Alpenglow cuts finality on Solana's test networks from 12.8 seconds to about 150 milliseconds

TowerBFT produced irreversible settlement on the order of 12.8 seconds ^[7]. Votor's fast path aims for roughly 100 milliseconds when a large share of stake agrees in a single round, and its second path aims for about 150 milliseconds when another round is needed ^[8]. Divide the old figure by each target and the fast path is 128 times quicker, the slower one about 85 times ^[20]. The Solana Foundation's own upgrade materials describe the 150-millisecond number as an expectation measured against a given stake map, not a constant that will hold in every geography or under every load ^[9].

Anza is changing how long a transaction stays reversible and leaving what it costs alone. Execution, programs and fees are intended to stay the same ^[10].

The testing work falls on application teams first. Devnet is where most programs are exercised with valueless tokens ^[18], and Anza's devnet post told developers to "test your programs and integrations now during our observation period leading up to mainnet-beta migration" ^[14]. According to Crowdfund Insider, millisecond settlement would change how wallets, exchanges, bridges and liquidation engines treat confirmation ^[19]. Operators will watch whether "confirmed" and "finalized" can safely collapse into one state ^[17].

Solana co-founder Anatoly Yakovenko answered Anza's devnet countdown with five words: "You are not ready for this." ^[15] The same account says the speeds remain largely unproven under full mainnet conditions ^[11].

If the observation window closes cleanly, mainnet-beta could cross on Agave alone, the client both public migrations leaned on because Firedancer and Frankendancer still lacked Votor support ^[16]. The window could also stretch while those clients catch up. Or finality could come in slower once real balances, real latency and a full validator set are involved, the conditions under which hidden assumptions surface ^[12].

I think the mainnet date will be set by client support more than by latency. Anza's sequencing puts mainnet-beta after an observation period with no live-network date attached ^[1], and client diversity is on the list operators will watch ^[17]. The counter-case is that Anza has already taken two public clusters through the handoff leaning on Agave ^[16] and could do the same with the third. A mainnet-beta date announced before Firedancer supports Votor would prove this view wrong.