IMHO, L2Beat's/Vitalik's Stages are incomplete. For Stage 2, in addition, 

- Alternate sequencing triggered within 1 epoch (i.e. 6.4 minutes). Currently, Optimism is 12 hours, which is not acceptable (preferably, there's real-time distributed sequencing) 
- Security council has pause rights only; upgrades even for onchain provable bugs must go through governance, emergency path with >X% approval

Then, there should be an optional Stage 3, or Stage X, where:

- Based or based-like sequencing is mandated
- Security council is dissolved
- Smart contracts upgrades require >Y% (80%?) approval from the protocol's full governance, in addition to >30d delay

For big-tent high-value L2s like Arbitrum One, Stage 3 should be the target. For most, Stage 2 should be perfectly fine, and even Stage 1 enough for many lower-value app-specific L2s.