Kerr County issued four wireless emergency alerts as flood risk rose early Thursday, and the city of Kerrville issued one. Three of six new sirens in Kerr County sounded. In Comfort, two recently installed sirens sounded twice. The improvements were real. They were also local. An Associated Press review found no wireless emergency alert listed from a Uvalde County agency, while cautioning that officials there may have warned people through other channels. [1]

That uneven record answers the question left by Friday's account of more than 570 rescues. Rescue boats, closed roads and a broken bridge approach showed what a warning had to outrun; they could not show who heard one. Saturday's AP audit begins to supply the missing delivery record without turning five communities into a verdict on Texas.

The distinction also carries forward the paper's first flood assessment, which separated forecast, alert, delivery, evacuation route, transport and rescue. AP found progress at several links in that chain. It also found households awakened by friends, by water on the floor or by repeated calls rather than by a listed local wireless alert. [1]

This is what an improved warning system looks like before an after-action report makes it tidy. One county sends alerts it withheld during the previous year's catastrophe. A town adds sirens. A private company installs river-triggered towers. Another county uses social media and door knocks. Some residents leave. Others linger. Still others discover the flood inside their homes. Success and failure can occupy the same storm because a warning is not a statewide object. It is a message delivered to a particular person, in a particular place, while a usable road still exists.

Alerts reached places, not Texas

The National Weather Service sent 38 alerts to selected southwest Texas communities between early Tuesday and about 9 a.m. Thursday: 14 tornado warnings and 24 notices that flooding was occurring or imminent, some describing a life-threatening threat. Those notices went through broadcast outlets, weather radios and mobile phones. Local authorities could add more precise instructions. Kerr County's alerts told residents along Quinlan Creek to move to higher ground and warned of extremely dangerous flash flooding. People enrolled in the county's CodeRED system also received text messages. [1]

That is a layered system rather than a single alarm. A Weather Service warning can define the hazard across an area. A county message can name the creek or evacuation instruction. A siren can reach someone asleep, outdoors or away from a phone. A door knock can find a household outside the digital network. Each layer has a different denominator, and one working layer does not prove that the others reached everyone.

Kerrville resident Suzanne Sutphin Gschwind told AP that the contrast with the previous year was striking. She described multiple texts and calls from local authorities, a weather channel and her doorbell camera, sometimes arriving about every two hours. Her experience is evidence that warnings multiplied for one resident. It is not the receipt list for every resident along the river. [1]

The Uvalde-area accounts show the other side. AP's review found no wireless emergency alert listed from agencies in Uvalde County, although the agencies may have used other methods. Jaclyn Gonzales said a friend called around 2 a.m. to warn of a possible tornado; when she got out of bed, she felt water on the floor. In Batesville, Kat Sprawls said a friend's call around 3:30 a.m. was the warning that water was approaching, and that it took five or six calls to wake her because her phone was set to do not disturb. [1]

Those details do not prove that no official warning existed. They establish the narrower and more useful fact: the listed wireless record did not contain a Uvalde County agency alert, and two residents described learning of danger through friends. A complete review must now identify every channel the county did use, the intended audience for each one and whether the message arrived with time to act.

Zavala County offered another channel. A sheriff's department official said the department updated Facebook with evacuation information and sent officers door to door in affected areas, including Batesville. [1] That may reach people missed by wireless alerts. It may also arrive later, cover fewer homes or depend on roads and staffing. The proper comparison is not digital good, analog bad. It is reach, time, language, clarity and action for each method.

Sirens worked where the water and hardware met

Texas expanded its outdoor warning network after the deadly 2025 floods. New sirens in Ingram and in Kerr and Kendall counties were used during this week's storms. Twenty-eight additional counties are eligible for flood-warning money, but most remained at the stage of preparing implementation plans for review by the Texas Water Development Board. [1]

Three of Kerr County's six new sirens sounded. The manager of the Upper Guadalupe River Authority said the other three were in places that experienced only minor flooding. That explanation matters. A raw count of three unused sirens could suggest hardware failure; the cited account instead places them outside the severe local threat. The next audit should still publish the activation criteria, water readings, operator logs and test history. It should not manufacture a malfunction from an unused device. [1]

Comfort had kept one warning siren for years and recently installed two more. The volunteer fire department sounded them twice, about an hour apart, because some people remained in place. [1] A siren receipt therefore opens another stage rather than closing the case. Did residents hear it indoors? Did they recognize the tone? Did they know which direction was safe? Did people stay because they doubted the threat, lacked transport, cared for someone who could not move quickly or had already lost an exit?

