Bexar County Capital Projects, and the Column That Hides a Third of Them

Construction_Cost is a text column. Most of its values read 4000000.00. Eight of them read 94000000 $, with a space and a dollar sign on the end. Cast the column to a number, which is what a spreadsheet does when you open the file and what ::numeric does in a query, and those eight rows return nothing at all.

The result is $440,953,000 across 70 projects instead of $667,303,815. $226,350,815 vanishes, across 8 projects. The largest of them is Alamo Ranch Pkwy at $94,000,000, which is the biggest single project in the county program and is completely invisible to the naive read.

Nothing about the smaller figure looks wrong. It carries no error, no gap, no warning and no missing-data flag, because from the arithmetic’s point of view nothing is missing: eight strings simply were not numbers. That is the whole class of defect this site exists to publish, and it is the same one the bond ledger records in mirror image, where the money was written $17201615.67 with the symbol in front and a parser that stripped commas but not symbols reported a finished program as zero dollars.

The City of San Antonio open-data catalog holds 166 datasets. Searched on 2026-08-30 for budget, finance, spending, contract, vendor, payroll, salary, expenditure, checkbook, payment, procurement, revenue, audit and employee, not one dataset matches. The City’s whole finance domain is published as PDF. The bond ledger tool only reaches a vendor name at all by joining into sibling packages on a project number, and even there the current program names nobody.

Bexar County puts Contractor, Consultant and Construction_Cost on the same public layer as the project geometry, queryable without a key. That asymmetry is the reason this page exists: it is not that the County is unusually transparent, it is that the City is unusually quiet, and the contrast is only visible once somebody holds both.

It is also easy to miss. This layer sits on the County’s own map server, and the County’s open-data portal, which is the surface a search engine finds, lists twenty-two datasets of boundaries and census geography and no permits, projects or money at all. The same discovery problem hid a county permit register from us until August 2026.

No schedule, no delay and no variance. The four date columns mix real dates, quarters, bare years and phrases like “Estimated Winter 2026” and “Early 2027”, plus one value with a five-digit year. A duration across those is arithmetic on four different kinds of thing. The shapes are counted instead.

No merged company names. The record holds J3 and J3 Company LLC, Curran Construction and Curran Contracting, and a contractor spelled with the County’s own typo. Deciding which of those are one firm is a judgement rather than a reading, and a wrong merge is invisible once made. The counts on this page are counts of spellings and say so.

No project manager names. The source layer carries six fields holding County staff names, their direct email addresses and their direct phone numbers. They are never requested, so they are absent from the file, the download and this page. The bond ledger makes the same refusal about the City’s contact column.

And no money is allocated between precincts. Two projects span more than one, thirty rows name TxDOT or the City rather than the County as the lead agency, and a third of the ledger records no cost at all.

sa-county-capital.json carries every project, both readings of the money column, the eight rows the naive parse drops, and the date-shape counts. sa-county-capital.csv carries a row per project.

Every row in the CSV holds construction_cost beside cost_as_written, so the parse can be checked against the County’s own text rather than trusted. That pairing is deliberate: this page is about a column being read wrongly, and a download that published only the parsed figure would be asking for exactly the trust it spends the rest of the page arguing against.