At 18:08 UTC on 19 July 2026, this frozen Sui minute compressed 7,350 eligible transactions into a near-continuous burst: the median gap was 0 ms across those 7,350 transactions, and the mean was 8.1 ms. It was also an abrupt failure event. 2,338 of 7,350 transactions failed—31.81%, versus 8.36% across the stored corpus—and the minute carried nearly 3.8 times the preceding minute's 1,924 eligible transactions.
The failure was front-loaded rather than evenly distributed. The first 40 seconds contained 2,294 of the minute's 2,338 failures (98.12%). In the final 20 seconds, there were only 44 of 2,338 failures. The repertoire changed with it: the opening five-second bucket was led by execute-swap-route (83 of 430 transactions), cancel-order (80 of 430), and place-limit-orders (72 of 430). At +5 seconds, 397 of 1,694 rows were only explainable at the call-chain level; at +10 seconds, that rose to 461 of 1,022. By +25 to +40 seconds, order placement and cancellation were again the leading described activity, while failures fell to 7 of 312 transactions in the +40-second bucket.
That shape looks less like one contract suddenly failing everywhere than a burst of incompatible attempts moving through several protocols. The largest failure group, 725 of 2,338, had no recorded stopping point. Among identified stops, 278 of 2,338 reached the framework balance-splitting primitive, 230 of 2,338 stopped while returning a quote-side flash loan, 178 of 2,338 stopped while removing an order-book entry, 112 of 2,338 failed a minimum-amount check, and 108 of 2,338 failed a slippage check. The evidence supports a mixed failure wave, not a single proven root cause.
One concrete example is 9VKLy9Qk, a seven-step execute-swap-route transaction at checkpoint 300528407. It opened a temporary flash obligation, passed CETUS through a directional router leg touching USDC, and reached the repayment step. It then aborted at the framework balance::split call at command 4; two submitted steps never ran. The sender-bound movement is −0.000811584 SUI, and the transaction narrative says the application changes were rolled back. Sui's own system paper describes the same all-or-nothing rule for aborted Move transactions, while noting that gas is still charged. [The Sui system paper](https://docs.sui.io/paper/sui.pdf)
A second route, FUTA1H4V, shows why “swap” is too strong a word here. It composed an outer quote-side flash loan, a Cetus route, and a partner flash swap, with a submitted amount of 49.624759 USDC. It reached the outer return_flashloan_quote call at command 9 and aborted there, with −0.001110264 SUI at the sender boundary. The route was attempted and structurally understood; the stored evidence does not establish a completed exchange or a profit.
The minute was not only failure traffic. Across the 6,703-row semantic population remaining after 647 rows were excluded, the leading asserted behaviors were place-limit-orders (1,326 of 6,703, 19.78%), cancel-order (1,223 of 6,703, 18.25%), and execute-swap-route (782 of 6,703, 11.67%). Together they account for 3,331 of 6,703 rows. Near the end of the minute, H2yfp4iS was a non-failing, sponsored DeepBook owner-proof transaction: it submitted 7,500 SUI at a stated 0.7461 USDSUI per SUI, with client order ID 86,192,258. Its detail does not establish that the order entered or remained on the book, so this is evidence of an order-placement call, not an executed trade.
The value record is equally striking but must be read carefully. Sender-bound holding movements report 350,548.201738145 SUI in and 1,025,539.559043655 SUI out across 3,715 transactions, a net of −674,991.35730551 SUI. A separate gas accounting reports −14.039921548 SUI across 5,543 transactions. Those figures are not additive: a gas charge can also appear in holding evidence. Other holding movements include a net +78,155.939057 USDC across 215 transactions, +96,731.241315324 WAL across 58, and +82,054.355909 DEEP across 54. There are no prices here, so these are boundary movements, not profit or loss.
The minute therefore reads as a short-lived execution storm followed by ordinary-looking order-book maintenance. It is consistent with automated, high-throughput submission—the 0 ms median gap supports that inference—but the data establishes addresses and transaction sequences, not a person, company, or bot. The useful exits are 9VKLy9Qk for the balance-split failure, FUTA1H4V for the nested repayment failure, H2yfp4iS for the sponsored order-placement call, and sender 0xcde6db…6c0e04, the busiest address with 1,015 of 7,350 eligible transactions.