The same restraint applies to the official claim that the changes saved lives. Gov. Greg Abbott said preparations after 2025 improved the response and that lives had been saved. [1] The lower toll in the cutoff-safe record is welcome, but it is not a causal evaluation. Rainfall, river behavior, occupancy, timing, location, evacuation and chance all affect outcomes. A credible finding would connect a named warning to receipt, departure and avoided exposure without assuming that every survivor proves the system worked.

Private companies are filling parts of the map as public systems develop. River Sentry, founded after the 2025 floods, said it had installed 104 water-triggered sirens along the Guadalupe River at private sites such as camps, hotels and recreational-vehicle parks. Hononu received a state contract intended to ease agency purchases of its water-level sensing technology. Watch Duty expanded from fire tracking into flood monitoring. [1]

These are additions to the chain, not substitutes for an accountable public record. A privately owned siren raises questions about maintenance, testing, public coverage and what happens when a site closes or changes hands. A sensor produces data; an agency still must decide when that reading becomes a warning. An app can notify users who installed it; it cannot define coverage for residents who did not.

Receipt is only the middle of the chain

The temptation after a disaster is to count outputs. Four alerts. One city alert. Thirty-eight Weather Service messages. Three operating Kerr sirens. Two Comfort sirens. One hundred and four private towers. These numbers are better than a slogan, but they still measure equipment and transmissions rather than protection.

Protection begins earlier, with a forecast tied to a local river and a decision threshold. It continues through authorization, message composition and translation. Delivery then must reach a phone, radio, siren, app or doorstep. A resident needs to understand the instruction and have a route, vehicle, shelter and enough lead time to use it. Rescue begins when one or more of those stages no longer keeps a person away from the water.

That sequence explains why more warning can coexist with surprise. A phone may suppress overnight notifications. A local message may never be sent. A broad weather warning may omit the street or creek that makes danger legible. A siren may sound without a spoken route. A resident may receive a clear order after the low crossing closes. None of those possibilities should be assigned to a household without evidence. All belong in the audit design.

Language belongs there too. AP's review identifies channels and messages but does not provide a recipient-level account of language coverage. [1] Texas should publish when equivalent instructions appeared in the languages used by affected communities, and whether a translated alert carried the same urgency, location and route as the English version. Sending a message is not equivalent to making the instruction usable.

Transport is the next denominator. A family with two cars, an older resident who no longer drives, a camp moving children and a traveler in an RV face different evacuation problems. A warning review that stops at receipt can praise a system that told people accurately they were trapped. The record must align each order with the roads and means available when it arrived.

Twenty-eight plans are the next test

The counties still preparing implementation plans are the most consequential line in AP's audit. [1] Installed sirens provide visible evidence. Planning counties provide promises. Each should publish the hazard map, proposed coverage, device type, procurement, funding, maintenance owner, activation rule, test schedule, language plan and completion date. Otherwise eligibility becomes the comforting word placed between last year's lesson and next year's flood.

The same publication standard should apply to systems already operating. For each affected community, an after-action table should align the first forecast, watch, Weather Service warning, local alert, siren activation, door-knock deployment, evacuation instruction, road closure, gauge threshold, emergency call, dispatch and rescue. It should include the number of people believed at risk and the limits of each dataset. A row with no record should remain blank rather than be filled by confidence.

No verified X post was recovered for this article. The paper therefore cannot describe statewide warning triumph or statewide government failure as observed platform consensus. AP's reporting supplies a better disagreement: officials and some residents describe improvements, while the delivery record and testimony expose geographic gaps. [1]

The defensible conclusion at Saturday's fixed close is neither that reforms worked nor that Texas learned nothing. Kerr County and Kerrville sent alerts. New sirens operated in Kerr County and Comfort. Weather Service messages circulated. AP found no listed local wireless alert from Uvalde County and reported residents warned by friends. Most of 28 eligible counties were still planning.

That is uneven improvement, not a rhetorical compromise. It is the finding the evidence permits. The next task is to turn every warning into an inspectable path from water reading to household action, community by community. Floodwater crosses county lines. Protection still depends on what happened before it reached each door.

-- DARA OSEI, London