At 21:58:21 UTC, the M7.7 event clock started on the USGS record [E4]. On BMKG’s own origin clock, its first tsunami warning followed 2m16s later [E1]. The earliest modeled arrival sat another 1m44s beyond the warning, leaving four minutes from origin to the first modeled arrival [E1]. In a near-field event, that is the whole mechanical drama: an alert can reach shore before the island has time to explain what the shaking did [E1][E3].
Observed waves then reached multiple gauges, and BMKG later ended the warning [E2]. No cutoff-valid source in this record attributes a confirmed death to the tsunami; the fatality count belongs to the earthquake impact ledger [E2][E3]. A recorded wave and a death roster are different instruments, with different clocks and causal questions [E2][E3]. Nothing in the sequence establishes that the warning itself was mistaken [E1][E2].
By the frozen snapshot, BNPB counted 47 deaths and extensive building damage [E3]. Earlier provincial allocations had differed, so those rosters cannot be added together or treated as simultaneous truth [E3]. The 47 is the end-of-window national snapshot, not a license to rewrite earlier allocations backward [E3]. The casualty map therefore carried two uncertainties during the day: local counts were changing, and some of the channels carrying them were damaged [E3][E6].
Telecoms made that second uncertainty physical. Komdigi identified 200 sites or BTS with disruptions [E6]. Electricity was also a moving constraint: PLN reported 429 substations restored, a recovery count from a network still being worked back into service [E5]. When casualty geography arrives through impaired cellular and power infrastructure, apparent silence can describe access as much as impact; that is an inference about the reporting channel, not a correction to the death roster [E3][E5][E6].
Health and transport nodes were taking the same hit. At RSUD Komodo in Labuan Bajo, all patients were moved out when the quake struck, with care shifted outside after the hospital was damaged [E7]. BNPB’s frozen impact picture also recorded damage affecting airports, ports and roads [E3]. Each break adds another handoff between an injured person and the national ledger, so the hardest-hit picture was being routed through infrastructure that could fail at the moment it was needed [E3][E7].
Reporting bias is not the default explanation for where deaths clustered. The M7.7 origin and extensive building damage leave other mechanisms live: building vulnerability, occupancy, slopes and rupture physics could shape casualty geography before communications enter the story [E3][E4]. The available record does not rank those mechanisms, and it does not show that districts with broken access were necessarily undercounted [E3][E6]. A clean test requires later district rosters matched against restored access and victim-level reconciliation; until then, infrastructure damage identifies a possible distortion channel, not the cause of the deaths [E3][E5][E6].
The discriminator is mundane and hard: watch whether district allocations settle as roads, power and telecoms return, and whether the reconciled victim list shifts toward places that were hardest to reach [E3][E5][E6]. If it does, access was part of the early reporting shape; if it does not, the counter-case strengthens [E3][E5][E6]. The warning system had minutes to turn a seismic clock into a coastal instruction, while hospitals, ports, roads, towers and substations had to survive the event before they could report through it [E1][E3][E5][E6][E7]. The small interval remains: the warning had already reached the coast while the island was still becoming legible [E1][E3][E6].