We used AI to build this. Here is exactly what that means.
Most products bury this in fine print. We think the opposite way: if the process is honest, showing it builds trust — and if a figure is wrong, hiding the process would not make it right. This page is the whole truth about how this service is made, what it can and cannot promise, and how to check it yourself.
1. AI disclosure
Artificial intelligence has been used in the creation, orchestration, and ongoing operation of this service in its entirety — including the collection and transformation of source data, the derivation and verification of published figures, the drafting of explanatory text, the design and construction of these pages, and the composition of communications sent on our behalf (including email). The "Ask anything" feature generates answers with a large language model at the time you ask. Where a specific output is produced live by AI, the interface says so; everything else was produced with AI assistance and reviewed through the automated checks described below.
2. Nature and limits of the data
All figures presented by this service are derived exclusively from records made publicly available by agencies of the State of Texas and the United States government (see Sources & methods for the complete register, including the exact file, column, and arithmetic behind every published number). This service adds synthesis and presentation; it adds no non-public information.
All data carries a margin of error. Figures are provided "as is" and "as available," without warranty of any kind, express or implied, including without limitation warranties of accuracy, completeness, merchantability, or fitness for a particular purpose. Underlying records are published on a lag and may be restated by their publishers; derived figures reflect the files as of the date stamped on each page.
3. Verification — and its honest limits
The service is grounded in both deterministic and non-deterministic design, orchestrated across thousands of automated checkpoints: source files are fingerprinted; every headline figure is independently re-derived from the state's own file by tests that do not share code with the system that produced it; rebuilt artefacts are compared byte-for-byte; and the live site is checked against the audited codebase. Notwithstanding these measures, errors can still occur. No representation is made that any figure is free of error, and no automated system — AI-assisted or otherwise — should be treated as infallible.
4. Your responsibility; assumption of risk
Any use of, or reliance upon, information from this service is at your own risk. Verify independently any figure that matters before quoting it, publishing it, or acting on it. This service is for informational purposes only; it is not an audit and does not constitute financial, legal, investment, or professional advice, and nothing here is a finding of wrongdoing by any person or entity. To the maximum extent permitted by law, TAG ai and its contributors disclaim all liability for any loss or damage arising from actions taken, or not taken, in reliance on this service. We share public data in a form people can actually read; the decisions remain yours.
5. Check us — please
Independent verification is not merely permitted, it is the design:
- Feed it to an AI of your choice. Copy any figure from this site
into the LLM you trust and ask it to validate against TEA's published
files — or connect an AI assistant to our machine-readable interface
(Model Context Protocol endpoint at
https://txisd.dev/mcp), which serves the same data with every caveat attached. Two independent readings of the same public record beat one. - Re-derive it by hand. Sources & methods names the file, the column, and the arithmetic for every measure — the same trail our own tests walk.
- Read the code. The entire system is open source (MIT): github.com/GusAI40/texas-isd-finances.
6. What we learned building it (the short version)
AI has come a long way in a very short time — and the lesson of this project is that the value is not the model, it is the orchestration: first-party sources only, one identity key carried through every join, refusal as a feature (the system says "unknown" rather than guess), every claim framed against what it includes, and every published number re-derived from the state's own file before it ships. The state published all of these numbers for years. What was missing was a way for humans to frame the information — that is the layer we build.
7. Corrections
If any figure looks wrong, tell us — gus@ubntag.com. We investigate against the primary record and publish corrections with credit. Two published claims have already been corrected this way; the record of both is public.
Operated by TAG ai · 100 Plaza Place, Suite 300,
Northlake, TX 76226 · gus@ubntag.com
Artificial intelligence was used to build and operate this service. Figures
derive from official public records, carry a margin of error, and are provided
"as is" without warranty; use is at your own risk.
How this was made & terms ·
Sources & methods ·
Privacy