At 19:34 UTC on 19 July 2026, the frozen Sui corpus recorded 1,601 transactions in 59.775 seconds of activity. 613 failed: 38.29% of all transactions, roughly 4.6 times the 8.36% failure rate across all 1,000,000 transactions in the frozen corpus. The sharpest clue is that the minute was not simply busy. It was a burst of flash-routed activity that repeatedly broke at value preparation or repayment, while one unusually large liquidity-position close moved hundreds of thousands of SUI and USDC.
For activity shares, the denominator is 1,533 transactions with an asserted transaction-level interpretation; the other 68 are excluded from those shares rather than being treated as “other.” The leading asserted behavior was execute-swap-route at 243 of 1,533 (15.85%), versus 5.72% across the frozen corpus. Update-price followed at 218 of 1,533 (14.22%). No single sender explains the spike: the busiest sender contributed 169 of 1,601 transactions (10.56%), below that sender’s 19.64% share of all corpus transactions.
Where the minute broke
The failure sequence has a recognizable shape. The largest abort label was unrecorded, with 222 of 613 failures (36.22%). Among named sites, framework balance::split accounted for 134 (21.86%), and return_flashloan_quote in one analyzed package accounted for 85 (13.87%). Those two named sites alone cover 219 of 613 failures (35.73%). The minute also contained a run of 71 consecutive failures.
The system structurally understood 1,092 of 1,601 transactions end to end (68.21%), down from 82.31% in the previous minute and below the frozen-atlas rate of 92.19%. That measure is separate from execution success: a transaction can be understood precisely even when it aborts. The named route FSEdcLZf is an example. It opened short-lived flash credit, passed NAVX through a conversion into SUI, and attempted to return the flash receipt. It stopped at balance::split on its third reached command; one submitted step did not run, and application changes were rolled back. Its recorded gas charge was 0.000399296 SUI. Reaching the route calls proves the attempt, not a settled exchange.
The timing supports a bursty-execution interpretation, but not an identity claim. The median gap was 0 ms, the 90th-percentile gap 219 ms, and the longest idle period only 467 ms. In other words, many transactions arrived in the same millisecond while the minute was active almost continuously. That pattern is consistent with automated submission or coordinated batching; the evidence does not establish who, or whether several addresses share an operator.
The money tells a second story
Sender-attributed holding movements show 661,204.04152169 SUI in and 636,806.654069862 SUI out, a net +24,397.387451828 SUI across 982 transactions. USDC moved in at 545,702.342687 and out at 588,014.977902, a net -42,312.635215 across 49 transactions. Gas was separately charged at 1.481806664 SUI across 1,276 transactions. These accounting categories must not be added together: gas can also appear in holding evidence.
The SUI number is dominated by one concrete row. Transaction 5GE5r5mg at checkpoint 300551677 called a close-liquidity-position wrapper. Its sender-attributed boundary movements were +399,995.19509906 SUI and +330,329.665721 USDC. The system describes the call as bundling liquidity removal, reward handling, fee collection and position consumption behind an exit wrapper. That explains why this minute’s SUI holding movement was about 98 times the corpus’s per-asset 99th-percentile movement. It does not establish profit, price, or the economic value of the position.
A contrasting route, HXsSkJQL, used four connected calls: quote-side flash credit, a Momentum conversion, a Cetus conversion, and flash-loan repayment. The exact receipt from the opening was passed to the close, and the sender-attributed boundary movement was +0.00813621 SUI after a 0.001413668 SUI gas charge. Even here, the stored explanation explicitly withholds an exchange outcome.
There was ordinary market-structure activity amid the failures. Sponsored transaction 3PQvnTpA passed an owner proof to a DeepBook limit-order call with 143,693 DEEP as the order size and a 0.02338 SUI-per-DEEP resting price. Its gas owner differed from its sender. The record identifies a placement attempt and a resting price; it does not prove that the order entered the book, remained there, or traded.
So the minute’s anomaly is a combination rather than one event: a high-throughput burst, an unusually failure-prone family of flash and balance routes, and one large position-close movement that overwhelms the aggregate SUI total. The clearest next reads are 5GE5r5mg for the outlier close, FSEdcLZf for the balance-split failure, HXsSkJQL for the route that reached repayment, and the reusable route pattern sui/execute-swap-route/roster-f0117/v1.