Between 21:55:00.200 and 21:55:59.800 UTC, the frozen corpus records 1,819 eligible transactions. Exactly 498 failed: 27.4%, compared with an 8.36% failure rate across the corpus. The activity was densely packed—median gap 0 ms, mean gap 32.8 ms—and the five-second bucket beginning at 21:55:35 contained 458 transactions, including 198 failures.
The spike was not one uniform operation. Of the 1,819 rows, 1,680 had enough execution evidence for semantic interpretation; 376 of those failed, or 22.4%. The remaining 139 stopped too early for a safe transaction-level purpose. The engine still supplied complete structural explanations for 1,429 rows, but that does not mean those transactions succeeded.
One sender, several fixed stopping points
Sender 0x7744ad…a7c7fa submitted 136 eligible transactions and 110 failed. Its failure pattern is unusually concentrated: 41 stopped at pool::flash_swap_internal, 26 at pool::swap_in_pool, 17 at pool::swap, 10 at validation::check_min_amount within an asserted swap-route pattern, and 8 at vault::withdraw. Another 8 reached the same minimum-amount check without an asserted behavior.
The sender’s median gap was 0 ms and its 90th-percentile gap was 259 ms. That cadence is consistent with automated execution, but the address evidence does not establish a bot, owner, or shared operator.
FyE9zpG6 shows the first route family clearly. At 21:55:00.200, it began a partner-attributed flash-credit opening and aborted at pool::flash_swap_internal with Move abort code 11. The recorded route had 12 steps, but the remaining 11 did not run. Its boundary movement was −0.00027066 SUI in the holding lane and −0.00027066 SUI in the gas lane. This proves an attempted route and a failure point, not a completed exchange.
GqLv1TbJ shows a different stopping point in the same broader family. It withdrew a SUI coin, constructed an opposite USDC balance, and reached an exchange-related call before aborting at pool::swap_in_pool with code 1009. The transaction submitted 756.764902171 SUI as an input, but that is not evidence that the amount was spent or that output was produced. The remaining six steps did not run; the boundary gas movement was −0.000659972 SUI.
Sui’s transaction semantics are all-or-nothing for application effects: when execution aborts, earlier application changes do not persist, although gas is still charged. [Sui’s system paper](https://docs.sui.io/doc/sui.pdf) describes this rollback behavior.
A second failure mode: no safe purpose at all
The minute’s largest single failure label was unrecorded, with 227 of 498 failures. In total, 312 failed rows had no asserted behavior. The system withheld a purpose rather than guessing one.
GqZ1BnrX, from the minute’s most prolific sender 0xcde6db…6c0e04, is a useful example. It stopped before any submitted command could be safely attributed because a withdrawal request exceeded available funds. Its two-step shape contained owner-proof generation followed by an order-cancellation call, but no application action was asserted and no sender movement was recorded. That is a failed attempted shape—not proof that an order cancellation was reached.
The same sender produced 297 eligible transactions in the minute and 130 failures; 122 of those failures were unrecorded. Meanwhile, the interpretable activity remained mixed: limit-order placement accounted for 259 of 1,680 rows, asserted swap routes for 232, price updates for 217, and order cancellations for 215. Swap-route activity was elevated here—13.8% of interpretable rows versus 5.7% across the frozen corpus—but a route label still does not establish an exchange.
What moved while this happened
Sender-boundary movements show +109,291.293326394 SUI in and 105,415.543118096 SUI out in the holding lane, for a net of +3,875.750208298 SUI across 1,016 transactions. The gas lane is separate: 0.314231458 SUI in, 5.162444767 SUI out, and −4.848213309 SUI net across 1,289 transactions. These lanes must not be added together.
WAL moved in the opposite direction at the minute level: 290,339.999205869 WAL in, 697,103.134442931 WAL out, and −406,763.135237062 WAL net in the holding lane. These are boundary movements, not prices, profit, or proof of a common economic purpose.
The largest individual holding movement was not marked failed. 4SvFUEzy recorded −429,916.498463106 WAL and −23,367.220665941 SUI while running an intended liquidity-position opening route. Its detail supports the opening attempt and the boundary movements, but not whether a position actually opened or what contribution was accepted.
The best explanation supported by this minute is therefore a burst of tightly repeated route attempts encountering several recurring guards, alongside a separate stream of under-explained order-related failures. The evidence supports retry-like automated execution and rollback at known stopping points; it does not establish the cause, the operator, or any completed trade.