Pick the right wait time before you treat the next on-chain transfer as settled. Same block, two integrators, different correct answers. April 2026 was the worst drainer month on record - $629M+ across 20+ incidents per Chainalysis. Several of them came down to a single design choice nobody flagged at code review: the integrator treated an unfinalized transfer as settled. The reason that keeps happening is that "wait for N confirmations" gets copy-pasted across three completely different security models without anyone naming what the primitive actually is. Bitcoin's 6 confirmations is a probability estimate. The attacker has to spend mining electricity to re-org. The wait gets longer as hash concentrates, not shorter. Ethereum's "finalized" is a cryptoeconomic slashing guarantee. Reverting a finalized block would burn about one-third of the total stake. At May 2026 prices that is roughly 30 billion dollars at risk to overturn one block. The wait is ~12.8 minutes (2 epochs of 32 slots at 12 seconds). But Ethereum also exposes "safe" (justified, ~6.4 min) and "latest" (~12 sec) for builders who want a softer guarantee and faster UX. Solana's "optimistic confirmation" at ~12.8 seconds is a supermajority vote. In production it has never been rolled back. That is a strong operational statement, not a cryptoeconomic one. Alpenglow (the May 2026 testnet rollout) targets ~400 ms finality and is going to compress this dial significantly inside 2026. Three questions decide which version of this you should run today. One. Is the action reversible by you, in under an hour, at zero customer cost? If yes, the optimistic / safe-head is fine. Refunds, replay-able quotes, indexer reads. Two. Is the dollar loss from a reversal larger than the gross monthly revenue you make from this flow? If yes, wait for full finality and add a margin. Bridge mints, large withdrawals, irreversible payouts. Three. Both yes? Run two paths. Optimistic UX for the read path, finalized lock for the write path. That is how exchange-grade integrations work on every chain. If you cannot answer those three, more confirmations is not the fix. The integration needs a loss-function review before the next deploy. I built a one-page wait-time decision card with the Bitcoin / Ethereum L1 / Ethereum rollups / Solana variants and the loss-function triage. Link in the first comment. Where I might be wrong: the Solana "never rolled back in production" claim is true as of April 2026 mainnet data. If Alpenglow ships and the protocol changes underneath that claim, the operational versus cryptoeconomic gap that the current triage relies on starts to disappear inside 2026. Follow for daily insights - where blockchain meets AI, one satisfying swipe at a time. For educational purposes only. Not financial advice.