The minute from 20:49:00 to 20:49:59 UTC carried 1,771 eligible transactions across checkpoints 300571132–300571378. 414 failed—23.38%, versus 8.36% across the frozen corpus. Of the 1,713 transactions with enough execution evidence for semantic labels, 363 failed, or 21.19%.
The failures arrived in waves rather than evenly. Between 20:49:10 and 20:49:19.999, 241 of 578 transactions failed. Another 71 failures appeared between 20:49:40 and 20:49:44.999. The minute was also extremely bursty: the median gap was 0 ms and the 90th-percentile gap was 234 ms.
What broke
The largest named failure group was execute-swap-route: 187 of the 363 semantically classified failures. The most common stopping points were the flash-loan quote repayment call, with 98 of all 414 failures; balance::split, with 45; and a slippage check, with 25. The system could not assign a recorded abort location to another 145 failures.
This is also a useful warning about reading explorer labels. The engine structurally explained 1,472 of the 1,771 transactions, yet 201 of those explained transactions still failed. Understanding what a transaction attempted is separate from whether it completed. Sui’s transaction model is all-or-nothing for application effects after an abort, while gas is still charged. [The Sui paper](https://docs.sui.io/paper/sui.pdf) describes the same rule.
Three routes, three different stories
Transaction CxWXfjqv is the clearest example of the minute’s dominant failure mode. It opened a flash credit, passed SUI through calls touching WAL/SUI, WAL/USDC, and USDC/SUI, then presented the returned SUI and receipt to return_flashloan_quote. That final call aborted with code 14680251. The transaction submitted 41.321325091 SUI, but the route does not prove an exchange or a completed repayment. Application effects were rolled back; the sender-boundary record shows only −0.00048168 SUI in gas and holding accounting.
3jbT1AnF followed a different route. It withdrew 44,693.243877074 WAL, converted values into balances, created an empty USDC coin, and reached a pool swap. The pool stopped it with code 1009, leaving six later steps unexecuted. Its recorded sender movement is −0.001853916 SUI, again without a proven application-level exchange.
By contrast, ErUAsm25 completed its two connected calls without an execution error. Its boundary movements were +400 MAGMA, −118.01517 USDC, and −0.001552 SUI. Those movements show what reached the sender boundary; the stored explanation still does not establish that the route was an economic exchange.
The surrounding market machinery
The minute was not only a failure storm. Among the 1,713 transactions with semantic labels, the leading behaviors were cancel-order at 288, place-limit-orders at 272, execute-swap-route at 259, and update-price at 209. The busiest sender was 0xffd4f0…d2f0ad, with 285 of 1,771 transactions. Its activity was almost entirely order placement and cancellation, and all 285 transactions used the separate gas owner 0x59e8e0…bdc610; the addresses’ relationship is not established by this evidence. Overall, 459 of 1,771 transactions were sponsored.
The sender-boundary movement record shows substantial activity, but it is not a profit statement. Holding-accounting totals were −5,853.063536788 SUI, −250.881172 USDC, +86,250.252682463 WAL, −85,586.994089 DEEP, and −9,947.058208 USDSUI. Gas accounting separately recorded −1.992766292 SUI. These categories must not be added together: gas can also appear in holding accounting, and there are no prices for comparing the assets.
The minute’s central fact is therefore narrower than “swaps failed.” A dense burst of multi-step routes repeatedly reached repayment, balance, pool, or slippage guards, while order-book maintenance and price updates continued around them. The evidence shows attempted paths and rolled-back failures—not a common operator, a proven trade, or a measured loss.
For the closest inspection, start with CxWXfjqv, 3jbT1AnF, and ErUAsm25; then compare the repeated sui/execute-swap-route patterns with the minute’s dominant sui/cancel-order/deepbook-owner-proof/v1 activity.