Reporters built the border tower death map one Texas sheriff's office at a time

In most of those counties, the fullest record of a local death is kept by a justice of the peace, a locally elected administrative judge ^[13]. Some of those judges never go to the place where a migrant died, and formally declare the person deceased over FaceTime with the first responders on the scene ^[14]. Their reports sometimes leave out the scene and location details ^[14].

Location is the field the whole analysis turns on, because the question was why so many people died near towers meant to track and apprehend them ^[2]. So MIT Technology Review filed with the county sheriffs instead, beginning in July 2025, on the grounds that those agencies are often first on the scene and their reports can carry more detail about where remains were found ^[15].

Getting the files took periods of near daily calls to the offices, and some never arrived at all. Some counties charged nothing; others charged between $300 and $1,300 ^[17]. Had all 17 charged, the fees alone would have run between $5,100 and $22,100 ^[19].

The sheriff route improved the location detail and shrank the count. Deaths handled by local or state police do not appear in sheriff files, so the Texas figures undercount ^[16]. Several counties also warned that their own holdings were incomplete, with an unknown number of records lost during a move, damaged, destroyed or misplaced during transitions between sheriffs ^[18].

Border Patrol estimates more than 10,000 people have died during crossings since 2000, a figure the publication says is widely considered an undercount, and there is no national protocol for how records of recovered remains are handled ^[7][8]. Against that, the inclusion test was deliberately narrow: a case was analyzed only if reporters could confirm the person was believed to have been crossing, where the remains were found, and roughly when the person died ^[9].

The methodology account describes how the journalists assembled their own dataset. It does not say whether CBP or the tower vendors ever ran the same comparison ^[20].

For anyone who has to answer for a deployed system, the useful part here is the shape of the measurement problem. Two things decide whether the system can be evaluated at all. First, whether the outcome it claims to affect is recorded at the same resolution as the deployment, meaning a place and a date. Second, whether that record sits inside your organisation or in someone else's filing cabinet. When the resolution is coarse and the cabinet belongs to a county clerk, an effect claim stays an assertion until somebody pays the per-record fee and makes the calls ^[17].

MIT Technology Review says it went back to 2015 so the data would cover different border policies, presidential administrations and tower technologies ^[6].