This minute began like a malfunctioning relay. Of 5,805 eligible transactions, 2,081 failed—a 35.85% failure rate, more than four times the frozen corpus rate of 8.36% (83,649 of 1,000,000 transactions). The failures arrived in a wave: 790 of the first 1,531 transactions failed in the opening five seconds, while only 1 of the final 140 failed.
The chronology suggests several overlapping failure modes rather than one proven system-wide outage. In the first ten seconds, flash-loan repayment and balance-splitting failures dominated. Between 17:02:15 and 17:02:35, slippage checks became prominent. Later, failures shifted toward order-book maintenance. The evidence does not establish that these were caused by one operator or one shared incident.
Among the 5,420 transactions with supported semantic meaning, the leading activities were placing limit orders (1,009, or 18.62%), executing swap routes (860, or 15.87%), and cancelling orders (840, or 15.50%). Of all 2,081 failed transactions, 689 carried an asserted swap-route behavior and 156 carried an asserted order-cancellation behavior. Another 385 rows were excluded from meaning-based counts; their exclusion is not evidence of success or failure.
One early route shows the mechanics clearly. Digest 3zXAAvbv opened short-lived flash credit, passed a borrowed CETUS balance through a router touching USDC and CETUS, and attempted to return the balance. It stopped at balance::split; the remaining two submitted steps did not run. Its only retained sender movement was −0.000811584 SUI in transaction-wide balance change and gas accounting. The route was attempted, not proven to have completed an exchange. When a Sui transaction aborts, application effects are rolled back while gas remains charged, consistent with Sui’s documented all-or-nothing execution model ([Sui system model](https://docs.sui.io/doc/sui.pdf)).
Digest Bz6MyaQh shows a different stop. It opened a quote-side flash obligation, reached a Cetus conversion touching USDC and SUI, then a DeepBook conversion touching DEEP and SUI, before aborting at return_flashloan_quote. The transaction submitted 2,018 SUI as the repayment amount, but that input does not prove that the amount was spent or that either conversion completed.
The minute also contained a striking successful exception. At 17:02:20.148 UTC, 9VWpzgiw closed a pool position after two separately supported reward-collection calls. The sender’s boundary movements were +318,505.67615826 SUI, +233,704.861023 USDC, and +0.345992917 CETUS. The retained evidence does not reveal the reward amounts or assign those movements to a particular call, so this is not evidence of a payout or profit. It is simply a large, positive boundary movement amid the failure wave.
By the final second, the remaining failure was back in DeepBook. Digest 9udAqvxQ created a delegated trader proof and attempted to cancel an identified order, but stopped at big_vector::leaf_remove; the cancellation’s application effects rolled back. The transaction records the attempted order ID, not a completed cancellation.
Four exact revisions have their earliest stored calls in this minute, after four hours of preceding observation, but their complete corpus usage totals only 19 calls. They genuinely enter this window's observed repertoire here; the minima still do not establish when the packages were deployed. That timing is worth preserving as a lead for further investigation, not treating as an explanation for the spike. The stronger conclusion is temporal: a dense burst of flash-route and balance failures arrived first, slippage failures followed, and order maintenance continued after the broader failure rate had collapsed.
Further exits: digest 3zXAAvbv, digest 9VWpzgiw, and pattern sui/execute-swap-route/coverage93-r027-committed-or-repayment-abort/v1.