Finality is usually described as a property a chain has. It reads better as a budget a chain spends. Every design buys a kind of certainty with a fixed amount of time, and the price of that purchase is the number an exchange desk, a bridge and a user each experience differently.
Two kinds of certainty
Probabilistic finality is the older model. A block gets harder to reverse with every block built on top of it, and the cost of reversal rises with each confirmation. Nothing is ever finished. The curve simply gets steep enough that no rational attacker follows it.
Economic finality makes the same argument explicit. Reverting settled state would require burning capital that exceeds the value of the reversion, so the guarantee is a price rather than a timestamp. A chain is final when attacking it costs more than the attack is worth.
Proof-based finality is the third case, and the one people usually mean when they say a chain is final. A validity proof or a passed challenge window converts a statement that is merely expensive to reverse into one that cannot be reversed without breaking a rule the verifier checks. It is the strongest of the three claims, and it is always paid for in latency.
Why a deposit desk counts confirmations
An exchange crediting a deposit is not verifying finality. It is running a risk model. The desk picks a confirmation count, watches for reorgs deeper than that count, and treats the residual risk as a cost of doing business. The credit appears fast because the venue, not the chain, is carrying the exposure.
That is a rational trade for a venue with a treasury and a fraud team. It is a poor trade for anyone who reads the credit as proof. Between the credit and the settlement there is a window in which your balance is an entry in someone else’s database, backed by their solvency rather than by a state root.
The count itself is a policy number, set per chain and revised when reorg behaviour changes. It is not published as a guarantee, and it should not be read as one.
Why a bridge waits
A bridge moves a claim rather than a balance, so it cannot settle on a risk model it controls. It has to wait until the claim is enforceable, which on an optimistic rollup means waiting out the challenge window in full.
That is why the clock is measured in days, and why the number differs by design. Ethereum.org’s L2 directory, in its October 2026 listing, shows Base and OP Mainnet at 7 days to mainnet, Arbitrum’s dispute window at 45,818 blocks, about 6.4 days, and ZKsync Era at a 3-hour minimum. The first two are waiting for the right to object to expire. The third is waiting for a proof to be produced and checked.
Neither is a defect. Both are the published price of a settlement model, and a user who moves size without reading that price is guessing at the risk.
| Route | Exit to mainnet, Ethereum.org, October 2026 | What the clock actually buys |
|---|---|---|
| Base | 7 days | Expiry of the challenge right on an optimistic rollup |
| OP Mainnet | 7 days | The same model under a different operator |
| Arbitrum One | 45,818 blocks, about 6.4 days | A dispute window counted in blocks, not days |
| ZKsync Era | 3 hours minimum | Proving and verification latency |
| Centralised venue | No chain window, venue policy decides | Custodial credit on the venue’s book |
What the window costs you
A finality window is not idle time. It is time in which your capital is committed, your price is not, and your optionality is gone.
Send the same notional through a 7-day exit and a 3-hour exit, the windows Ethereum.org listed in October 2026, and you are not comparing fees. You are comparing how long a position stays unreachable. On a volatile asset, a week of unreachable capital is a directional bet you did not choose to make.
The fee side of the trade is small by comparison. FeeEdge’s survey of 17 August 2026 put L1 ERC-20 withdrawals at $2 to $10 and above $50 under congestion, with Base transfers at under $0.01 to $0.10 and Arbitrum at $0.01 to $0.50. Those numbers are visible before you commit, though fee pages change more often than most traders assume. The window is not priced anywhere, which is exactly why it gets ignored.
Verdict
Treat finality as a budget line, not a checkbox. Ask what the chain spends to give you certainty, and how long your capital sits inside that spend.
An exchange credit is a fast, cheap promise from a counterparty. A bridge exit is a slow, expensive proof from a protocol. Both are legitimate, and the mistake is using one and assuming the other.
Before you move size, read the exit window next to the fee, and check who secures a rollup to see which party your window depends on. If a route cannot tell you its clock, it has told you something already.
0 comments
No comments yet — ask a question and we will answer it here.