How the records get here — and how we check them.

1. Principles

• Source on every figure. Every statistic carries its source, URL, and retrieval timestamp. No traceable source, no publication.
• Coverage labels, not marketing. If a dataset covers 184 of 254 counties, the label says so — on the page, in the byline.
• Leads, not determinations. Screens produce records for officials to review. We do not publish findings of fraud, intent, or final ineligibility.
• Records are never paywalled.

2. The ingest pipeline

County and agency portals, public APIs (Legistar, Granicus), EMMA/MSRB filings, state regulators (TCJS, ERCOT, RRC, TEA), and Texas Public Information Act requests. Documents are OCR'd into a searchable index — 1.84M documents, 312 GB of PDF as of July 2026. The registry of named sources lives on the sources page.

3. Verification — the homestead screen as the worked example

The statewide screen flagged 1,316 accounts from downloaded appraisal-roll data. Before publication, 642 live appraisal-record checks and 20 voter-file verifications reduced that to 662 referral-ready leads. A lawful carve-out (Texas Tax Code §11.42 mid-year sales) was applied so legitimately transitioning accounts were excluded. What remained was published as a review queue — with names and addresses withheld from the public explorer.

4. Refresh cadence & honest staleness

Each dataset carries a cadence label (daily / monthly / annual / static) and a last-retrieved stamp. When a refresh slips, the public build-status page says so and the automated watchdog opens an alert. An honest stale label is a feature of this system; a silently stale dashboard would be the failure.

5. Limitations

• Lag. Records describe the day they were filed, not today; retrieval stamps show the gap.
• Coverage. Not every county publishes equally; gaps are labeled per dataset.
• Screens are not verdicts. Downloaded data can't see clerical lags, corrected records, or lawful arrangements visible to the district.
• OCR error. 312 GB of scanned PDF carries character-level noise; figures that matter are verified against live records before publication.

6. Change log

July 22, 2026: expanded scheduled checks to all published build endpoints and added source-aware freshness reporting. July 21, 2026: restored independent refreshes for Harris County, HISD, Houston City Council, and METRO agenda sources. Questions about method: send a message.