The minute of 2026-07-19 20:15 UTC was not uniformly broken. It was punctured by a five-second collapse: 212 of 305 eligible transactions failed between 20:15:10 and 20:15:14.999. Across the entire minute, 354 of 1,397 eligible transactions failed—a 25.34% failure rate, more than three times the frozen corpus baseline of 8.36%.
The burst had a recognizable shape. Ninety-seven failures stopped at Sui's native balance::split, 30 stopped while returning a flash-loan obligation, and 47 had no named abort site. A second, smaller wave arrived at 20:15:45: 42 of 128 transactions failed there. The first five seconds were comparatively calm, with 15 failures among 239 transactions.
The route that explains the spike
Transaction DsNxrvcn at 20:15:12.963 is a useful specimen. Its five-step route opened short-lived flash credit, moved through NAVX-, BUCK-, and SUI-related conversion calls, and prepared to return the receipt. The record places the abort at command 3, balance::split. The application effects rolled back; the only clearly reported durable cost was 0.000472256 SUI in gas.
A nearby transaction, 7ouj3NkZ, took a different path to the same kind of failure. It opened a quote-backed flash loan, ran three connected Cetus conversion calls, and then reached return_flashloan_quote with a submitted amount of 103.623667039 SUI. That repayment call aborted. The submitted amount is an input to the call, not proof that the amount was spent or that the route produced an exchange result.
The route family was not uniformly successful or uniformly understood. 6LVZqhzw, near the start of the minute, opened flash credit in USDC and passed values through calls touching SUI, NAVX, CETUS, and USDC before stopping with InsufficientCoinBalance; two later steps did not run. The system could reconstruct the sequence but withheld a named transaction behavior. Sui's own CLI documentation describes coin splitting as producing a requested amount and a remainder, which fits the native operation named in these failures. [Sui CLI cheat sheet](https://docs.sui.io/doc/sui-cli-cheatsheet.pdf)
One sender address, 0x5b7122…d94208, accounted for 55 of the 134 failed transactions whose asserted behavior was execute-swap-route. That is meaningful concentration in this minute, but it does not establish a bot, person, or shared operator.
What was happening around it
The 1,336 transactions with an asserted behavior show a broader operating mix. update-price led with 218 transactions (16.32%), followed by cancel-order with 189 (14.15%), execute-swap-route with 180 (13.47%), place-limit-orders with 153 (11.45%), and record-activity with 142 (10.63%). The system structurally explained 1,122 whole transaction compositions, but structural explanation and execution success are separate facts.
The order and oracle traffic continued through the failure burst. For example, CMDYQx83 at 20:15:00.016 prepared values and ran a directional pool-script route with a recorded sender-boundary change of −8,307 NAVX and +116.350702304 SUI. That movement does not, by itself, prove an exchange. The minute also included 302 sponsored transactions out of 1,397, meaning another address paid the gas in those cases.
What actually moved
Sender holding evidence recorded net inflows of 2,580.808749409 SUI—3,739.486917825 SUI in and 1,158.678168416 SUI out—across 790 transactions. Separate gas accounting recorded a net −1.270066914 SUI, with 1.375664484 SUI out and 0.10559757 SUI in across 1,049 transactions. These accounting views are not additive: the same gas charge can also appear in holding evidence.
USDC holding evidence showed a net −3,676.693962 USDC, while WAL showed a net +12,376.669266692 WAL. MAGMA was especially conspicuous at the boundary: 49,999.75 MAGMA out against 52,987.477001693 MAGMA in. None of these figures establishes a price, profit, or completed exchange.
The best-supported reading is therefore narrower than “the market broke.” A concentrated family of programmable routes hit balance and repayment guards in a short burst, while price updates, order maintenance, campaign records, and other routes continued around it. The chain records attempts and rollback boundaries; it does not establish why those routes were submitted or who coordinated them.
For a closer look, start with DsNxrvcn, 7ouj3NkZ, or 6LVZqhzw; the reusable route explanations are sui/execute-swap-route/roster-f0038/v1 and sui/execute-swap-route/coverage93-r046-whole/v1.