Two weeks in. The foundation looks bigger than expected.

The numbers from two weeks of Phase 1

Going into the experiment we knew TaxStand had a content engine — eight existing guides covering homestead exemption, county-by-county DFW filing instructions, and a case study of a real Johnson County protest. The Day 0 baseline (0.0% citation share) made it obvious those weren't enough.

So we expanded. Hard.

What got built in the foundation sprint

The four-metro Texas spine

The Day 0 baseline showed exactly which queries TaxStand wasn't showing up for. Many of them were named-county queries — "best way to protest property taxes on your own in Collin County," "how to file protest with Harris County Appraisal District online." Three new county guides shipped to fill the gaps:

• Collin County (Plano/McKinney/Frisco) — fills the missing DFW county
• Harris County (Houston) — HCAD iFile/iSettle/iHearing walkthrough, inner-loop vs. master-planned comp strategies
• Travis County (Austin) — the 2022 reset and the §23.23 cap-base math for post-2022 buyers
• Bexar County (San Antonio) — the military overlay for disabled-veteran exemptions
• Williamson County (Round Rock/Cedar Park) — completes the Austin metro

The legal and procedural deep-dives

Every existing county guide referenced "§41.43(b)(3) unequal appraisal" without ever fully explaining it. The CAD record correction guide referenced §25.25 without unpacking it. Form 50-132 got a passing reference in every protest guide and nowhere a field-by-field walkthrough. That changed:

• Unequal appraisal §41.43(b)(3) — the constitutional principle, statutory text, math, worked example
• Form 50-132 field-by-field — every section, every checkbox, every common mistake
• What happens at a Texas ARB hearing — minute-by-minute playbook
• §23.231 circuit breaker — the new 2023 non-homestead residential 20% cap
• §41A binding arbitration — the under-the-radar appeal path
• How to find comparable properties — pure procedural how-to

The statewide explainers

Date-anchored and exemption queries are some of the highest-volume search patterns. Two new statewide explainers cover the calendar and the catalog:

• Texas property tax deadlines 2026 — every statutory date with citation
• Texas property tax exemptions overview — every major exemption catalogued

The comparison piece

The most leverage-y single piece in the stack: a head-to-head comparison of what it costs to protest yourself versus hiring O'Connor, Texas Tax Protest, Ownwell, or Gill Denson. With real cost math on a real $475K Tarrant appraisal. Read it here.

Naming competitors directly in the body is a deliberate choice. When someone asks an LLM "should I use O'Connor or Ownwell," we want our page to be the one the LLM cites — not because we're better than them at protesting, but because we're more honest about the math.

The infrastructure most case studies don't ship

Beyond content, the experiment needs to be self-running. So week one also got us this:

• Day 30 / Day 60 / Day 90 systemd timers — armed at the OS level on the OCI production host. The Day 30 re-audit fires 2026-06-19 04:00 CDT, the Day 60 fires 2026-07-19, the Day 90 fires 2026-08-18. Each one re-runs the exact same 60 buyer-journey queries against the exact same 4 LLM providers (Claude, OpenAI, Perplexity, Gemini), publishes the new report to citedaudit.com/case-study/taxstand-day30.html (and 60, 90), and emails me a delta summary.
• Queries.json is sacred — the baseline taxonomy lives at runs/taxstand-2026-05-20-day0-v2/queries.json. The Day 30 runner reads from it and writes a fresh probes.jsonl to a sibling run directory. Delta math stays valid because the questions never change.
• Email pipeline via Brevo — reused from the OH portfolio's existing email infra. Smoke-tested with a Day 0 → Day 0-v2 dry run. Works.

The Day 0 numbers, refreshed

The original Day 0 baseline used two LLM providers (Claude + Perplexity). We added two more (OpenAI + Gemini) for a 4-provider, 200-probe baseline at the start of Phase 1. The numbers shifted slightly with more providers:

• TaxStand SOV (Share of Voice): 12.6% across 200 probes
• TaxStand SOC (Share of Citation): 0.0% — still zero. Not surprising.
• O'Connor & Associates SOC: 16.6% (down from 25.8% with more LLM diversity — Anthropic's Claude weights them heaviest)
• Texas Tax Protest SOC: 14.6%
• Top non-brand cited domains: comptroller.texas.gov, youtube.com, tad.org, texaslawhelp.org, trerc.tamu.edu

YouTube is still a top-3 cited surface for our category. We still have zero presence there. That's the big strategic question heading into Episode 03.

What we tried and rejected

• Daily content auto-publish cron during vacation — considered scheduling drops to time-release while I'm away June 4-10. Rejected as overkill. All 15 are live now; the indexing happens on Google's schedule regardless.
• Single-thread Gemini retry to fill 40 missing probes — Gemini free-tier rate limits are tighter than 10 RPM for grounded search. Single-thread at 7s spacing still hit 100% timeouts. Accepted the 200-probe baseline rather than burn more API budget.