Transaction speed gets all the attention. Throughput numbers dominate marketing decks, and TPS benchmarks get cited in token launches like they’re the only metric that matters. But finality - the point at which a transaction is mathematically, irreversibly settled - is the harder problem, and the one that determines whether a chain can actually support high-stakes applications.
Probabilistic vs. Deterministic Finality
Bitcoin’s finality is probabilistic. After one confirmation, a transaction is included in a block, but a reorganisation could theoretically reverse it. After six confirmations - roughly an hour - the cost of mounting such an attack becomes economically irrational for most adversaries. The chain doesn’t promise finality; it makes reversal progressively expensive.
Proof-of-stake systems like Ethereum post-Merge can offer something closer to deterministic finality. Ethereum’s finality checkpoint mechanism, based on Casper FFG, finalises blocks every two epochs - roughly 12–13 minutes under normal conditions. Once finalised, reverting a block would require destroying at least one-third of all staked ETH, which currently represents tens of billions of dollars. That’s not probabilistic in any practical sense.
Solana operates differently again. Its optimistic confirmation model gives transactions fast apparent finality, but true economic finality - where supermajority stake has voted on a block - arrives later. The gap between these two states is where most of Solana’s outage-related disputes live.
Why Applications Care
For a DEX executing a swap, fast soft finality is usually enough. The risk of reversal is low, and the cost of waiting is real.
For a cross-chain bridge moving $10 million between networks, the calculation inverts entirely. Bridges that settled too early on probabilistic finality have been exploited through reorg attacks. The Ronin bridge hack in 2022 wasn’t a finality exploit specifically, but it demonstrated how assumptions baked into bridge architecture - about what counts as settled - become attack surfaces.
That’s why finality models are now a central part of how bridge protocols and rollup sequencers design their systems. Optimistic rollups on Ethereum inherit Ethereum’s finality but introduce a 7-day fraud-proof window before funds can be withdrawn to L1, which is a different kind of finality problem: economically safe, but slow by design.
The Underrated Variable
Finality time isn’t just a safety metric. It’s a latency floor for any application that needs to compose across chains or settle in a way that downstream systems can trust.
As more financial infrastructure moves on-chain, the difference between “probably settled” and “provably settled” is going to matter in ways the throughput race mostly ignores.