We pressure-tested the SEO Believer strategy against Charlie's five known weak points (deep dives uneven, the whitespace proxy is indirect, no city-vertical data, the scoreboard is copyable, solo bandwidth versus the plan). Then we ran adversarial reviews of the launch plan with four separate LLMs to find what we were missing before spending money or time.

• Bandwidth cap: 15 hours per week on SEO Believer, multiplied by Claude Code plus a proven offshore model. The leverage is code-fixable wins, not headcount.
• Case Study Zero: Rebuilding Seminars (already signed, generating real data, relationship no longer private). Fastest published before/after. No stranger pilot needed first.
• Scoreboard demoted: the public prediction scoreboard stays, but as the proof layer, not the headline positioning.
• YouTube channel: a channel on local SEO and GEO for service businesses is the real moat. Audience plus authority compound and are far harder to copy than a web page.
• Lab credential is niche-only: the medical-technologist background sells to medical, dental, and scientific buyers. Do not anchor general service-business positioning on it.
• Rank-grid engine: build, do not rent. The reviewers urged renting Local Falcon first; we built our own RankGridEngine instead, as a reusable owned capability. It was built to Codex's 15-flaw spec (27 tests passing) and runs on mocked data at zero budget, with a cheap one-city validation (a $25 Local Falcon month or a manual Maps check) before any client use.

Pulled from Ahrefs API, all growth-to-peak, zero paid ads:

• Domain Rating 0.9 to 62 (peak June 2023)
• Organic keywords 0 to 31,310 (June 2024)
• Referring domains 0 to 1,187 (September 2025)

The "0 to 69" recollection was wrong. Use 0.9 to 62. Full charts and method note on the Bone Voyage proof page.

1. Dentistry, implant-cost (Boulder) , lead. One displaceable cost authority.
2. Orthodontics, braces-cost (Boulder) , same single-source-authority shape.
3. Med spas (Boulder) , proof vertical (RESTOR-type candidates).
4. HRT / TRT, local clinics, nationwide , St. Louis first, daughter on the ground. Portable city-to-city.
5. Cosmetic surgery , opportunistic only.

Dropped: personal injury law, wealth management, LASIK, fertility/IVF.

The Google-page-1 proxy overstated the vacuum. A direct 6-engine test (Perplexity, Claude, Grok, ChatGPT) showed the candidates ARE named by at least one engine on high-intent queries.

The real play is "named-by-one to named-by-all, plus own the structured-data layer." More honest, still winnable. For HRT, telehealth (Hims, Defy) is NOT the competitor; the competitor is multi-location franchises (Gameday Men's, Ageless Men's). Details on the direct and extended test pages.

Full method on the HRT metro targeting page.

Consensus findings:

• They recommended renting a heat-map tool (Local Falcon) before building one. We did not adopt this. See "Decisions locked" above: we built RankGridEngine as an owned capability with a cheap validation step.
• The plan front-loads proof work into the leverage-free window (no revenue, no help yet).
• Case Study Zero cannot show a real before/after in 90 days. Reframe to baseline plus written predictions plus leading indicators.
• Gate on 7 to 14 day leading indicators (crawl, index, schema validation, first AI mentions), not 30-day rankings.
• Add a real pipeline/revenue metric: discovery calls booked, pilots closed. Not just citations.
• Riskiest unstated assumption: that AI-citation drives revenue. That claim is vendor-sourced and unproven at the city-vertical level.

Codex found 15 technical flaws in the engine spec, including search_places location drift, rank_absolute versus rank_group confusion, a parser and cost-guard that were specified but never built, and a national scan that is roughly 49,000 tasks and therefore not cheap. We addressed these flaws in the build rather than abandoning it: the engine was rebuilt to that 15-flaw spec, has 27 tests passing, and validates on mocked data at zero cost. Full writeup on the adversarial plan review page.

• Revise PLAN.md per the review: one anchor in weeks 1 to 7, reframed metrics (baseline plus predictions plus leading indicators). The engine was built to Codex's 15-flaw spec rather than rented.
• Funding DataForSEO is lower priority for now: the engine validates on mocked data, so only a real national scan needs funds.
• Confirm St. Louis with a fuller read before pitching (the first sample was thin).