The minute's most revealing event was not a failure. One of the 12 exact revisions with earliest stored calls in this minute became the opaque middle of four elaborate routes. Ten minutes of earlier corpus traffic make this a real entry into the window's observed repertoire, but not proof that the package itself was newly deployed.
At 13:12:05 UTC, sender 0xae4ba5…90548c reached router::swap_exact_in_direct_w1 in revision 0x011354dc…e66fa3. Three more transactions from the same on-chain address reached that revision before the minute ended. All four had different 24-step transaction shapes. They prepared supporting values, split or transferred assets, entered one or more router calls, passed balances through settlement steps, and ended with further transfers. One route reached the new function twice; another combined it with swap_exact_x_to_y_w1 from a different revision.
The boundary evidence shows 0.732319152, 0.733954467, 0.733055844, and 0.734222623 SUI leaving sender-held value across those four transactions—a combined 2.933552086 SUI. That is a sender movement, not proof of an exchange, output, profit, or loss. The calls reached the router functions, but the package had no authored analysis and no recognized transaction-level activity. The engine could describe the sequence while withholding what the route meant as a whole.
That gap matters because the revision was not an isolated curiosity: it was observed in 148 transactions across the frozen corpus after its first stored use here. In this minute, however, all four uses remained unexplained at the purpose level. The evidence resembles a multi-router swap workflow, but it cannot establish which assets were exchanged or whether any exchange completed.
The rest of the minute was an unusually dense market-maintenance burst. There were 1,166 eligible transactions between checkpoints 300447278 and 300447557, with a median inter-transaction gap of 0 ms, a 90th-percentile gap of 213 ms, and a longest idle period of only 642 ms. Of the 1,166 rows, 1,155 supported a semantic activity reading; the activity breakdown was led by update-price at 212/1,155, cancel-order at 176/1,155, place-limit-orders at 158/1,155, and record-activity at 149/1,155. Together those four accounted for 695/1,155 rows.
Failure was not the defining anomaly. Ninety-seven of 1,166 eligible transactions failed, an 8.3% rate—almost identical to the frozen corpus baseline of 8.36%. The leading stopping points were an unrecorded site in 19/97 failures and a flash-loan quote repayment call in 18/97. A representative cancellation, AzQqBUKy, generated a temporary DeepBook trader proof, attempted to remove order 544492544823684811098111, and aborted at big_vector::leaf_remove; the attempted application change rolled back, leaving only a 0.000745424 SUI gas charge. Reaching a cancellation call is therefore not evidence that the order was cancelled.
The cleanest reading is that 13:12 was a knowledge boundary, not a confirmed router outage: a revision entered the corpus's observed use after ten minutes of preceding traffic, several complex routes reached it, and the available evidence could not yet name their purpose or outcome. The sender address identifies who submitted those four transactions, not a proven operator or organization.
For a closer look, start with 7yeHQsvxM5JuYEYoFamsGyDT4m7zgcJYp1BE7M9sRjzG, the first observed route through the revision; compare C3mbjEZQarRUUcmnWo7B8F1Ks6ETkagoHeNZyJBSYtyv, which adds another router revision; then contrast both with the failed DeepBook cancellation AzQqBUKyPZoTgr75NyjarhJwHvoAZXQqDmoSaCmU3UB9.