L1 和 L2 的边界正在消失,但方向和直觉相反:不是 L2 越来越像 L1,而是 L1 的执行层自己会变成一个 rollup。 最近以太坊社区讨论的两个方向,单独看都不新,叠加之后才产生了新的东西。 Native rollup 让 L1 直接验证 L2 的状态转移。L2 不再需要维护自己的证明系统和验证合约,代价是执行环境必须与 L1 完全一致,回报是 L1 后续的任何升级都自动继承。 Single slot finality 则把 L1 的 finality 从大约 15 分钟压缩到单个 slot。 两者解决的并非同一个问题。今天的 rollup 名义上已经结算到 L1,但实际被信任的是它自己的验证合约以及背后的升级权限,native rollup 消除的正是这层自定义的信任假设。延迟方面,一笔 L2 交易需要先等待自身的证明生成与批次提交,再等待 L1 的 finality。Native rollup 让验证在 L1 区块内同步完成,直接消除了前一项;SSF 则压缩后一项。缺少任何一个,L2 的结算都会被剩下的那项限制住。 两者同时到位之后, L2 的区块空间与 L1 已经相差无几,若再叠加 based sequencer,则趋于完全一致。 代价同样明确:交出定制执行环境的权利,也就交出了在这个维度上做差异化的空间,L2 的角色从"另一条链"回落为"L1 的扩展"。对已经拥有成熟技术栈的头部 L2 而言,这是一个真实的取舍。 值得注意的是,这套架构里所证明的是 L1 而不是 L2。L1 由区块构建者为整个区块生成证明、节点验证证明,而 L2 的执行本就发生在 L1 区块内部,天然被这个证明覆盖。L2 自始至终不需要生成任何证明,也不需要为 L1 更换证明系统做任何适配。 这也让最后一件事变得清晰。这几年 rollup-centric 的路线与对 ZK 的巨额投入,随着路线图转向 L1 扩容,很容易被解读为走了弯路。事实恰恰相反:native rollup 的扩容能力建立在 L1 的证明化之上,而 L1 的证明化,依赖的正是这几年被 L2 赛道反复验证过的那套证明技术。rollup-centric 并没有被放弃,它的产出成了下一个阶段的输入。
Riffing on this, many analogies collapse when you consider that L1 is likely to eventually turn into a rollup/L2 of itself. So it's not the fundamental nature of a rollup to not be "value accretive" to ETH or Ethereum. And the right lens to think about it is state, and one's ability to host the creation of valuable state and to manage access to this state. This is also why better interop is valuable. The more external domains/sequencers have the ability to compose with L1 state, eg, leveraging its liquidity, the more value accrues to it, vs "islands of state" bootstrapping their own economies without Ethereum's added value. With this in mind, it seems like it should follow that based sequencers and sync composability are then the way to go, to make external state as tightly coupled and composable with L1 state as possible. This is only true if systems built on these principles retain the ability to host the creation of valuable state. And this is decided much more by go-to-market/distribution, UX and other such factors, some of which are more actively impeded by a stricter relationship to the L1 sequencer than not.

Jul 27, 2026 · 8:56 AM UTC

3
6
13
6,442
Sort replies: Relevant Recent Liked
Replying to @qzhodl
只是Native rollup的落地按照计划表,要等到最后才实施,太久远了
1
64
今年年底 Native rollup 可能会有个 devnet
1
55
Replying to @qzhodl @qc_qizhou
一个slot 的finality ,怎么做到的 @grok
1
39
Replying to @qzhodl
已关注,回关秒回~
13