At 14:29 UTC on July 19, 2026, Sui’s failure rate jumped sharply: 509 of 1,969 eligible transactions failed, or 25.9%, versus 8.36% across the frozen 1,000,000-transaction corpus. The failures were heavily front-loaded. 473 of 509 occurred during the minute’s first 20 seconds; only 36 arrived afterward.
This was not one address repeatedly hitting one error. The busiest sender accounted for 417 of 1,969 transactions, but only 26 of those 417 failed. The minute also included 356 sponsored transactions, where the gas owner differed from the sender.
The repayment guard was the main choke point
The largest failure cluster stopped at flash_loan::return_flashloan_quote: 137 of 509 failures, or 26.9%. Another 113 of 509 had no specific stopping call recorded. The next major sites were the framework’s balance split, with 49 of 509 failures, followed by a DeepBook vector-removal call with 31 of 509 and a slippage check with 29 of 509. Those five sites together account for 359 of 509 failures, or 70.5%.
The flash-loan cluster was strongly associated with routed execution. 133 of the 137 failures at return_flashloan_quote carried the execute-swap-route behavior; four had no asserted behavior. Across the minute’s 1,958 rows with behavioral explanations, the three leading behaviors were cancel-order at 339 of 1,958, execute-swap-route at 330 of 1,958, and place-limit-orders at 313 of 1,958. Together they represented 982 of 1,958 explained rows, or 50.2%.
The package analysis gives the repayment guard a concrete meaning: the supplied quote coin must be strictly greater than the stated repayment amount. A representative failure, 7AxMSCZ9, opened a quote-side flash loan, created an empty intermediate coin, passed value through DeepBook and Cetus, and then stopped at command 4 while returning the flash-loan receipt. Its recorded Move error was code 14680251. The route’s application effects were rolled back, leaving only a boundary change of −0.00038326 SUI for the sender. The same charge appears in both the holding and gas evidence, so it must not be counted twice.
That distinction matters. Sui’s transaction model treats an aborted transaction as having no application effects while still charging gas. [Sui’s protocol paper](https://docs.sui.io/paper/sui.pdf) describes the same all-or-nothing behavior.
A second example, 5chiJGU3, shows that the minute was not failing at one universal command. It opened a flash obligation, routed DEEP through successive adapters toward WAL and SUI, and then stopped earlier at command 3 in balance::split with code 2. The final repayment step never ran. Its sender record shows −0.000721996 SUI, again without proving that the attempted route completed or that an exchange occurred.
The successful routes remain carefully bounded
The minute did contain committed routes. Hb5pALLM ran a two-step direct router path with a boundary movement of +99.998710268 SUI and −75.115159 USDC for the sender. But the retained detail says the step-bound output was unavailable and explicitly does not establish an exchange. Opposite-signed asset movements are evidence of sender-boundary changes, not proof of a completed trade or profit.
The broader movement picture is similarly large but not economically interpretable without prices. In the SUI holding record, 217,111.528645686 SUI entered and 54,997.076721054 SUI left across 1,038 transactions, for a net of +162,114.451924632 SUI. In the USDC holding record, 112,443.563985 USDC entered and 116,849.744318 left across 81 transactions, for a net of −4,406.180333 USDC. These are governed boundary movements, not earnings calculations.
Only 1,611 of 1,969 eligible transactions received a complete structural explanation in this minute. That figure is not a success rate: failed transactions can still be structurally understood, while a transaction without a complete explanation is not necessarily failed. The striking fact here is the timing and mechanism of the failures—a dense early burst dominated by flash-funded routes that often reached repayment or balance handling before aborting.
For a closer look, start with 7AxMSCZ9, 5chiJGU3, and the committed contrast Hb5pALLM; then compare the recurring coverage93-r004-committed-or-repayment-abort/v1 route with the strict-remap-440e561741be/v1 balance-adapter path.