01:18:58 It doesn't have to be "PoS" if it's argued a valid finality layer. 07:02:22 What is slashing in this context ? 07:11:08 I think it's something to do with punishing misbehaving PoS nodes 09:06:20 Slashing refers to burning the stake of misbehaving nodes who put up stake as evidence they wouldn't misbhehave. 09:08:34 I think if we do want to discuss a slash-less finality layer, which can be done entirely off-chain, it'd best to posit it as a centralized entity issuing checkpoints which 51% of known miners agree to follow. The entire security commentary can be done against the black boxed centralized entity, with a way to decentralize it (last 100 miners) discussed after. 09:15:51 If you assume the network is synchronous with a latency of 2 minutes (which honestly is probably the case without a full net split/eclipse attack), then you do get a solution where at worst, finality halts (halting services requiring finality) and only if network synchrony is broken, a panic. 09:16:48 I wonder if there's existing research on the time to detect equivocations somewhere. I'd assume so. 09:18:49 I'll probably type up sketches on this in a MRL issue and try to nerdsnipe Rucknium.