Thirteen exact revisions have their earliest stored calls in this minute, about nine minutes after observation began. That is a genuine change in this bounded corpus's observed repertoire, not proof that any package was deployed at 13:10. Two of the revisions left an unusually clear fingerprint: a burst of flash-backed route attempts that reached repayment or balance preparation and then stopped.
The minute contained 1,676 eligible transactions, including 132 failures. For behavior shares, 1,655 transactions had an asserted behavior: cancel-order led with 376 (22.72%), followed by place-limit-orders with 332 (20.06%) and update-price with 208 (12.57%). The failure rate was 7.88%, up from 6.10% in the previous minute but still below the frozen corpus rate of 8.36%.
The key sequence began at 13:10:30.544 UTC. Transaction 9WZbv23S, using revision 0x520c61…1e926f for the first stored time in this window, opened short-lived flash credit, routed WAL through a WAL/SUI exchange-related call, prepared the return, and aborted at para::r_c_f_s_b with code 401. Five more rows using that revision followed the same narrow family of loan-return failures.
Two hundred milliseconds later, 4vKswxMk used another revision not previously stored in the window, 0xc17bf1…c740f0. It opened a flash-backed position, routed WAL into SUI, and reached the repayment call before the framework's balance::split aborted. Twelve transactions in this minute followed that same route shape and failure site. Their application effects were rolled back; their reached calls prove attempted execution, not a completed exchange.
That sequence fits the minute's failure map. The largest abort site was the quote-side flash-loan return, with 28 of 132 failures. Framework balance splitting accounted for another 17, while 17 failures had no recorded site. Price freshness and minimum-amount checks contributed 12 and 11 respectively. The evidence supports repeated route attempts encountering repayment or input-balance guards; it does not establish the underlying cause.
A busy order-maintenance backdrop
The route failures were embedded in a much larger DeepBook maintenance stream. Two senders accounted for 624 of the 1,676 eligible transactions: 0xcde6db…6c0e04 sent 384, and 0xffd4f0…d2f0ad sent 240. Their repeated activity included cancellations, placements, and price-related calls. The cadence was highly compressed: the median gap was 0 ms, the 90th-percentile gap was 209 ms, and the longest idle period was only 640 ms.
The individual rows show why call names need careful reading. In 2SNkE4Jc, an owner-proof workflow cancelled an identified DeepBook order and recorded 1,200 SUI as resting size removed from the book; that does not show the order had not traded earlier. In sponsored transaction A4GqmfkQ, the sender submitted a 169.5 WAL order at a resting price of 0.029524 USDC per WAL. The evidence identifies a placement attempt and a resting price, not an execution price or guaranteed durable order state. There were 451 sponsored transactions in the minute, with another address paying gas for someone else.
What moved
Separate sender-boundary holding records aggregate to +63,068.518954541 SUI, +74,559.771145 USDC, and −440,816.598965397 WAL. Gas records separately show 1.795003842 SUI out and 0.223466022 SUI in, for a net gas movement of −1.571537820 SUI. These are per-asset boundary movements, not prices, profit, or proof that the assets exchanged. In particular, the route evidence does not establish an economic swap merely because multiple assets appear in the same transaction.
For the closest look at the burst, start with 9WZbv23S and 4vKswxMk; for the surrounding order activity, compare 2SNkE4Jc with the place-limit-orders pattern.