Aptos runs transactions in parallel first and sorts out the collisions after
Most parallel chains require transactions to declare in advance which data they touch.
@Aptos does not, per its docs, taking the opposite bet to Solana and Sui.
The engine called Block-STM runs a block's transactions in parallel while tracking what each one read and wrote. Clashes are caught in a validation pass, and the losers are re-run until the results match running everything one at a time in the block's preset order.
The trade-off lives in busy moments: heavy contention on the aptos:native network means more re-runs, a cost both Aptos Labs and independent researchers acknowledge.